📚┃suno-school

[half dragon]

A feature similar to create a Persona; more like a rebrand of Persona

[novel osprey]

no it wont allow me to make a voice based off of uploaded audio...

and there seems to be no more record voice feature

[half dragon]

Yeah, it will not allow one to create voice from an upload. It's been so for quite sometimes.

[novel osprey]

so how does one create a voice based off their own then?

[half dragon]

Use Custom Model feature. Upload your voice, or recorded live on Suno, to train your custom model, then use that to generate your songs.

[novel osprey]

ya i get it now ty, only problem now is its now picking up my mic for some reason.. it works with everything else but when i try to record live into suno nuthin... maybe ill try it on my phone again mannnn arg lol

[small bridge]

How does it would if I already have an annual (and over six months left on it) pro vers and I see now studio is 20% off. Is there a way to upgrade now? or will it apply when the year is up? How does it work? Will I pay full price 20% off and start the new one now and just lose the rest of the pro ver?

[half dragon]

I did this way back in the day... upgrading from Pro to Premiere... it upgraded immediately so you can use Premiere right away and took advantage of the sale.
The downgrade is the opposite, it'd only happen when the current plan expired.
I suspect it'd still work the same.

[small bridge]

What happened to your money you spent on pro? was it wasted? like it doesn't apply to the purchase?

[quiet solar]

I believe the way Suno handles it is charging you the annual fee for Premier starting immediately, then separately refunding you the pro rata amount for how much you have left on Pro (e.g. if you've had Pro for 6 months already, you'll get back half what you paid).

[broken phoenix]

HI. I am looking for a basic prompt guide for clear sounding instruments./ voice is there somerhing here? many thanks !! happy sunday

[slender fable]

I’m trying to remix a cover of mine translated to Japanese, but the results either don’t match the original tune, or do match the tune but are in the wrong key. What the hell? How do I get it to match the tune AND be in the right key???

the new interface sucks, by the way. The “use last prompt” button is gone, the library of uploaded audio/samples is gone, and the remix function is so weird where it shows up at the bottom with just a simple prompt bar and the option to go to “advanced”, which doesn’t have all of the features anyway. The sliders are completely absent. I can’t even edit the song name before generating it either.

[slender fable]

I’m going to forward this to the iOS bug reports.

[quiet solar]

To get the sliders, you need to choose Remix: Reuse Styles on iOS. I presume they removed the option to switch from Simple to Advanced on standard cover, to simplify for the majority of iOS users not doing more complex things.

[slender fable]

That will remix the song/tune and not just copy the prompts and lyrics?

[quiet solar]

If it’s a straightforward song, you can specify the key signature in the prompt and v5 models will usually respect it (v4.5 less so). If you’re using 4.5 you may just have to go through multiple attempts if you need it in the original key.

If you’re using v4.5 or v5 models, try setting sliders to something like w20 s50 a60-70 if you’re looking to totally change the lyrics but retain the melody relatively unchanged.

v5.5 is more sensitive to sliders, so try something like w30 s50 a40-50 as a starting point for the same.

If you need the melody and instrumentation copied exactly, and just the lyrics changed, you might want to consider splitting the stems and editing the final version together rather than relying on prompting alone.

[slender fable]

Um…I’m on the free tier.

[quiet solar]

Then your options are a bit more limited; you’ll have to rely on trying multiple times.

[quiet solar]

Ah, you’re right - that doesn’t bring the original song over.

Guess it’s a job for the desktop version.

[polar spade]

• Well... it uses what we would consider are equivalent to mastering techniques to separate the vocals and instruments to make a feeling of much fuller sound.

• Another term might be "to clean up" and output , whether in older model or current outputs. So it can filter out low volume moments, typically unwanted noise, like a hiss or click.

• But also may turn noises into another instrument.

• also very low fidelity sounds might be enhanced and brought forward in the mix, so a faint trill of a synth arpeggio might show up as if it just suddenly appeared, but was actually digitally buried before in the music.

-- from there, we have to determine if we like the results or not, as sometimes the sound may actually become more muddled, phasing, warbled, or off in its energy. It may filter out low moments which were intended, such as a whisper, or a sound effect, or sound which was to taper off, where it gets clipped in Remaster. So listening carefully is needed afterwards to make a good judgment.

As most things, you get two variations. And you can choose from three models and three strengths from a couple of the models.

[azure crane]

Thanks, yes, it just takes a little tweaking and the result can be really good 😊.

[hoary stream]

yea i tried, it's just making it louder on the bass, a bit more crips on the vocals

[sleek tapir]

Hi some help; I am trying to cover in Studio a simple recorded guitar riff with a different style, using a different guitar prompt style, but I get a full song (bass, drums, etc...) instead of a single guitar cover. Any suggestions?

[torpid wolf]

Hi everyone. I'm Raul. I’m pretty new to Suno and this is also my first time using Discord. I’d like to ask the community about a technical issue I’m having with adding my own voice. Is this the right place for that? Apologies in advance if this isn’t the right channel or if I should start a new thread.

[polar spade]

It does say "upload error" .

If you just use the upload tool in Suno, does it upload OK from that?

This is an upload test. Dont click +Voice. Just click +Audio and /upload to test if your file will upload, without the Voice verification check.

[torpid wolf]

I tried about 20 times this morning and kept getting that error. I just tried again to take a screenshot of the error for you, but it actually worked this time! Thanks a lot for your help

[covert granite]

I'm still having problems with 4.5 5 and 5.5 in covering a song that is long.... I give it a 5 or 7 min song and the cover is 2 1/2 miin long...... it just ignores entire sections of the song I've played with sliders and prompts till I'm blue in the face.........

help?

[potent charm]

I wrote a new 1st verse well really finished the verse but this is annoying trying to blend it into the song

[quiet solar]

Yoshi, are you breaking Suno with complex songs again? 😂

[quiet solar]

If the new verse is consistent with your existing verse structure, then covering the existing song and adding/editing the lyrics is probably your best starting point, if you haven’t tried that already.

[potent charm]

https://suno.com/s/uy6SyAco0MwZjOFz

https://suno.com/s/2GlEo1c1wb36BHue

@quiet solar I have a blended version I did in the studio but I need to finish blending when verse ends and hook comes in

[covert granite]

no this is a very simple blues song.....slow and laid back with clear divisions and separate instruments. and it insists on sax sections harmonics instead of a simple quartet... it just can't resist putting in three saxes playing in harmony. and the biggest problem of all is I can't convince it to go to 5 min...... even when the original is 5 min long. even if I set the follow audio to 90pct or 50 pct it doesn't matter the only thing that helps partialy is the lyrics..... if I have someone sing some fake lyrics it will go 8 min...... but it does not like instrumental music

[polar spade]

Glad it worked. Keep in mind any large files, say over 45MB, could also be timing out, if the server is busy.

[slender fable]

Except that the sliders don’t work on the website and the default setting for audio influence is 25%, which is absolutely ridiculous. The other sliders’ default settings are 50%. Why is audio influence set to 25%? And why can’t I change them?

[quiet solar]

The sliders definitely do work on the website.

They were restricted to Pro/Premier users on launch and, reading Suno's tier description, it seems they're still intended to be.

I believe Cover was originally intended as a tool to switch genres and fully remix rather than clone music, at which 25% is a pretty good default, but having access to the sliders opened up opportunities to use it in different ways.

[quiet solar]

Have you tried 'even faker' lyrics, like just putting in punctuation so that nothing is actually sung, but that there's something in each section for Suno to latch on to?

I'll be honest, I don't know if it'll work or how well, but I recall someone experimenting with something similar around a related problem, so it might be worth a try. Otherwise, I'm really not sure what else to suggest.

What I can say is that a 7 minute song might be a challenge in itself, because the maximum song generation is 8 minutes and some of that capacity is allocated to interpreting audio uploads - I've seen 2m-3m cut off that time when uploading audio that was 4 minutes long and trying to extend sections through cover.

[quiet solar]

Just trying to get my head around how you've approached this so far...

It looks like you've used the original song as Inspo for the reworked verse 1, which you've created in a totally separate song that's just verse 1, which is now about 5x longer. And what you want to do is take that new verse 1 and use it to replace verse 1 in the original, totally separate, song.

That's going to be very difficult without editing via Suno Studio or an external DAW from this starting point. But doing it by editing would be fairly straightforward if you're familiar with audio editing, since there's a convenient tiny gap of silence between verse 1 and the hook in the original.

If you didn't want to do it by editing, and want Suno to blend the whole thing together itself, I'd suggest an alternative route:

Use Remix -> Cover on the original song, and in the lyrics box replace the original verse 1 lyrics with your new verse 1 lyrics. Leave everything else unchanged in the lyrics and style prompt. Then since you're using v5.5, set the sliders under 'More Options' to something like

• weirdness 30%
• style influence 50%
• audio influence 50%-60%

Because it's such a dramatic change to the structure of verse 1, you might need to play around a bit and gradually turn down the audio influence in search of the sweet spot if Suno struggles to reinterpret the new verse on those settings.

[quiet solar]

It can help a bit if you set that stem as Guitar in Studio, but yeah - that can happen.

If you're happy with how the guitar sounds in the new version, you can select that cover, go into Clip and split the stems from it, importing only the guitar and leaving out everything else Suno created.

[potent charm]

100%. I have it at 90bpm, I am going to cut the silent part and maybe a half second and try to blend it into the beat or the way you explained in the editor.

Life be lifing and I wanted more on the first verse and it took me a little while to write. I had to read a few more of their arguments during the era. Then mix my own opinions into like I did the other verse lol. I really thought about even adding more to the second verse but I am over 5000 word caps lol

[ocean cape]

Can I bring a lyric and a song for you to analyze? The structure and the result? I have some doubts about whether the way I do it actually yields results or if I'm embellishing it too much (the term v ortex is just a name to find in the files, but does it actually have any personnel or is it just a few meta tags?

[potent charm]

Can I check it out

@ocean cape

[quiet solar]

It's your song, so approach it the way you feel is best.

If you were asking me, I'd be tempted to fade out / crossfade the new verse 1 from 1:54-1:56 and use that as the connection to the hook, so you can make use of that little gap of silence. You could do something similar on the end of the Intro with either a quick fade out on 0:42-043 or a very sharp cut just after the 'Twin Flames' line but before the echo kicks in, and transition into the new verse 1.

Good luck! 🙂

[potent charm]

I will do it both ways I have time tonight my kids will be sleeping over the grandparents.

[quiet solar]

You're welcome to share a link to the song, but we'd need to know what you're trying to do (and why you feel it's not working) if you wanted any advice on how you could do things differently.

If you wanted to share the song just for feedback, that's better suited to https://discord.com/channels/1069381916492562582/1083129341870878780

[potent charm]

Don't laugh at me. Literally if I come up with bars I voice record them because the moment the 1 year old or 4 comes and takes my attention away or whatever the monster feel like I completely forget what I was about to write lol 😆. I left my laptop open a week ago and my 4 year old went on a typing spree.

[quiet solar]

Haha...I feel your pain!

[potent charm]

What kind of annoyed me at first when I used 5.5 is that my saved artist or vocals started to sound like fetty WAP

[cunning depot]

hello guys

what kind of sounds make you feel like being in a summer walk in the sunset, but in the middle of a melancholic movie scene

I mean, not the typical summer vibe

but a romantic one

[potent charm]

Miami Florida push it to the Limit

20% techno
60% higher tempo

@cunning depot

[cunning depot]

the scene is: looking someone you want to see so bad after a long time and caught a glimpse of him while driving

so it feels more like cinematic scene

[frozen dagger]

lo-fi pass, or music drops out to something to evoke the heart beating strong and hard. a deep inhale and then a moment where you don't breathe... but in music. 🙂

[cunning depot]

i've tried with pop, my voice is soft so I need to make it feel deep but the results sound generic to me

this is my prompt: "cinematic pop, melancholic sunset feeling, summer fresh air, marimba melody warm and resonant, vibraphone with bow ethereal, sustained cello underneath, sparse piano notes with long reverb tail, soft brushed drums distant, soft hi-hat played on the edge for a warmer, less sharp tone, breathy female vocals with underwater reverb, slow tempo, modal harmony non-resolving, dorian mood, rhodes piano"

[potent charm]

@cunning depot I DM you 4 templet that I put together. I am more hip-hop r&b but

[frozen dagger]

the marimba and everything else looks good. if you're trying to create a moment within the song then you may want to create the prompt tags inside the lyrics

[Bridge | lo-fi pass, music drops, Steady Heart Bass Pulse]
Lyrics written here.

I would come up with a better prompt then my example, but even that might do something. The style prompt you have looks good. (I would maybe try to define/articulate the vibe in different ways. "Summer fresh air" for example the ai could jsut break that down into smaller parts like "air" and do some airy pads or something. (which could fit, but isn't contributing to the creation in the way we think.) Sounds like you're trying to guide Suno on song direction, which can be done with tags in the lyrics box.
The style and sound is ultimately up to you. Which sounds like you have that for the most part.

[cunning depot]

slowcore sound pretty well for this, but I feel like it needs more instrumentation than just an acoustic guitar and some drums to make it feel better, but I feel like the rhythm or the harmonic progression sounds generic

there´s a sound pattern I recognize in suno music

and it's been difficult to me to break that pattern

[potent charm]

There is no music i just made a beat to see if I understand what they are asking

https://suno.com/s/y2bwvP93tj1I9dua

[cunning depot]

I was listening to those beats and sound pretty well

[frozen dagger]

there is, we can see things happening at the same time stamps in songs, measured similar and structed differently. you can try prompting for more aleatoric and randomnes with style prompts, or recreating parts of the song to try and change the pacing/vibe of the pattern.

[cunning depot]

The thing is, I'm working on two EPs. The first one sounds more like dream pop magical and ethereal, while the second feels more grounded and real. They explore the same emotions, but the first presents them in an idealized way, whereas the second brings those feelings down to earth.

and the 1st one is already finished, this song is made for the second one

[frozen dagger]

I haven't heard anything. So I don't know how I can really help further, but that could be what you're really wanting is some just listener feedback? If that is something you are interested in, and ready to share then I would suggest maybe just asking the community in https://discord.com/channels/1069381916492562582/1069381916492562585 for some ears from peers. Something that ai just can't really do. If not that's perfectly ok, just putting that out there. Ask other Suno artists what they think?

[cunning depot]

i thought my layering was poor or something

like, maybe I need to add more details

[frozen dagger]

Sounds like you have a case of, "something's missing, but I can't quite put my finger on it." In which case those things usually come to us in time.

[cunning depot]

yeees

[frozen dagger]

Ask for feedback, if you're ready, or just relax and take a break. Think about other stuff. Let it evolve in the back of your mind. You finished one already. 🙂

[ocean cape]

Following the rule I was looking for, the intention was to avoid leakage for the interpretation of sound and instrumental instructions, but in fact, should I enclose everything in square brackets? My question is:Does this structure actually generate something more relevant? The focus is on at least being accurate in separating instrumental and vocal/singing instructions; another point is whether adding time actually strengthens the model.I was actually generating songs of six to nine minutes, which was the maximum I could achieve (I wasn't successful in going from seven and a half minutes of music to nine, because I would have to generate...). The continuation of the cat with more credits because the model made the music cut. Here is the link and the prompt :

https://suno.com/s/zxcnG9bIMAkte1P9

When I translate the lyrics into English, I have some words that slip out, like a foreigner speaking Portuguese.

Cut off, interrupted, could it be due to the abrupt ending of the description?https://suno.com/s/1yEnW8d9mexUkgFK

[graceful condor]

anyone can tell me anything else what i should add in this song.

[high snow]

Can someone help me I’m in studio and I uploaded an audio and for some reason it’s slowed down by a few seconds and I have absolutely no idea why. When I uploaded it, it didn’t do that or it wasn’t like that.

[tame venture]

Funky fresh your good to go soul brother

[silent mortar]

Check the BPM setting. Check if it's on manual or follow

[high snow]

I found it and got it fixed

Ty

[crude citrus]

How do you upload a beat like from my guitar and all that from like another app on mobile because I don't know

[solemn topaz]

Yeah, you can upload or record audio in Suno and use extend/cover—but just so you know, it doesn’t continue your beat exactly. It kind of reinterprets it and generates something similar.

So if you're trying to build directly on a beat from another app, it might not match perfectly—but I can help you find a workaround if you want 🙂

[crude citrus]

Thx bro I'll dm u soon

[plucky terrace]

I'm experiencing a consistent audio issue in Suno: every generated track has a brief reverb or spatial glitch around the 49-second mark, regardless of prompt or genre. Could this be due to segment stitching or mixing inconsistencies in the model?

[solemn topaz]

That’s interesting—and yeah, it could be a stitching/mixing artifact.

Mind sharing one of the track links where this happens?

Would help to hear the exact glitch + check what model version, prompt, and lyrics you used. If it’s consistent, we can probably pinpoint whether it’s a generation boundary issue or something else.

[plucky terrace]

V5.5

Thank you. Also, how can the 10K–20K high-frequency noise issue be solved?

[solemn topaz]

It would’ve helped a lot if you could share the track links as well—then I could check the lyrics, style prompt, and model version behind them. That context makes a big difference when diagnosing issues like this.

Also, give me a moment to listen through these properly—I want to hear exactly what’s happening on your end before suggesting fixes.

About the 10K–20K high-frequency noise: that’s usually a mastering/artifact issue (kind of a “digital hiss” or harsh air band). Once I hear your tracks, I can suggest more precise fixes—but it’s generally handled with EQ cuts, de-essing, or a gentle high-frequency roll-off.

[plucky terrace]

https://suno.com/s/6ihXDnbjX6KjWRO7

https://suno.com/s/oNV5Rz0acpoW3Xz6

https://suno.com/s/64zEj6H104mxxSCZ

[solemn topaz]

Honestly, I’m not really hearing any clear reverb glitch or issue in these at 49 second mark — at least not on my phone speaker.

It might be something that’s more noticeable on headphones or studio monitors, especially if it’s in the high-frequency range. I’ll check again properly with headphones once I’m home and see if anything stands out more clearly.

In the meantime, let’s see if others here can spot anything unusual—fresh ears always help with stuff like this.

For now, I don’t want to jump to conclusions without actually hearing the issue properly.

[runic rose]

anyone know how to get SUNO to follow your song sequence EXACTLY for covers?

[plucky terrace]

All three of my tracks have the issue at around 49 seconds. When listening on speakers, the first track is the most obvious, and the other two also still have the same issue, just less noticeable. On a phone, it sounds more like a strong wind or air noise.

Thanks.

[frozen dagger]

https://discord.com/channels/1069381916492562582/1082775596767117312 Put a post in here. There are many others reporting same issue, including myself.

[plucky terrace]

@solemn topazIt seems like a lot of people are experiencing the same issue. It looks like it might be a bug.

@runic roseYou can adjust this by adding prompts in the lyrics field (including instructions), and also by tweaking the settings in the Weirdness, Style Influence, and Audio Influence controls.

[solemn topaz]

Could be—I’m on phone speakers right now, so I might be missing it.
I’ll recheck on headphones later. If a bunch of people are hearing it at the same timestamp, that’s usually not a coincidence.

[shell loom]

have you included a lot of lyrics?, that use to be a problem with short version, lyrics to short

👋🏼 Hello ladies & gentleman.👋🏼

[plucky terrace]

They are all instrumental tracks with no vocals or lyrics.

[solemn topaz]

@plucky terrace Also, in all my tracks so far on v5.5, I haven’t really noticed this issue—either I missed it, or my ears are just taking a day off 😄

If you’ve got time & @frozen dagger too, feel free to check mine as well and let me know if you hear anything similar.

[frozen dagger]

"Exact" is tricky for Cover. If it was a Suno made song I would suggest trying to use the same model that generated it, if not, I'd still try exploring the other models (if possible). As mentioned adjusting the sliders in advanced options. (High Audio influence and lower weirdness little). Sometimes no prompt is better with high audio influence, and sometimes using the same prompt. Any changes made to lyrics or stlye prompt will make changes tho. (even no prompt changes will make changes; that is kinda the nature of Suno; iterative)

[shell loom]

not had too much of instrumentals being short, couple of times maybe, but they seem to go back to normal, maybe it a server capiciate issues.

[frozen dagger]

your public 5.5 sound clean. PRANA CHITRAM — The Last Frame is the only onme that even comes close to slightly hearing it, if that is what it is, it's layered in the background.
In this example @plucky terrace gave https://suno.com/s/dNn7MIO4yhGTQXa7 you can hear it pretty well when turn volume up as it goes from 0:48 -> 0:49 sounds like a thunderstorm rolling in for 1 second. Gotta be something about the way Suno pieces things together, and usually does that at same time stamps. (similar structures)

[plucky terrace]

https://suno.com/s/oNV5Rz0acpoW3Xz6

There are actually many tracks with a similar issue, and it always appears around the 48–49 second mark. Even when doing remixes of different songs, the same problem shows up at the same timestamp. However, not all tracks seem to be affected.

If the previous track sounded like a “rolling thunder grandfather,” then this one sounds like a “rolling thunder baby.”

[frozen dagger]

ya that got dunked underwater for a sec. seen it described a couple of ways. sometimes its "lowkey" other times not... i know i have one somewhere that played and i was washing dishes while listening to the gens and had to stop and come back to my pc like, "what the heck was that?!"

[plucky terrace]

“Rolling thunder baby” is born.

[solemn topaz]

@plucky terrace Alright, checked again on headphones—and yeah, I hear it now. It’s like a brief “wind hit the mic” kind of artifact.

@frozen dagger I think I didn’t notice it earlier because I don’t do a lot of instrumentals. In vocal tracks, these things tend to get masked by layering. Like in PRANA CHITRAM, everything’s dense enough that it blends into the overall vibe.
Real talk—tools like Suno AI aren’t really production-grade yet. They can give you some amazing results, but you either polish them manually or use them as a base/reference for actual composition.

Personally, I just generate with my own lyrics and style in mind and keep rolling until I get something close to what I want.

If you’re expecting clean, production-ready output straight from generation… that’s going to take time—could be soon, could be years.

But yeah, in this case, it does look like a legit bug that needs fixing.

[random stag]

How do you publish your songs on Spotify? Also, how can you prove you own the copyright to the song?

Or am I wrong and we don’t?

[neat nexus]

You have to get a Distrubutor.

Suno says your bill for services is your proof of license.

You can get some information here: https://help.suno.com/en/categories/550145-rights-ownership

[teal saddle]

If someone could help, please? I recorded my voice, added it to creation, include my lyrics and song style, moved weirdness to 25, style to 80, and audio influence to 80, but when I generate, my voice doesn't show up. Any advice is greatly appreciated. TIA.

[neat nexus]

Try using the defaults?

[teal saddle]

So don't worry about advanced options?

[neat nexus]

Not saying not to use them ever, but since you're having issues, it's the first thing I'd try going back to defaults with.

[teal saddle]

Thank you. Will try that

[quiet solar]

So you can prove 'ownership' of the song by demonstrating it was created while on a paid Suno plan (see Hepcat's post), which gives you the rights to distribute it commercially.

Proving 'copyright' (ie, preventing others from using it without paying you royalties) is a whole other issue, and isn't something you can do through Suno as that depends whether the work is inherently copyrightable. If the lyrics and music were generated purely from AI prompts, even with minor editing, it's very unlikely to be eligible for copyright protection.

The distinction is going to get increasingly important as AI disclosures in music become more frequent. 🙂

[random stag]

What is a Distributor?

[fluid robin]

LANDR and Distrokid for example

[neat nexus]

You can always try searching "Understanding music distribution" and see what comes up. I won't mention any of the players in that field because I don't wish to promote of suggest any official support for any of them.

[teal saddle]

Well, I tried it. Still doesn't add my voice.

[neat nexus]

Sorry.

You'll have to keep working with it.

Have you tried keeping your original upload, using it as half of a mashup and using one of the outputs you liked (other than the vocals) to see if that works?

[polar spade]

You are Covering your recordings of your voice or Extending the recording of your voice? Does it say Part 1 on the output file?

[frozen dagger]

When you say "added it to creation", are we saying you used the Suno Voice feature (with the voice verification process), OR are we saying we are doing a Cover/Remix with an upload of your voice in the audio?

[teal saddle]

I used Suno voice feature with voice verification.

[frozen dagger]

Thanks for clarifying that. So now I'm curious what "my voice doesn't show up" means, as in what IS happening? Is the Voice selection disappearing from your UI (it does this with Mashup and other forms of Remix), OR it's generating a song that either doesn't match your "Voice" as it sounds closely, or way off base like random voices; maybe even male and female voices?

[frozen dagger]

@teal saddle and how you are creating as Dasvon was asking.

[teal saddle]

It doesn't match my voice and it generates random male/female voices. Basically, it generates a song without giving any hint of my voice.

[frozen dagger]

Are you using Cover to cover a song that has vocals already or is this an original generation; or an extension?

[polar spade]

Ahhhhh..... 🤔

[teal saddle]

No. I did the voice recording with verification, clicked on create using that voice, put in my lyrics and music style and clicked on generate.

[paper dragon]

Hi dear people, i am a novice suno user using pro plan. I am very puzzled with the edit system of a song, sometimes when i regenerate a verse, i get a cloned result twice in different versions that repeats itself, how do I evade that?

[polar spade]

Have you found the sliders, if Pro sub.? You can try extremes .

Click Cover to Cover the recording. Styles to 5 % and Audio to 98%... and test. We're forgetting about whether it gets the music correct at this point.

[teal saddle]

Yes. I put weirdness at 25, style at 80 and audio influence, I have gone from 25 to 85 trying generations.

[frozen dagger]

Have you made songs using this same Voice with success before, or ....
If not had any success with it sounding like your "Voice" then there is a kinda trick where we use the Custom Model to create a song with the same audio upload we used to create a voice, and create a voice with the generation it makes. It's not difficult, but cumbersome and haven't done this since Voice started. When Voice started users said same thing and this workaround was one method that some found to work for them.

Without jumping through all of those hoops first I would say that the genre and singing style of the upload do matter. A sang-sung vocal voice in a screaming metal song or opera is going to be put through some ai-imaginative changes to match the style. If it matches close enough and that's not the issue, would look at the vocal upload used to create the Voice. Want to have the sample include the range and style in it the way you want performed (to the best of your ability) with as much audio context as possible, or sometimes if you're trying to create a very specific voice you want to just give it the a sample that represents only what you're after, not the whole range. Could try doing a new recording and trying again. If the audio sample was short, just duplicate it and make it bigger. The same idea for this is why the Custom Model trick works; we are reinforcing the influence.

[midnight plume]

Man HOW do you edit lyrics now? I put in an old song to remaster it with a new AI voice. It generated and changed some of the hook in the chorus that CAN'T be changed and I can't figure out how to edit them now since 5.5.

[polar spade]

Style at 5%. Audio 98% , to test if it gets closer.

[quiet solar]

Generally, the best way is to use Remix->Cover on the original song and make your lyrics edits in there, provided they're in-keeping with the original structure of the song.

If you're on Pro/Premier, also set the sliders in Create (under More Options) to something like weirdness 30%, style 50%, audio 50% as a starting point.

[midnight plume]

Thank you! I will try all of the above.

[polar spade]

It's been reported this weekend -- as some users have noted the same. As well as on Reddit.

[digital osprey]

Wow, neat !

[viral pond]

i usually use the editor for replacing parts, but how in the world does the replace work in studio ? it just generate rubbish 😄

[polar spade]

If you want to post this post in the Bug Channel, I'm sure the extra details will be helpful.

[digital osprey]

It’s done! I hadn’t seen that there was a dedicated channel for that.

Do you know whether there is already an estimated timeline for a fix, or is the issue still under investigation at this stage?

[polar spade]

No issue so soon has a fix-by-deadline time. All usually come under the "as quickly as is assessed an issue and the cause, a fix will begin" moto.

[digital osprey]

Thank you for the clarification.

[random stag]

I wish there was an appeal process when it says my hum is a copyrighted song. There is no way my out-of-tune hum sounded like anything that’s been published!!! I feel like this happens way too frequently .

[severe narwhal]

I had it with a lyrics with 'blah blah blah' in it, not allowed : Copyrighted 😂 😂 😂

[polar spade]

If you're just humming or da-da-da ing have you converted that to an instrumental first? Or it wont let you do that?

[quiet solar]

I expect that's because it's only about 1-2s worth of audio. Suno generally works better when it has at least 10s to blend audio from the beginning of a section to the end (see below) - top 2 are around 10s generations, bottom two are the same length you used (and are completely useless).

That said, I've never been a big fan of Replace in any of the Suno models.

[quiet solar]

I mean, it is repeated multiple times in a song by Kesha (and is also the name of that song), so while that is a little bit of an odd copyright block, it's not entirely unreasonable. 😉

[cunning olive]

Any idea how I can generate a track with no guitar?

[quiet solar]

Specify the instruments you do want, don't use the word "guitar" anywhere in the style prompt (e.g. don't add "no guitar").

Guitars are very prevalent in a lot of genres, so in those you may still get guitars creeping in from time to time because they're so common in the reference material, but it shouldn't be a difficult exercise.

Feel free to share the prompt you're using if you want any more specific input.

[viral pond]

ah in the editor it does not really matter how long you select since it always generates more than you selected on either side. but i mean i think the prompt matters, but I just want it to use the existing style to generate the replacement, so all tis song, vocals, synth, fx and stuff confuses me when i want to replace something

[quiet solar]

Can't say I'd noticed it extending beyond the timeframe specified in Studio, but maybe I haven't yet had the pleasure of Suno deciding to ignore the selection I've made. xD

Depending on what exactly it is you're trying to change/tweak, you might be better off changing Replace to Cover. Although the same rules apply - you're much better off with a 10s+ selection as super short ones will inherently cause Suno difficulties. But so long as the tempo is locked, you can just crop down to the shorter piece you actually want to replace.

[viral pond]

sometimes I can mask it with other sounds, but other times its impossible, i often just end up replacing it in the editor, but then I have to manually find the exact position in studio, which not always matches the time in the editor

[ocean cape]

I have an idea that might help you; describe the type of music and stylistic and vocal references.A method specifically for this type of problem that I also have, many times a rhythm pleases us, and these stumbles in copy are annoying lol, anyway I think I can And references, if you can be as thorough as possible, I'll do it my way and share the chat link (Gemini) and you'll understand perfectly.

[quiet solar]

Question - why switch to the editor rather than doing the whole exercise in Studio, where it would be much easier?

[quiet solar]

Oh, no need. It wasn't about trying to replicate the style of the Kesha song.

This was a conversation about someone else having the lyrics 'blah blah blah' get blocked for copyright. I was just explaining why that wasn't as daft as it might first appear.

[ocean cape]

Ah, okay, I understand. But getting back to that, I missed the context, and taking advantage of the opportunity: how can I replicate a specific sound when the input vectors are equal or very close? (Before, I used to say...) Now it seems to me that he makes modifications and continues the task, I have some very specific, and big, questions lol I would have to...

[Style: Instrumental_Symphony, Avant-Garde_Orchestral, Cinematic_Score, Poly-rhythmic_Tension, Acoustic-Fingerprinting, Wide_Stereo_Field, Mastering_Grade]

[Marca_de_Tempo: Zero_Minutos_e_Zero_Segundos]
[Acoustic_Fingerprinting: Resina_Friccionada_em_Madeira_Centenaria]
[Dynamics: Pianissimo_Absoluto]
[Phase: Vulnerabilidade_de_Fase_Sussurrada]
[Sopro_Madeira_Friccao: s-s-s-shhh]
[Ambience: Catedral_Vazia_Reverb_Denso]

[Marca_de_Tempo: Zero_Minutos_e_Trinta_Segundos]
[Pan_Dialetico_Forense_Iniciado]
[Eixo_L: Violinos_em_Harmonia_Institucional_Rigida]
[Eixo_R: Violoncelos_em_Dissonancia_Gravitacional]
[Eixo_C: Silencio_Profundo]
[Contencao_de_Caos: Cordas_em_Ataque_Medido_e_Sufocante]
[Flow: Crescente,lento,arrastado,puxado,tenso,vibrato_pesado]

[Marca_de_Tempo: Um_Minuto_e_Quinze_Segundos]
[Engenharia_de_Prosodia_Instrumental_Mimetica]
[Melodic_Doubling: Oboe_Lead_Mimetiza_Lamento_Humano]
[Sonoplastia_Sintatica: Fagote_Responde_em_Contraponto_Cinico]
(Mais_Grave_Nos_Metais!)
(Corta_Frequencia_Alta!)
[Seismic: Low_Frequency_Oscillation_Swell]

[Marca_de_Tempo: Dois_Minutos_e_Dez_Segundos]
[Desconstrucao_Metrica: Polirritmia_Forcada_Entrando]
[Grid_de_Foley_Institucional: Timpanos_Quantizados_Agindo_Como_Martelo_de_Juiz]
[BPM: Setenta_e_Cinco_Mudando_Subitamente_Para_Cento_e_Dez]
[Dissonancia_Harmonica_Ironica: Acordes_Maiores_Luminosos_e_Ensolarados]
[Textura: Contraste_Brutal_Com_Tuba_Sufocada]

[Marca_de_Tempo: Tres_Minutos_e_Quarenta_Segundos]
[Colapso_Latente_Induzido_Por_Sopro]
[Metais_Plosivos_Loop: k-k-k-k-t-t-t-t]
[Injecao_de_Erro_Transiente: zxzxzx-qyw-zero]
[Physics: Structural_Fracture_Nas_Cordas]
[Texture: Crushing_Granite_and_Glass_via_Pratos_de_Choque]

[Marca_de_Tempo: Quatro_Minutos_e_Cinquenta_Segundos]
[Eixo_C: Trompa_Solo_Heroi_Desperta]
[Authority: Booming_Brass_Authority]
[Mimetismo_Sincronico: Envelope_de_Decay_Curto_e_Seco][Friccao_de_Radio_AM: Filtro_Lo-Fi_Subito_nas_Violas_de_Fundo]
(Sufoca_O_Reverb!)...

It's divided, this is the structure I use.

[pure panther]

Ayeeee, that helper tag looks good on you 🙂

[sick badge]

I have a love song duet I'm trying to make where the woman sings the first verse, the man the second, and both sing the chorus, but for some reason specific lines in a verse changes from man to woman. I've done everything to "force" it to go one way including putting [Verse - Female vocalis] at the top of verse 1, [F] and/or [Female] next to the lines of verse 1, and put "Female verse 1" in the styles box, but one specific line of this verse ALWAYS goes to a man. Am I doing something wrong?

[pure panther]

@sick badge Check this out: https://discord.com/channels/1069381916492562582/1466598465969590465

[sick badge]

@pure panther Thanks! I'll look through that. This comment interested me from that thread: "If it's only the one line, that's a great problem to have. Have you tried covering it with just the one singer so that you can get that line sung correctly in isolation, so you could edit it in to the version that's almost right?"

I never thought I could "copy and paste" from a different version. Is that in the Pro version or Premier only?

[pure panther]

Cutting in Suno is premier only.
However, the way I personally do a duet, including the copy pasting, is to do two songs where the first one uses the first vocalist and the second one uses the second vocalist (use the first song as a cover to get the second one) until you get two very similar versions (one with one vocalist and one with the other)
Then I get stems for them both (vocal + instrumental should be enough, just keep one instrumental)
Then I put them Audacity (a free and simple but very powerful audio editor) And put the vocals where I want them.
When that's done, don't worry it will sound very jank it doesn't have to be perfect, I upload that back into Suno
Finally I cover that version until I get a version I like.
Here's an example of what I made doing that: https://suno.com/s/nycUv4wCc6PEoo5p

[sick badge]

Very nice! I'm a Christian so I think that's cool to make. Yeah, and I'm making mine complicated unfortunately. I'm going for Wagnerian, Jim Steinman style so Suno's AI already doesn't like me. lol. Here is the song I'm trying to make. About 40 seconds in, you hear "and I felt the floor rise", and immediately it took me out of it. lol" https://suno.com/s/Dorc75jDVk44R6Xy

[midnight plume]

I got that to work but an issue I can't fix is repeat play. No matter which way I have the toggle set it loops around and starts the song again. Any ideas?

[rough depot]

Ugh what part of 100 percent audio influence doesn't it understand https://suno.com/s/j0lrbD0bV6mEAAky

[midnight plume]

Also, on almost every song, right at the 40 second mark, there is an almost inaudible dropout. Almost every song.

[rough depot]

It is REFUSING to maintain the tempo and phrasing of the original song!

The original song in question https://suno.com/song/aa903047-0538-4b2a-9ad2-165a9affa89d

[frozen dagger]

This is a new bug. Right-click (or click the "...", from the song with the error and "report bug" there, giving the song a thumbs down also helps. There is a bug report on discord as well https://discord.com/channels/1069381916492562582/1082775596767117312 if you want to post it there with details, but it is a known issue.

[frozen dagger]

What is looping the song player? (mobile or desktop)

[rough depot]

0 weirdness, 0 style, 100 audio still coming out weird. what is the ACTUAL point of the cover feature

[midnight plume]

Desktop

[frozen dagger]

And we are talking about the Suno player looping on repeat? The player repeat funtion is the icon at the end after the Next Track > arrow.

[midnight plume]

Actually what is happening it is going to the next song and starting it without my asking it to.

[frozen dagger]

ah ok. that is the player's funtionality. I turn repeat 1 on so the song i am playing will repeat and not go into other songs. It will continue to play songs from your own library where the selected song was played from (playlist, workspace etc), or if there is nothing in your library it will start playing other Suno user's songs that starting from the "similar artists" tab that it creates.

[frozen dagger]

I can only change the repeat function on desktop, the icon isn't there on the mobile player. I thought maybe this was the issue.\

[pure panther]

Nice! That's already a very pretty song 😄 I hope you're able to get the vocals sorted out 🙂

[rough depot]

[name]esque should be allowed in 100% of situations. homages arent infringement.

[sick badge]

I'll keep on it. Thanks!

[rough depot]

Grrr https://suno.com/s/IDB8FnUH97JOCPBS

It's singing too fast and it's doing melisma thing despite the fact that I literally put melisma into the negative prompt can we please fix this God damn ai?

Melisma = INSTANT DISLIKE

[plucky terrace]

What happened? It’s stuck here and not moving.

OK。。。。It’s fine now—it was stuck for a long time.

There’s another piece of bad news: yesterday, the composition would definitely run into an issue at 49 seconds. Today it still happens, but not as often.

😢

[quiet solar]

Cover was originally intended for reinterpreting songs into different genres - in earlier versions, I avoided it like the plague for trying to match existing audio for that exact reason; it was far more prone to variation than Extend.

The world has changed a bit since then, particularly with the addition of the sliders, but Cover was never designed to be an audio cloning tool. 100% audio influence on Suno's side means "give the existing audio input 4x the default weight when creating the new song". Sliders at extreme settings like 0/0/100 also tend to introduce problems and oddities far more often than those nearer the defaults, especially on v5.5.

As for the "singing too fast" issue, we've discussed variations of this before multiple times. I could already see which section of the song it was going to be before I listened to it:

[Pre-Chorus]
The brave outlander and the fourth princess!
Have fallen deep in love a bond they now confess!
They ride now with hearts aligned flying ever high!

The headache here is that it's a three line section in a four-line structured song, so Suno is having to decide how to space it for you. And the pre-chorus tag is probably influencing it to do something different to the verse, which is contributing to the decision to double-time the first two lines.

If you want to nudge it to solving it the same way it did in your v3, it might help to try restructuring it as something like this, and relying on the audio input to steer how it handles the (former) pre-chorus:

[Verse 1]
The psychic storm is broke, the apex mind undone,
A single chambered round, and so the day was won!
The twisted machinations, have all been blown away,
The heroes stand unbowed to greet a better day!
The brave outlander, and the fourth princess!
Have fallen deep in love, a bond they now confess!
They ride now with hearts aligned
Flying ever high!

[quiet solar]

Well, if you'd like some good news, Team Suno have confirmed they're aware of it and are looking to fix it.

[plucky terrace]

Great, I can finally send Rolling Thunder Baby to kindergarten and never see it again.

[delicate fable]

I noticed the ai hasn't been trained on older more mature voices because it can't generate one. It's always younger adult voices no matter how many propts you give it.

[plucky terrace]

maybe...try...dying?

You can try SMOKY, HUSKY, WEARY.

[viral pond]

because it just does not replace the selected area properly, it just makes something entirely new.. unsure what I do wrong, surely I would use studio if it replaced properly

[quiet solar]

Yup - that's what it does when it comes to replacing a section. It's why I've never really liked the Replace tool in any of it's iterations.

As far as I know, the Replace tool in Studio and the Editor is the same underlying tool. The main differences are that the Editor has a nicer way of showing the lyrics for lyric editing, while Studio's version gives you access to the sliders.

The issue I have with Replace is that it's default level of deviation is pretty high, and it's ability to connect seamlessly to the start and finish of the section it's replacing is pretty low. So the minute you're using Replace, you're setting yourself up to do another pass using Cover or Remaster to fix the problems caused by Replace.

At that point, if it's a simple lyrics change, you might as well skip Replace and just do the edit in Cover in the first place...and if it's a musical change, maybe it's just easier to do an Extend from that point in the song and re-rerun the rest of it, since Extend (until very recently) was pretty faithful to what had gone before.

The experience you're describing is what led me to shift my workflow to Extend (and editing the final result in Audacity) in the first place, and then over to using Cover and a combination of Studio and Audacity for v5+.

[rough depot]

That didn't stopThe AI from using the phrasing it used in the original iteration

It's a pre chorus if I move that to a verse it's not going to have that ascending vocal

[slate juniper]

What is a good prompt to create instrumentals that leave space for vocals? If I prompt [Instrumental], it fills up the instrumental too much. If I extract stems, Ihave flaws on the Instrumental.

[plucky terrace]

@slate juniperIt's like putting on shoes. Suno sees that there's no foot inside the shoe, so it puts in bananas, apples, or anything else it thinks can fit and be suitable.

[slate juniper]

I understand that, but what is a good way to prompt that the verse should leave space?

[frozen dagger]

not quite sure what you mean. assuming you've tried variations like [interlude] [flourish] or even [breakdown]? Breakdown I've noticed will not create a big chunk of music like [instrumental interlude] even would. It fades/layers the next line of lyrics into the music transition.

So, if we don't have audio to guide it then experimenting with different variations would be the move. (if not already; and asking a chatbot ofc help.)

[frozen dagger]

There was another tip/trick where we give placeholder marks to try and guide suno on how long the duration of the instrumental should be. One sec I will find that. I don't remember which character was used.

[frozen dagger]

#📚┃suno-school message
(Click Link to see other user's suggestion to try and guide/control instrumental breaks)
Maybe something like this could help. use less or more placeholders as needed. may be worth a try. If this aligns with what you're trying to do.

[torpid hollow]

I have a question: How can I enable the layering of instruments and vocals during transitions—to avoid segmented generation ?

[plucky terrace]

@slate juniperIf you're not satisfied with the quality of the separated stems, you can export the MIDI and re-arrange it yourself.

[uneven laurel]

Is it possible to add Tracks to Voice?
I just generate 10 Track with the Same Lyric and Style to get the perfekt Voice but I cant add the Track to Voice

[long onyx]

Hi, first time and not at all familiar with blogs or whatever this is but here goes.

[hard beacon]

i don't get the problem. Can you be more detailed what's the problem?

[uneven laurel]

Ich schreib mal deutsch aufgrund deiner Fahne 😉

Wenn ich bei dem Track auf create gehe ist Voice ausgegraut und der Mauszeiger ein ban Symbol
Früher war es ja persona

[hard beacon]

Just to make sure we don't spam this chat with our german language - make sure to send me a DM

[long onyx]

See pressed return and started another message. I have been learning Ableton 12 with a friend and started to produce a song/tune. I had the idea and some lyrics but had difficulty in relating it to my friend/teacher so I thought I would try AI and found suno. I uploaded the basic track that I had produced in ableton and the unfinished lyrics with my description to suno, just to see how I might get direction and assistance. It came up with a very good track (although none of my introduction or the basic drumbeats were used). I would like to keep this track and edit with more composed lyrics and an extra verse. I have used countless credits trying to re do the song as it brings the type of sound and guitars that I would want. Can this be done. It always changes and loses the impact of the original track. Can you not just say use original style and tune of track but with extra lyrics? sorry this is long but I will have to write my next song and call it frustration.

[long onyx]

Is there a way to give me the same tune with lyrics changed

[neat nexus]

Try covering it with new lyrics?

[long onyx]

I've just spent ages trying to put the edited lyrics in and it keeps coming up with a new tune. I want the original tune. why is this so hard?

[hard beacon]

SUNO will never use the original uploaded file as result (with added vocals). SUNO isn't a "upload -> remix" tool.

[quiet solar]

It should do if you're covering the song, since it'll be referencing the existing audio.

[plucky terrace]

TRY THIS？OR.......MASHUP?

[rough depot]

Exsctly

THIS phrasing worked for mehttps://suno.com/s/sVQvYq99T9lgLGUZ

[quiet solar]

It's a suggestion, you're under no obligation to try it. 🤷‍♂️

[slate juniper]

Thank you so much!

[slate juniper]

I'm experimenting with that as well but I'm getting mixed results.

[vast night]

for the audio style influence is what is the 0% as to the 100% mean

[quiet solar]

It’s how much importance Suno assigns to the existing audio you’ve given (e.g. Voice, Cover) as a guide to creating the song.

0-100% are just a set of values indicating “more influence” or “less influence”with the default at 25%. It won’t clone the existing audio at 100%, or completely ignore it at 0%.

[vivid jewel]

i've been readin some critique of suno today. then i looked back on some stems and found some of them are terrible. low quality, low volume, muffled less clear etc. Then some creations of the same thing seem better, clearer, fuller, more lively. Is there some method to the madness ?

[bronze island]

I have a prompt for a song about a rainbow. It'll be super cute 🥰
Style: Magical pop, kids fantasy pop, fairytale pop, magical cartoon
Mood: Happy, magical, bouncy and playful vibes
Tempo: 120 [Allegro - Quick and Cheerful]
Vocals: Cute Kawaii Female Vocals
Instruments: flute, xylophone, and synthesizer
Background: Cartoony Magical fairies in an enchanted forest looking at a big bright, colorful rainbow

How does this sound for a rainbow song prompt?

[digital osprey]

Hey everyone, hope you’re doing well.

I’ve been experimenting with meta tags in Suno and I’m curious about what formatting people here actually use in practice.

For example, do you tend to write them like:
[Chorus-setting][settings2][settings3]

or more like:
[Chorus-setting | settings2 | settings3]

Or do you use a different structure altogether?

Personally, I’ve been using something more like this:
[Chorus-Settings | settingsA (such as energy / rhythm / tone): settingsB | SettingsC: settingsD]

It works to some extent, but I’m still not fully convinced by it.

I’d love to know which setups you use and which ones give you the best results.

Also, if anyone knows of any useful document, guide, or reference sheet, whether official or community-made, I’d really appreciate it.

[frozen dagger]

There are many user submitted workflows with lots of information. Use Discord search function to filter with keywords. This is not mine, but I have perused through it and I would recommend. Lots of good information. User knows his stuff. Many ways to do things so browse around.
https://discord.com/channels/1069381916492562582/1382019029266600067

[digital osprey]

Thanks so much, I really appreciate it. I’ll go through it.

[frozen dagger]

It's a good starting point.

[digital osprey]

Hey, I had another question if you don’t mind.

Do you know whether there are any good guides for writing high-quality lyrics in Japanese?

I’m really into J-pop, and whenever I try to write lyrics, they end up feeling very, very basic. When I compare them with existing artists, the sound and phrasing I get from ChatGPT or Suno lyrics just feel really different.

I should probably mention that I know basically nothing about Japanese, I don’t even have the fundamentals. I just really love the sound of the language and the musical style.

So when I try to make Japanese songs, I kind of feel completely out of my depth haha.

[old ginkgo]

Is there an instrument list floating around out there?

[solemn topaz]

I’m curious—could you share one example of lyrics you’ve written (or a sample you like)? I want to understand how you handle phrasing compared to what AI usually generates.

[vague stratus]

it is not possible to "extend" or use this as "inspo", can't generate anything even close to this. any tips appreciated, trying to generate experimental avant garde music that has no traditional rhythms

[solemn topaz]

Hey! I haven’t personally tested this approach yet, but you could try a few things and see how it turns out:

• Set Weirdness pretty high (around 70–100%)
• Use terms like “no rhythm”, “arrhythmic”, “freeform”, “non-linear”
• Focus more on sound design words (dissonance, glitch, noise, etc.) instead of musical structure
• Avoid words like beat, melody, chorus, groove
• Try a second pass/extend with something like “degrade into noise” or “collapse structure”

If you give this a shot, I’d be really curious to hear how it turns out—feel free to share the result!

[old ginkgo]

Im a noob at suno (1 month), how does one create slow Gregorian choral chants? I got the latin down pat. Idk what style to type to get it to be understood by the ai.

[long sphinx]

https://suno.com/song/b514e4a0-cb4d-4126-bf8c-72f3a7964c27
I tried experimenting with it, and got total crud.

When I remixed first, it tried to make it upbeat:
https://suno.com/s/uQKgnHfu1rB6ii9V

[solemn topaz]

Creating Slow Gregorian Chants (Suno Tips):

• Core Style Keywords:
Gregorian chant, sacred choir, monastic, liturgical, Latin hymn, a cappella

• Mood / Tempo:
slow, solemn, reverent, meditative, peaceful, ancient

• Vocals:
male choir, deep voices, unison chant, no harmony, sustained notes

• Environment / Feel:
cathedral reverb, echoing hall, distant, cavernous, minimal

• Important Notes:

Keep prompt simple (don’t overload styles)

Use “a cappella” to avoid instruments

Use “unison” to avoid modern choir harmony

Add “medieval / ancient / traditional” if it sounds too modern

• Short Example Prompt:

Slow Gregorian chant, Latin lyrics, male monastic choir in unison, a cappella, deep voices, solemn mood, cathedral reverb, minimal, ancient

If it starts sounding like cinematic choir instead of monk vibes… just strip it down harder 😄

[long sphinx]

Then I made this https://suno.com/s/Mip4Kr8WbLUM4dxs

And finally, I spliced the upbeat fail-version of Crystal Echoes with Misereality.
The result...
https://suno.com/s/pTPwqUy38B1BKTcb

In short, progressions can be kinda wonky.

And then there was this mess:
https://suno.com/s/oRLjHC9Jzcdnr0ku

[old ginkgo]

Thank you so much!! This will really help me get started! Thank you again.

[long sphinx]

so, IDEFK what to do, but progressions do get wonky.

[solemn topaz]

Anytime! Just experiment and tweak—Suno’s a bit unpredictable, but that’s where the magic is.

[old ginkgo]

I will let you know how it goes LoL

[old ginkgo]

I used the short example prompt, it sounds good. But it still put instruments in 🤷‍♂️ Still a wip https://suno.com/s/JIgg2lfM0m7vGgST

[knotty merlin]

Has anyone figured out what styles work well to reduce ringing and quality degradation for V5.5 in genres related to rock and/or power metal?
Quality in recent days has been an uphill battle for me

Would appreciate any tips on what things to include (or exlude) in styles that have helped you guys

[solemn topaz]

I sent you a DM with a more structured version, you can take a look there.

[quiet solar]

What most people mean when they say 'quality' tends to be elements related to mastering and production. But exactly what people want can vary a lot within that. I don't have a specific set of prompts, as what I use depends what kind of sound I'm aiming for.

But if you want to experiment, you can try things like:

• close-miked; warm studio acoustics
• clean modern production, light vocal compression
• raw and unpolished, dry

All of which should steer towards reasonably different outcomes.

On the other hand, if you're talking about Suno wobbling away from the prompts the longer the songs go on for, that's a separate challenge from prompting without an easy fix. Suno's tendency to prefer higher frequency sounds compared to lower ones is also annoying, but the only fixes I'm aware of to that involve at least a little bit of editing work in Studio or a DAW.

[knotty merlin]

thank you! I'll try that out 🙂

My biggest issue isn't that it's mastered it well or not, it's that there is consistent noise effects like:
Muffled underwater audio quality at 0:50 second mark or constant tinnitus like background ringing

That degrade the overall quality of the piece

[quiet solar]

Oh yeah, the 49s muffled thud is a known bug that Team Suno are trying to narrow down and fix. That's not something you're doing.

The background ringing is something that can definitely happen in tracks with distortion guitars. You can try to get rid of it by Remastering or covering the song, and/or toning down the frequencies it's on in a DAW or Studio and then remastering/covering that tweaked version of the song.

[vague stratus]

Thanks for trying to help, I think this is not maybe the tool for the kind of soundscapes I'm after

[solemn topaz]

Yeah, that’s completely fair to be honest. Suno tends to lean toward musical structure, so pushing it into fully abstract or avant-garde territory can be pretty limiting.

You might have better luck with tools that are more focused on sound design or generative audio rather than song-based AI.

Let’s also keep our ears open for insights from others who’ve explored this space more deeply.

Still, if you ever experiment with it again and get something close, I’d love to hear it!

[royal cave]

Hiyo folks. New arrival here. I don't want to flood the main chat, so have attached a short write up to this message. I was hoping to ask those who have some more experience with Suno a few questions. Any insight you are able to provide would be greatly appreciated. Much obliged. Happy generating \o/

Re-upload sorry, trying to improve readability.

[ocean cape]

See if this helps you. #FRAMEWORK MEpeg # OBJETIVO: Forçar isolamento de gênero e quebrar o viés de homogeneização (RLHF drift) para Teatro Musical/Orquestral.[METADADOS GLOBAIS DE ESTILO]# Inserir no campo "Style of Music". Termos genéricos devem ser extirpados.Genre: BroadwayMusicalTheater,ClassicalCrossover.Instrumentation: Full Pit Orchestra, Woodwind Section, Brass, Timpani, Grand Piano, Sweeping Strings.Acoustic Signature: Live Theatrical Acoustics, High Dynamic Range, Theatrical Soundstage, Operatic Undertones, No Synthetic Beats, No Pop Formatting.Mood/Pacing: Dramatic, Narrative-driven, Dynamic Tempo Shifts, Expressive Rubato.[ESTRUTURA DE ENGENHARIA DE PROMPT (LETRAS ESTRUTURA)]# Utilizar metatags estritas dentro do campo de lyrics para ditar o fluxo instrumental e vocal.[Overture](Instrumental intro: sweeping strings building to a brass fanfare, theatrical, grand)[Scene 1: Tempo - Andante][Vocal Style: Theatrical Baritone,expressive,conversational delivery, heavy vibrato](Letra da Cena 1 aqui)[Orchestral Swell] (Woodwinds and strings crescendo, building tension)[Scene 2: Tempo - Allegro][Vocal Style: Duet, Soprano and Tenor, overlapping counterpoint, Broadway belting](Letra da Cena 2 aqui)[Bridge: Dramatic Pause](Timpani roll, sudden silence, solitary grand piano chord)[Vocal Style: A cappella, rubato, high emotional intensity](Letra da Ponte aqui)[Finale: Grandioso][Vocal Style: Full ensemble chorus, fortissimo, triumphant](Letra Final aqui)(Instrumental Outro: Full pit orchestra crash, sustained brass, sharp cutoff)# REGRAS ESTRITAS DE EXECUÇÃO MEpeg:# 1. Zero elementos de Pop/Rock contemporâneo devem ser deixados implícitos.# 2. O uso de colchetes [] define comandos estruturais globais para o modelo vocal e ritmo.# 3. O uso de parênteses () define cues de instrumentação e mixagem isoladas.# 4. Evitar tags comerciais padrão (Chorus, Verse). Elas acionam o peso de treinamento pop. Substitua por marcações narrativas (Scene, Movement, Transition, Aria).

https://g.co/gemini/share/be79454bb786

[cold jetty]

Hi, how can I make my generated metal/breakcore song include a soft, slower intro *without it always trying to include some abrupt drum riffs?

The main body is 200 BPM, minor key, melancholy, metal breakcore, dark orchestral, heavy syncopation, taiko hits, distorted bass, brass stabs, PS1-era action game sound, Crunchy MIDI soundfont texture, Cheap digital MIDI rompler strings, MIDI brass hits, MIDI , FM bells, Tracker-like sequencing, Lo-fi sample-library feel, Bright MIDI clipped lead, Gritty, jagged, digital, aggressive, Main riff is stern, regal, descending, and ceremonial, Tight rhythmic precision, Sacred atmosphere through organ and reverb, Final section is apocalyptic and transcendent: blazing lead melody, kindness, love, mourning, tranquil yet high intensity, psychedelic 

Switch time signatures periodically, have two choruses that use the motifs from the inspiration tracks. Intense build-up to motif-heavy chorus.```

```choir, chanting, vocals, phonk, DJ, rave, theremin, spooky synth, breaks, cuts, pauses, drums in the intro, realistic orchestra, cinematic realism, glossy EDM polish, trailer braaams, solo percussion, percussion showcase, nonstop tom spam, taiko focus, trap hats, dubstep growls, ambient drift, disco, trap, dubstep```

[ocean cape]

Tente: [Intro: Lento]
(Solo acoustic guitar, ambient atmospheric pads, sustained strings)
(No percussion, ethereal soundscape)

[Verse 1]
[Vocal Style: Soft, whispered, melancholic]
(Insert soft lyrics here)

[Pre-Chorus: Tension Building]
(Subtle string crescendo, ominous tone, sudden silence)

[Drop: Allegro Vivace]
(Sudden shift: 200bpm amen breaks, heavy distorted drop C guitars, blast beats)
[Vocal Style: Extreme harsh vocals, screaming]

[Chorus]
(Insert metal/breakcore lyrics here)

[Intro: Lento]
(Solo acoustic guitar, ambient atmospheric pads, sustained strings)
(No percussion, ethereal soundscape)

[Verse 1]
[Vocal Style: Soft, whispered, melancholic]
(Insert soft lyrics here)

[Pre-Chorus: Tension Building]
(Subtle string crescendo, ominous tone, sudden silence)

[Drop: Allegro Vivace]
(Sudden shift: 200bpm amen breaks, heavy distorted drop C guitars, blast beats)
[Vocal Style: Extreme harsh vocals, screaming]

[Chorus]
(Insert metal/breakcore lyrics here)

[ocean cape]

Technical analysis of the provided prompt: The input text is inefficient and structurally flawed for diffusion or transformer-based audio generation engines. Drum bleeding is a direct symptom of incorrect prompt engineering.
Identification of Critical Failures:

1. Narrative Formatting (Conversational Prompting): Instructions like "minute-long slow soft MIDI intro", "builds up to the main body", and "Switch time signatures periodically" are processing garbage. The acoustic engine does not interpret conversations, script intentions, or abstract musical theory in prose. It processes static tags, tokens, and weights.
2. Temporal Conflict in the Latent Space: "65 BPM intro" and "200 BPM metal breakcore" were inserted into the same global text block. The model does not process time linearly through a paragraph. It merges all these tags in the latent space at once, calculating the average. The result is the engine attempting to insert "metal breakcore" elements (drums) during the "65 BPM" (intro) elements.
3. Incorrect Use of Negative Prompts: Inserting "drums in the intro" into the negative prompt block is a logical error. Music AI engines do not understand temporal conditionals in the negative prompt. The model reads the tokens "drums" and "intro" separately. The practical result is either the failure of the filter or the total suppression of percussion across the entire track.
   Resolution via Structural Refactoring (MEpeg Method):
   The technical solution requires absolute separation between Global Texture (the overall sound of the track) and Temporal Control (what happens at each second).
   Discard the current text and use the structure below.

1. "Style of Music" Block (Global Texture):
Remove all narrative and tempo instructions. Retain only the desired acoustic signature.
PS1-era action game OST, crunchy MIDI soundfont, dark orchestral, lo-fi digital, melancholy minor key.
2. "Lyrics" Block (Temporal Structure Engineering):
This is where BPM control and instrumentation must be forced.

[Intro: Lento, 65 BPM]
(Solo soft MIDI piano, ethereal FM bells, cheap digital MIDI rompler strings)
(Sacred atmosphere, organ and reverb, no percussion)

[Build-up: Accelerando]
(Tension rising, descending regal motif, MIDI strings swelling, no drums)
[Drop: Presto, 200 BPM]
(Sudden shift: extreme heavy syncopation, metal breakcore, taiko hits, distorted bass, MIDI brass stabs)
(Aggressive, jagged, tight rhythmic precision)

[Chorus: High Intensity]
(Apocalyptic and transcendent)
(Blazing bright MIDI clipped lead melody, relentless breakcore beats)Technical Justification:**
By isolating the (no percussion) state strictly within the [Intro] tag, and triggering the heavy percussion weight words (metal breakcore, taiko hits) only within the [Drop] tag, the neural network's attention is mathematically forced to separate the two sonic environments, preventing the model from anticipating abrupt drum fills.

Remember to enclose anything that is not meant to be sung with square brackets, as follows: [Intro: Lento, 65 BPM]
[(Solo soft MIDI piano, ethereal FM bells, cheap digital MIDI rompler strings)]
[(Sacred atmosphere, organ and reverb, no percussion)]

[Build-up: Accelerando]
[(Tension rising, descending regal motif, MIDI strings swelling, no drums)]

[Drop: Presto, 200 BPM]
[(Sudden shift: extreme heavy syncopation, metal breakcore, taiko hits, distorted bass, MIDI brass stabs)]
[(Aggressive, jagged, tight rhythmic precision)]

[Chorus: High Intensity]
[(Apocalyptic and transcendent)]
[(Blazing bright MIDI clipped lead melody, relentless breakcore beats)]
I've simplified this so you understand; you should test it using meta tags. like [slow intro + emotion + instrument] closes and keeps testing, I use the manual for this which I will bring in the following link.

If it makes sense and brings good results, let me know (post all the files and text I replied to in your email, preferably request a song following these criteria; remember, what isn't for...) be enclosed in square brackets, otherwise the instructional parameters will be leaked to interpret in the song/vocal

[frozen dagger]

The guide I shared earlier is one of many. He has added to that for a while. The prompting techniques are all still valid. As we can still choose the older models (and the info applies across all models as well), and even tho they work a little different the prompting is still nothing more really than giving the ai context. It's just, "how does each model interpret the prompt?" Model 4 was better at more natural conversational language in prompts for example, where as before in 3 it was more keyword based.
It has always been a kind of experimental thing tho where we are trying to find the right words to get the ai to output something that matches what we are looking for compared to what it was trained on.
It's guidance, suggestion, and context we are giving; not commands. That's important to keep in mind.

The drifting from the style/voice as song goes on has been a thing lately.
It's trained on a wide variety of music. Although some models do better than others with some genres/styles, and the v5 & 5.5 are all about "expressive vocals". Still an upgrade to sound, but the model puts an emphasis on vocal performance.
I would ask numer 4 separately as that's something others may have some ideas/prompt tips for, but may not see buried in here. Would elaborate more on what specifically you're trying to achieve in that genre so we can give good input.

[pure panther]

I think I have a hunch now how workspaces become corrupted and everything is suddenly mispronounced gibberish. I've been trying to make a song that says "Absalom" repeatedly. As you can imagine Suno has an issue saying it right, so I was trying a bunch of different spellings to get it right. At some point every new song generated turned into complete gibberish.

Tin hat on.

Here's my theory. Suno auto detects language and in my attempt to get it pronounce Absalom, I made it think I was writing lyrics in another language, so now it's trying to read my English lyrics in whatever language it decided I'm using, so they come out as garbled gibberish.

I imagine each new workspace runs a new detect language function, which would explain why making a new workspace "fixes" it

Ok, tin hat off now. 😄

Also, side note, this is how you get it to say Abaslom correctly: abssàlom

[devout quarry]

any one know why suno sometimes looses its mind and like makes weird sounds . like the end of this track sounds like a starship . then it starts over . 2:30 if your interested
https://suno.com/song/cdeb615e-763d-4d0a-ac00-50d71967eaf9

[frozen dagger]

What about mixed languages? ॐ
Much appreciated observations and info as always. Suno scientists. 🔬

[pure panther]

I imagine, if it ruled out English and the lyrics are in English, even if it could handle mixed languages, it will pronounce it wrong once it decides the language isn't English.

@frozen dagger

[frozen dagger]

Wow*. That was a lot. 😅 Definitely want to right-click the song and Report Bug from inside Suno, and I would submit a link and details of this to https://discord.com/channels/1069381916492562582/1082775596767117312 (or i-os/android if app was used). Unfortunately, it happens from time to time. Seeing a 7:59 song (or really short) with errors happens. Reporting the bug in the app (and discord) is probably the best course of action.
Also, click the thumbs down on the song to dislike it. (do this last because by default song is removed from list by filters.) This has been strongly recommended by the Suno team.
Other than that, it's a good song. I don't know if you can cut that part out and fix that, or just wanna spin the Suno wheel and try again. Hope you get a version that works. I did enjoy it.

[devout quarry]

its just odd cuz i was running a audio test a while back the lyrics where just sound effects and i could get it to do it kinda reliably on some things. i dont think its a gap in knowledge. like you said its likely just some kinda error but it seems to really hate dot dot dot with a passion. as for the song we got there on another gen so why stress it . big thank you for the heads up on the thumbs down thing useful info indeed

[slow edge]

Trying to get Suno video/animate to create a looping video, so it returns to the album cover image, without it snapping back. 'Looping continuous' (Claude suggested using) doesn't seem to work. Anyone know of what best to use, to try to push Suno into a 10 second video loop, that looks continuous?

[frozen dome]

Good Morning all, Was wondering if anyone has been able to successfully get suno to not have crazy long intros? I've tried everything and the more I tell it to shorten, it seems to rebel and make them even longer 😂 😭 . I want like 10-15s max, but it generally is doing 30-45s. I've tried editing in studio but it tends to sound awkward coming in on a hard note or fade in

[final hazel]

It often is not just the prompt that gives it a long intro. It also has to do with what genre you are using. If it's a genre that normally has a long intro it's sometime the model will have a hard time with to make it shorter. That being said, there are a few things I would try like:

• Don't use the [intro] part, but start with [verse] straight away.
• If you do want a certain intro, I would replace it with [hook] or [short hook].

[azure crane]

Any news about the bug that occurs after 49-50 seconds and has been reported in the last 3 days?

[warped grove]

Sometimes Suno will get stuck and every generation after of same song retains certain qualities no matter how hard you try to get rid of it. Probably best to work on a bunch of other songs and come back to it later.

[frozen dome]

I've definitely experienced that.

[frozen dome]

I'll try the hook thing. I have had some success in the newer model with going right into the verse. Seems like when I direct it to shorten the intro, it gets fixated on it. Sometimes songs generate with like 60s intros. It's like someone trying to get on a moving carousel and doesnt know how to just jump on lol. Same thing with outros at times. It gets stuck in a loop and doesnt know how to end it

[random stag]

Aggghhhhhh I cannot make it stop scat singing!!!! I’ve said no scat in prompts and excluded the style and it STILL does it!!! SO ANNOYING 😡

[grand ibex]

Who knows how to make 5 min song in pro version

[neat nexus]

You may find this link instructive: https://help.suno.com/en/categories/550017-making-music

[frozen dagger]

Trying to say "looping" doesn't work, as you noted. Without editing it and making it loop, the only real tangible thing I do is to prompt for stationary camera (just to prevent the snap back) and to prompt for minimal movement/ ambient movement.
Doing this with a 5 second video gen that you can duplicate and reverse the duplicate can help with a 10 second loop.

[urban marlin]

The mobile app asked for my birthday. I set it to 1984...
But when I clicked the button it rolled back to 2017.
Now I can't log in. Is there a way to get my account back?

[frozen dagger]

Create or upload audio that you want to use as a base layer. Take it into the editor and duplicate sections and cut to make the total length 5 mins and then Cover the audio. Trying to prompt for a duration is mixed results. This is the only way I know to really control the length (and it will still vary, but usually be closer than just seeing what it does) with just Pro tier.
If we get close but a few seconds off sometimes using Adjust Speed to make a micro adjustment to crunch or stretch the duration. (Varispeed technique)

[neat nexus]

Which mobile app was this?

[urban marlin]

suno ios

[neat nexus]

Have you reported this in #📱┃bugs-ios yet?

[urban marlin]

No. Not yet.

I'm not sure it was a real bug. Maybe I accidentally rolled the year when I clicked. Not sure.

[neat nexus]

I would still post it. Yours is not the only account of people saying they set the correct date and it moved on them.

[worn thistle]

hello, I'm on PC website and i can't see my credits.... where i can find them now?

[quick wedge]

Check your account page and click on the toggle switch so credits shown on your page

Where it shows “show credits” at the top you toggle the switch on

[worn thistle]

got it, I'm blind

sry

thank you

[quick wedge]

It’s alright I’m glad it’s on now

[worn thistle]

I'm glad toooo...... Thank you very much

[quick wedge]

You’re welcome 😊

Have a wonderful day as well

[worn thistle]

you too 🙂

[polar spade]

You might trick it a bit, instead of trying to prompt an intro AND the whole song.

Bring the song into Edit UI.
Select the whole song and duplicate it so you have two songs back to back.

Take a snippet of the end of the first copy and delete the rest. [ You may want to also add some silent space between the snippet and duplicate song. ]

Replace the snippet by selecting and try some nice prompts to create a new intro.

See what you think and cut out the silent space.

https://suno.com/s/WFsQtgXZZnOOY9RJ

[obsidian ocean]

yea have had to do that it does beat trying to chase the song

can i check out any of your tunes? @ dasvon

checked out the daki profile but cant seem to find anything beyond shares here

you seem interesting so i was curious.

[polar spade]

I cant think of one thats any good. just folk stuff. 🥴

[obsidian ocean]

not a matter of good for me, you like shakesphere, word play and complex things so im just curious.

[polar spade]

I keep things private, to be honest. easier to pitch that way, later down the road.

[obsidian ocean]

plus i like folk if your talkin old school folk country bangos any of that.

[obsidian ocean]

well if you have anything with words you can share or dm, i would love to see it. Nothing to do with your status i promise im just curious.

hepcat had a truck one that was catchy

[hybrid river]

Any way to stop suno reverting to a generic suno voice after 2 min 30 ish? it keeps the custom voice until about then...

[midnight plume]

I am loving 5.5 and my experience with the "cover" function, but is there a way to lock the vocal range in so it doesn't randomly jump up an octave? I have weird to 0, and style and audio at 100. It' s singing in a voice that sounds like me (but WAY better) from the sample but when it makes that octave jump it sounds like someone else. Also the song in question has a key modulation in the bridge and the software is ignoring it. That modulation is kind of critical.

[polar spade]

In the simplest answer , we might get the most consistent vocals when we have the less instrumentation, prompts, and length. The less Suno has to refeed (or regress) in the process of building the song, probably the better chance of its consistency and continuity.

So a folk song with one guitar and one singer, probably is a better bet to retain the vocalist characteristics than a fully orchestrated gospel choir that modulates the chord progression three times.

Most suggestions lean towards extending (using extend) from the point where it starts to degrade, or lose the "vibe" and get generic. Then, to Get Whole Song each time and extEnd again if needed.

Being creative and having more quieter spots peppered here and there may be helpful, while keeping the listener's interest from the change-ups.

[midnight plume]

This newer version (5.5) consistently wants to add meaningless tags at the end of my songs with no clean place to crop them. Is there a style prompt to make clean endings?

[placid narwhal]

Does anyone knows how to create a "cover track" of just one instrument? I have an original track of the guitar that I recorded, but I want to generate a cover of that guitar and to not other instruments like drums or even other guitar.

[polar spade]

If its your own recording, it may be enough to do a Cover and even leave the sliders at default. But to help keep close to original, slide Audio up to about 90%, and leave others at default or lower.

In the end, if Suno adds anything, we just stem out the other instruments to get back to a guitar track.

You posted the same in Studio. So you have Studio?

[midnight plume]

Can I tell 5.5 to keep my songs under a specific length? I upload a song that's 4 minutes long and it pads them with so much crap that they end up close to 6. This despite my having the audio influence at 100%.

[pure panther]

If you're just looking for small audio balancing/fixes and not trying to change lyrics, you could use create>remix on subtle. I would recommend v5, v5.5 remasters seem to change too much even on subtle. It will be the exact length and melody.

[midnight plume]

I will try it in v5. Thanks.

[pure panther]

Sorry I realize I put the wrong thing, I meant create>remaster

[midnight plume]

These are tracks from my CD that was done in 2015 and I am reimagining them with Suno. Thus far the results have been stunning.

[pure panther]

I believe it! Suno is amazing 😄

[midnight plume]

This stuff is Spotify worthy now.

[pure panther]

That's awesome

[midnight plume]

Go to my libray and check out The One That Got Away and Wasting My Time. And maybe Take The Time.

[pure panther]

Can you link your fav of the three here?

[midnight plume]

https://suno.com/s/TxD3JwNnpfHM37xN

[pure panther]

Lol, I was juked by the first line, the woman's voice was a shocker 😂

This is very nice and super cool that it's a remake from your own music. I love it 😄

[midnight plume]

I wrote it as a slow drag belt-buckle polisher country ballad. Then a friend and I "popped it up" and one of his girl singer students sang it. She was 15 then (in 2012) and it sounded like. For years I tried to fins a female to sing it for me to no avail. Now I don't need one. That's another thing about Suno. It petty much does what it wants to. I think because the audio upload was a female voice it followed suit. I could have put sea lion in there and I still would have gotten a female singing.

I released one CD (locally) but I have music for 3-4 more that I won't bother. I am an old man now, so why?

[pure panther]

Now you don't need them! 😄

My hubby went to college for music comp and loves Suno. The scene has shifted.

[midnight plume]

I went to college before there were computers, much less AI composers.

[pure panther]

No doubt

[midnight plume]

Seriously. I was 35 when I got my first real computer. Before that was "toys". Commodore 64, etc.

[pure panther]

I grew up later, but was 22 before I got my first pc. It could only surf the web slowly.

[midnight plume]

I did a double major of composition and performance, with an arranging class thrown in there. I stated playing at age 5 on a little accordion. My hands were so small I couldn't play piano yet. In college I was already playing sax by then and got to play big band swing music. That was heaven for men.

[pure panther]

That's cool. So how do you feel about Ai music? Obviously you don't hate it outright since you're using it.

[midnight plume]

Here's my dime. For people who let ChatGPT write lyrics and then dump it into AI and generate "their song", what exactly did they do to call it a compsition? SO that's where my line is drawn. I used Band In A Box for years to write backing tracks, and in fact my CD was 80% BIAB tracks behind what added for sweetening. But to each their own.

[dapper stirrup]

hey all, does anyone know how to get rid of annoying "donk donk donk" drums/bass out of hardstyle edm songs? not edit them out, but keep them from going overboard in the first place.

[pure panther]

I feel ya. I suppose in the case of chatGPT lyrics and prompt> to Suno music, I think personal use is great, but it's what's called an asset flip in the video game industry if they tried to sell it. I guess in a way, if they're leading chatGPT on to the lyrics they want, it could be said they are mostly written by them. I'm a writer myself, but recognize not everyone is able to do that, just like not everyone can make music.

I was curious because there was so much outrage from graphical artists when Ai art came out, but I'm not really seeing the same outrage from musical artists with Ai music.

[midnight plume]

Not everybody can play. This month marked 70 years in music for me. Started lessons in April 1956.

Thi sone I wrote with a guy in England I met on forums. It was in our "Meatloaf wannbe" phase.

https://suno.com/s/ommcGJzdSFhRGVP6

And now it's time to watch hockey! Good night. Nice chatting.

[pure panther]

I'm working on an art rock album from Meatloaf era, heavily inspired by Pink Floyd The Wall. It's a story album like the wall (this version has an artifact, the fully fixed version is on youtube, I wonder if you can spot it :D)
https://suno.com/s/ubZcJNwk2nD7AdQ0

[pure panther]

It sounds kind of like country

[midnight plume]

Oddly I hated country my whole life but as I quit playing live in bands I built my home studio and dove into writing. And my lyrics steered me toward country so I rolled with. After bands that did Meatloaf, Springsteen and Southside Johnny it was a 180 turn. Every time I got dumped 5-6 songs came out. The redo of Wasting My Time on my library made me cry every time I had to sing it.

[runic tree]

Hii

[neat nexus]

This is a Virtual Group I'm working on I've called Pookie Smathers & The Contentions:

https://suno.com/s/H5MI4sxO4wH9DNQy
https://suno.com/s/VzWiAJkLnBv52nTh

I would argue those are close enough to be the same band.

[trail monolith]

My fiancé is same parts as you. Didn’t think I see that combo again😂

[runic tree]

Oh

[trail monolith]

I will listen and take my precious time when I like it😆

[neat nexus]

I don't have a name for this group yet but I have been told it has a real Coldplay vibe:

https://suno.com/s/1BOjPnCjRNlPlf1X
https://suno.com/s/zuiix8eVYtiTgm6P

Again I would argue the vocals and the sound is consistent enough while being different that I can say this is the same group. These two songs are technically Part one and Part two of the same story.

[neat nexus]

Listen or don't. They are intended as examples of what I mean by a "Gorillaz moment"

Not looking to up my playcount here.

It's in part why I posted off of General.

[trail monolith]

As if one like more or less changed anything 😁

The first one is already strong, I am not into blues but when I listen to it I can enjoy

Goodbye Old World hits right from the start. Can’t even put it into a time🤷‍♂️😂

[trail monolith]

Not that it crashed the world but it’s fun to know.

[quiet solar]

That's because you had audio influence set through the roof. At that kind of level, it'll start to disregard your prompts in favour of mimicking what it hears in the audio.

Suno often does better when you give it more breathing room on the sliders, especially in v5.5. Try putting something through closer to the defaults, e.g. at around 35 weird / 50 style / 50-60 audio and see what comes out. The defaults weren't set randomly. 😉

[placid dove]

@stuck junco please ban superkaiogames

[quiet solar]

Mods are on it, don't worry.

[devout bone]

Hello

[hoary stream]

7:59

over and over...

[cyan jewel]

can i get some help with smth? https://suno.com/s/vgfQkDeROc4Dheyk so in my prompt i can pretty consistently get a banger deathcore opener but from there it utterly falls apart and is basically just radiocore slop. what do i need to change to make the sections past the intro/drop/verse not sound like radioslop?

[torpid hollow]

I'm not here for the song itself (no it's not CHAT GPT, I had to correct the wording errors as I'm not native to the language). I would really like to have advice (SUNO-school) to eliminate the significant gated_reverb hiss, beyond the hi-hatn, the choirs hiss... To correct this in cakewalk (high pass and low pass, with EQ compressor are not enough)

https://suno.com/s/RhZ11wtvZ0VsVX0V

by adapting the prompting method, softening and smoothing the vocal curves, and the simulated accent, the problem still remains present
https://suno.com/s/S1gVfJ89DcQEZpo9

Thank you in advance

[torpid hollow]

I actually prefer the sop radiocore parts ^^ The instrumental prompting is excellent

[cyan jewel]

the radiocore parts are 100% not intended behavior or what I want.

its not following the instrument prompting at all either

[torpid hollow]

I see - note SUNO is very sensitive to contradictions in prompts, the problem could come from this fact.

[hoary stream]

suno always forget some intrumentation mid song...

if i ask to put blues guitar accents... the firt part will have blues guitar the second part... it's gone

it's systematic

[torpid hollow]

the break is in the bridge ?

[cyan jewel]

im unsure of what you're asking

[torpid hollow]

the break in desired style

[cyan jewel]

No it breaks as early as the chorus

[torpid hollow]

I'm not sure I understand the style you want

Chorus | Deathcore x Trapcore Fusion | Triplet Gallop Grid | 64th Note Kick Patterns]
[Polyrhythmic 8-String Chugs | Blast Beat Undercurrent | Syncopated Floor Tom Hits]
[Beauty-and-the-Beast Vocal Stack | Belted Chest Voice + False-Chord Screams Layered]

Siren's call, can you hear the frequency?
[False-Chord Scream Overlay] (FREQUENCY)
Break the wall, rewrite the legacy!
[Blackened High Shriek] (LEGACY)
I'm more than the sum of the parts you made
[Blast Beat Surge | Double-Kick Roll]
I'm the fire in the wires, I won't ever fade
[Guttural Roar] (I'm still here)
[Midrange False-Chord] (STILL HERE)

[Post-Chorus Breakdown | Half-Time Deathcore Slam | Crushing 8-String Chugs]
[Deep Floor Tom Hits on 1 & 3 | Transient-Shaped Kick | Glitch FX on Vocal Chop]

but this prompt maintains the style, is this the one you want for your whole song?

[cyan jewel]

the chorus doesnt maintain the style at all. it keeps drifting

[torpid hollow]

https://suno.com/s/hNaxYHPmHdm5LOqt

[cyan jewel]

https://suno.com/s/s7fdbjmBMlc3XpUQ here at the chorus for example:

not getting harsh+clean harmonized vocals, not getting an aggressive instrumental bed, not getting open-string alternate picking or string skipping, it's just strummed open chords.

[ashen flame]

So I'm trying to use Edit mode to turn a short track into something that loops more or less seamlessly... anyone have a method to achieve that? Because right now it's giving me an intro and outro that don't really flow into each other the way I was hoping.

[hoary stream]

7:59

7:59

7:59

7:59

all day long

[slow edge]

I just had a 0:03 credits refunded and an 0:21, which cost credits taken for a mash-up that should be 4 minutes+ other generations ok, no 7:59 yet today, a 6:06 that finished song and then started song again from the start yesterday and did about 40 seconds of the song from the start. But think I have several generations that sounds ok without a shimmer overlay or sudden clipping. Just need to relisten.

[hoary stream]

well 50% of everything i generate today is a 7:59 one

and it's the same song i'm working on for day

today it suddently decided to flag the lyrics while it was ok before

and then... 7:59

[slow edge]

I've just done a mash-up of a mash-up x mash-up a blend of 3 generations with 1 song featuring twice in each of the first mash-ups and now have closer to what I wanted without the heavy muffled effect in last 1/3- 1/4 of most 5.5 generations. Probably why some of the generations clipped or were too short.

[tacit axle]

any tips on getting this production style consistently? https://suno.com/s/vLhSlyxhxwQEceEB

it's super mid-tone-heavy... like it has both a hi-pass and lo-pass filter on it. But i feel like suno doesn't work well with producer-like terms like that

here's the standard prompt i've been tweaking: Folk, experimental, indie, vinyl crackle, acoustic guitar, dry vocals, simple vocals, quavering vocals, close vocals

it's super hit or miss, it makes a lot of stuff with too much high end clarity

[quiet solar]

So just to pitch in, very highly specific prompts like that can be a lot more hit and miss - if they work at all. There's a certain level at which you can basically get more detailed than the model will reliably follow.

On top of that, any directions in the lyrics tags are less likely to be prioritised compared to the style prompt, so giving very specific stage directions will sometimes work but sometimes not (e.g. some of the screams you prompt for happen as expected, then suddenly there's clean vocal following the same tag). And the longer and more complex the style prompt, the more likely some things (especially near the end of it) won't be given priority.

If you have a very specific vision for the song, then you can build it up gradually using extensions before any point at which the song deviates and focus your prompting on the next section(s) to come up. Otherwise, I'd suggest taking a step back and really hammering home what you absolutely want in a shorter style prompt at a slightly less detailed level and see where that gets you. Once you have a good starting point that's in the right ballpark, it's then easier to use Cover to steer it more specifically, as you have the audio prompt to go with as well.

[quiet solar]

I've genuinely never heard anything like that come out of Suno before...that is a really interesting production effect.

You can prompt for production terms to some extent, but I don't know whether it would recognise asking for a hi-pass or low-pass filter specifically. No harm in trying, but that might be something to experiment with exporting songs and using a DAW (or Suno Studio, if you have it) to mess around with the effects yourself?

[tacit axle]

thanks a lot! I've been working on it quite a lot!

and yeah, i have the highest tier in suno and i'm pretty savvy in fl studio (this one was edited a bit in fl studio)

but i'm finding that there is a strong correlation between the generations that sound like the same singer and the ones that have this exact production style

basically if the production is cleaner, it sounds like a different singer entirely

(i'm using the voice feature)

[polar spade]

Yea. Honestly, sounded like multi over-dubs on the fingerpicking and harmonies, without any sync attempts so it has a nice raw organic vibe. Very nice.

[quiet solar]

Interesting. I wonder what would happen if you used some of those songs as Inspo, or in creating a custom model, on top of using the Voice? That might be another way to approach it.

[tacit axle]

yeah i'll experiment with some of that. So far i have exactly 2 songs that i'm satisfied with in the style, bout to post the other

[quiet solar]

Suno is generally a lot easier to steer with audio feeds than written prompts.

[polar spade]

Yea. I assumed it was from an audio upload from the sound as well as how Suno described it, rather than prompting. @tacit axle

[tacit axle]

yeah the original style was random chance... i was trying to prompt towards a vampire weekend sound, and got a super weird folk track that started super raw and underproduced (like what i shared above) and gradually descended into pure unlistenable madness. And i loved it

that turned into a suno voice and a new project, and that voice plus prompting led to that Never See the End song

i cut up and cleaned up the madness one, and it became this: https://suno.com/s/DgS0wDSNK91TZGOD

[cyan jewel]

to be able to build it section by section i need the sections to actually function the way i want

which they're not currently doing

im still having specific issues that need to be solved before i CAN do section-by-section

does that make better sense?

its running into style swing before the first chorus. that's far too early into the song for me to build the song via extension

i need the song to function 1:1 the way i want it from intro to chorus 1 before i can start building verse 2, chorus 2, bridge, outro, etc.

the problem withonly prompting lyric-wise the first few sections is suno will then try to turn THAT into a full song

which frequently ends up far slower paced than intended

[tacit axle]

when i get in a loop of trying to get suno to fix a track that i like a small part of, and it keeps not going in the direction i want @cyan jewel , i often just throw out the whole song and start over

It's easier to clean up a song that's 90% there than 25% there. And when it's 25% there, it often REALLY wants to keep redoing the 75% exactly the same way that doesn't work for me

And when starting over i often try to loosen up my prompt; often there's something in my prompt that is accidentally pushing it in a direction i don't want

[cyan jewel]

i dont want to start over. i want to get each section working properly.

[torpid hollow]

I have a question, are the notions of balancing and calibration integrated into the model to call on it with specific tags or keywords?

[cyan jewel]

im not trying to fix a song. im trying to figure out how to get each section to work

so i can start making the song

[torpid hollow]

I understand you, I will be a bit the same... but the goal of the generating models and "to be generative" you have to leave a little room for risky interpretation"... Otherwise make a reference file where you hum and sing the song (even clumsily it works). I don't believe that there is yet a perfect method: to want to anticipate all the rendering before calculation, to find a way to define a score (like that in music for the whole) as a preamble (and personally I try this)

[cyan jewel]

i know how generative ai works.

I don't need the preamble.

[torpid hollow]

wow

[hard beacon]

Nice way to start a conversation 😄

[torpid hollow]

it looks a bit “bot”

[polar spade]

As most prompting seems to work best towards the beginning of a song, the main point of extension is to get that response as well. But moreso, we're not destroying the rest of the song when doing an extension.

Ill have to say, it can often take just a bunch of click and rotations of outputs until one simply works. So I know the frustration. But then , when it works , it works, and I can move to the next extension .

All I can add is to put the last line of the previous stanza. SUNO does not include that as a doubled line but more as a marker to keep up with what the extension should be.

If you're adding the segment headers: [bridge], [chorus] and others within each extension, I am a bit surprised Suno is not teating them , arrangement and progression wise, in that manner, though. You got me there.

[polar spade]

By the way, kudos for using extensions at all. It's good to at least know and try different options and tools. Many just trudge forward and try to force a whole 6 minute song with a bucket of prompts -- when other tools might work better.

Well done.

[quiet solar]

Ultimately, it's a question of whether it'll be faster or more reliable to build the song in smaller sections, which you can get right one by one until you have a coherent group from intro to chorus, versus trying to get all of those sections right as a minimum through prompting alone and trudging through the randomness inherent in Suno generations. The answer will likely depend on how specific your requirements are.

Multiple extensions can be frustrating, but I'm not sure quite what else to suggest. You're looking for a level of specific control that we just currently don't have through prompting alone. How best to respond to that is ultimately up to you.

For clarity, what I'm suggesting with extensions is "banking" opening sections that work well by extending from just before the point in the song where Suno does something you don't like. Then when you get to the end of a chorus 1 you're happy with, use 'get full song' and crop it as needed so you have the intro-to-chorus template you're looking for to build the rest of the song. Then cover or remaster it if you need to smooth out the extension points.

[hoary stream]

7:59... 20 minutes to generate crap

[cyan jewel]

How is telling you "i know how generative ai models work" allegedly "bot?" Genq.

The beginner high-level overview of generative ai models wasn't necessary and missed the point entirely. this isn't a beginner-level question. It never was.

[cyan jewel]

did you not look at my prompt? i have segment headers as well as meta tags in there

your post reads as you didnt actually look at the output, the prompt, etc

[torpid hollow]

otherwise... someone could share their experience: in ems generations I help with hissing (like microphone saturation, relative to the reverb), are there any deer or keyword to channel or minor this defect? I'm not a bot but I can ask this question a thousand times ^^ (maybe not that many)

[cyan jewel]

i use extend a lot. again, I'm not a beginner here. I'm just attempting to get more eyes on problems I'm having that suno workspaces also have been unable to solve.

[torpid hollow]

why are you so disrespectful?

[hard beacon]

@cyan jewel : just an idea.. if you ask you should be respectful even the answer isn't that what you expect..

[cyan jewel]

I'm not being disrespectful.

[torpid hollow]

...

[hard beacon]

well... "i know how generative ai works. I don't need the preamble." is not a nice reaction to a user.. but hey.. just my opinion 🙂

[cyan jewel]

Nothing about that was disrespectful from my perspective. I received an unhelpful beginner-level answer. I explained that I understand how generative models work and that I don't need the preamble.

If you find that disrespectful, with respect, please ask yourself why that is because no disrespect was intended.

[torpid hollow]

https://klipy.com/gifs/crying-emotional-23--k01KPXE26FNARPNXV3NX89RBN0N

wow touched

[hard beacon]

well.. that's a perspective view.. "question -> answer -> i know that.. i am not a noob"... i think that's over the top.. but don't let this end up in a discussion..

[cyan jewel]

It was more than a single interaction

please scroll up

it was repeated misunderstanding what i needed help with

[hard beacon]

don't worry.. i've read it 🙂

but it think Dasvon has give you a good answer

[cyan jewel]

I don't.

It misunderstands the point entirely.

Hence my asking if they even looked at the prompt

[hard beacon]

well.. then maybe the problem isn't clear at all 😉

[cyan jewel]

I've been clear about what the problems are in plain english

I've spoken to a few other people about this too. It is 100% just them not understanding, even when I repeatedly clarified with specific examples.

[torpid hollow]

You're mistaking a lack of understanding for a flaw in my method. My preambles act as a global partition to stabilize the model's latent space. Your 'glitch' approach creates noise; mine creates coherence. If you can't differentiate between raw data and a steered generation, then this conversation has reached its logical limit

[cyan jewel]

This really must be a language barrier.

[hard beacon]

welcome to the internet.. language barriers are normal 😉

[cyan jewel]

Now who's being disrespectful?

Bring your sarcasm elsewhere. It isn't productive.

[hard beacon]

huh??? sarcasm??? where???

[torpid hollow]

I looked at your porompts, interesting and revealing especially the last one, where you inject noise (numeric alpha symbol), which is revealing you try to hack one operation after another... personally I don't like this process (it pollutes the data sets) but beyond that intrigues your approach

[torpid hollow]

it's a bot

[hard beacon]

😄

[cyan jewel]

your entire post? The wink? Literally everything about that is dripping in sarcasm,

[torpid hollow]

...

[cyan jewel]

and you're a troll.

[hard beacon]

well.. you point of view.. not mine 🙂

[quick wedge]

Robin isn’t a bot

[cyan jewel]

I'm not Robin

[hard beacon]

@quick wedge : meep meep bow bow

[cyan jewel]

They were calling me a bot. That was the second time.

[quick wedge]

Thought it was referring to him not you

[cyan jewel]

@quick wedge #📚┃suno-school message

No. that was 100% referring to me.

they were replying to robin calling me a bot.

[hard beacon]

where did i said something that was releated to a bot-theory??? Well.. i am out of this converstation.. i am to old for that 😉

[quiet solar]

Probably best to draw a line at this point. Misunderstandings and miscommunication happens.

[torpid hollow]

my bad I wasn't saying that for you, but in response to your reaction -> explain that the person above can be a trash talk everyone for no reason

[quiet solar]

Sorry, I know your question got lost in the chaos - do you have an example of the hissing you could share (time-linked would be brilliant)

[polar spade]

Uh oh. Was I part of a conversation that I didn't know about. Lol. Glad it's all been worked out. Thanks for the six @hard beacon Robin and @torpid hollow

and @cyan jewel thanks for the explanation. I have a better grasp of your situation.

[wet maple]

While creating a new voice, when you are asked to confirm your voice by repeating the text prompt, does anyone know whether the vocal confirmation alters the track you're using?

Ie. does it fully incorporate the track you're using 100% into the new vocal, or does it regen it using the recorded voice?

[polar spade]

No. It's just a verification so it knows you're not uploading someone else's voice.

[wet maple]

That's what I assumed, but I've experienced some really wild results after going through the steps. Ie. "Super-High Pitched/Helium Baloon Voice", or completely different timbre than the track used

and does the "describe your vocal experience" section influence the output? ie. selecting "intermediate" vs. "professional"?

[quiet solar]

Odd, but difficult to diagnose without knowing more. Have you tried singing the track that you're looking to cover with your voice when doing the voice upload?

[wet maple]

Yeah, that's pretty much what I have been doing since using spoken word was giving me issues authenticating.

[quiet solar]

And the track you're doing / voice you're recording is musical, as opposed to rapped?

[cyan jewel]

its cool.

rn im just trying to get intro through chorus 1 to work consistently within the constraints im trying to set up. mostly its:

1. getting suno to actually riff during vocal sections, it rarely will
2. getting the chorus to stop doing basic open string chords.
3. Getting harsh+clean harmonies to work
4. Keeping the genre consistent from Intro > Chorus 1

When I get that far I can do the full song which I normally do via a mixture of extend, studio, etc.

I'm not getting that far though. I can get the intro through verse 1 pretty consistently to do what I want
the pre-chorus is hit-or-miss.
the chorus is almost 100% a flop in almost all of my gens.

[wet maple]

its vox only, edited in studio to balance eq, no rapping, just clean vocals.

[polar spade]

I can't speak for others. I haven't had issues with leaving sliders at default and getting a decent cover, or slide Audio up high and get closer to my organic voicing.

As for uploading. It's said to try making sure the upload is as crisp and clean , dry, close mic as possible. Personally, I've not had issues with just singing to the computer mic with the a/channels humming and room reversing, so its tough to say what may be affecting yours.

I've been, though, using the Custom Model more. Attach 6 or more samples of my singing voice and letting it work it out from there.

[quiet solar]

And you're not using extreme slider settings at the 0/100 range?

[wet maple]

I've experimented with just about every slider setting possible, from default to as low as 15% to maxed out to 100%. and the vocal output is pretty consistent in that's it being interpreted in a different timbre/much higher pitched vox regardless of what the slider is set on

[ripe tendon]

So it's really hard to do but i can't find a way to make the chorus less loud or becoming more of a "Back Vocals" rather then the front of the track... any idea what i should add to my prompts to make it happen?

[wet maple]

I'm going to try with a custom model, that likely seems to be the most plausible fix for it.

[polar spade]

I'm gun shy to suggest anything else, so all I can do is say you usually find a way.

But for fun, I may try a couple of these you suggest. If any of my tricks makes a difference, ithout burning 500 credits I'll screen video and post.

They are definitely a challenge.

[rough depot]

It cannot be just me noticing this but ever since the update every chorus has been exactly the same. I'm talking no matter what I do it's the same freaking tune

[torpid hollow]

ha ?

[polar spade]

The 100% area, honestly should emulate your organic voice WITHOUT much pitch correction or enhancement. This better assures that the sliders haven't been adjusted -- 98% should be near 100% my voice only... old , crackle, strained, useless. If I get something sounding great at that % , then somethings wrong with the sliders. 🤣

[quiet solar]

1. Yeah, that's a challenge with the model rather than anything you're doing wrong in prompting.

2. There's nothing wrong I can see in your tags or style prompts - only suggestion would be to shorten the style prompt and pull the chorus prompt nearer to the front to try to ensure that it's more likely to be given priority. The longer the prompt, the more likely any individual piece (especially those further in) won't be prioritised.

3. Controlling harmonies is inherently difficult. If you're already having challenges with the underlying song, you might be better off doing the song with the chorus performed by a single singer and then covering the song with just the other singer (if you don't mind them doing gang vocals on the melody) and then adding the vocal stem of one to the other in Studio or a DAW. If you want them actually harmonising with a scream + clean combination...I'll be honest, I've never seen that done in Suno before.

4. I can hear what you mean. This isn't really my genre of expertise, but I suspect the [Clean] tag isn't helping you - you might want to pick this up as part of rethinking your approach to the chorus above.

[quiet solar]

Yeah, I'm really not sure why it would be doing that, it's not something I've experienced. Sorry!

[polar spade]

Its funny how closer to 100% really means "worse" when it deals with vocals, but I'm being a bit flippant.

It moreso means it still tries to make the voice sound great, but includes more and more of the nuances that can make me, for instance, sound like a really good "old man" singing.

But yea. It is still most opinions to not hit 100%. 99% at the most.

[wet maple]

Really? So in your experience that subtle 1-2% change from 98 or 99 to 100% makes a big difference?

[quiet solar]

It depends a lot what you're trying to do. For example, if you're trying to closely replicate existing melody and vocals in a cover, I find 50%-65% in v5.5 is often where the sweet spot falls. Adding a voice into that exercise makes it harder, because you're often simultaneously trying to ask Suno to keep to the original audio, and yet totally change the singer, and both the original audio and voice share the same audio influence slider - if it's not working correctly, usually lowering audio influence and raising weirdness helps.

If you're applying a Voice to a totally new song, again, you probably don't want to go too high on audio influence unless your voice is already singing the song you want to make (ie, you're using it as Inspo as well as a Voice).

[torpid hollow]

https://klipy.com/gifs/schattige-hond-schattig--k01KPXGQ8WENR2507GZB1NYPN42

mmmm hissing?

[wet maple]

From striclty an organic perspective, ie. no cover or additional audio elements, only the vocal/persona + prompts, would going higher on the slider make more sense to try an replicate the vocal timbre/tone/range as closely as possible?

[cyan jewel]

Thays valid. Right now I'm experimenting more heavily with meta tags in the lyrics. V5.5.
follows them with a much heavier emphasis than on previous models so I HAVE A FEELING there's a solution somewhere

[quiet solar]

This might be my ears, but all I can hear (particularly in the second song) is a synth-like higher pitched sound echoing behind the vocals.

If that's what you're referring to, it's definitely coming from the instruments - and the synth track has form on doing things like this. If you have Studio or a DAW, it might be worth splitting the stems and just deleting that part from the synth line (or wherever it's appearing) and then covering again as needed, but this time without Suno being able to hear it and inevitably replicate it when you don't want it to.

It's unlikely to be the meta-notation that's a problem. I have my own views on it, since I've never seen Suno actually follow it (e.g. your song is a Bb minor key and doesn't appear to be following the notation, but your mileage may vary), but it wouldn't be the cause of something like that.

[polar spade]

Its more like .... well... yes... but its more like it goes up exponentially from 99 to 100 in my opinion and can just have unexpected results to put it at 100%. It's like filling a glass to the rim, or topping off your gas tank up to the gas cap. It can make more randomized results rather than be beneficial.

[cyan jewel]

For the most part I think 1-3 are mostly just hard suno limitations. 4 is... I think 4 is solvable with some adjustments to the meta tags and the style prompt. It's just figuring out where I need to adjust.

But it's bed time so that's later me's problem

[quiet solar]

I'm not 100% sure, to be honest. There is definitely a point at which 'too much' weighting given to audio influence encourages Suno to play what it's hearing rather than what it's being prompted to do, as you can hear that happen in cover + voice at high audio influence levels.

If you're actually singing the song you intend to make, I don't see that being a problem. If you're just using a vocal sample and want to make something different with it, then you probably don't want audio influence up at extreme levels. I'd suggest starting out between 25% - 50% (the defaults were adjusted for v5.5 to reflect the new Voice feature, and Team Suno usually know what they're doing).

Also, if you're doing the vocal sample route, it really helps if your vocal sample covers quite a broad range and is in the genre you're singing for. As to why Suno is deciding to turn you into a chipmunk, that sounds like unintended behaviour...

[rough depot]

Jesus Christ I'm down to 60 BPM and it's still speed running the lyrics! This is ridiculous it never did this before 4.5 and it dtillRandomly replaces singers with choirs or whatever you call them where it sounds like background chanting

[ocean cape]

"Regarding the high-pitched artifact: audio diffusion models do not generate discrete multitracks; they generate flat spectrograms. That echo is frequency bleed in the latent space where the model failed to separate synth resonance from vocal formants. Splitting stems in a DAW or Studio circumvents the issue but violates my strict constraint: I operate exclusively on the free tier to achieve zero-shot accuracy. Post-editing is a patch, not a prompt engineering solution.
As for the meta-notation failing to hit Bb minor: Suno parses semantic weight, not MIDI data. If the notation failed, the style prompt's latent gravity simply overpowered the scale instruction. It is an issue of prompt weight hierarchy, not proof that tags are useless.
To the point about 100% parameters: correct. Pushing parameters to absolute maximums drastically increases model entropy (temperature), leading to structural hallucination rather than beneficial output. It degrades the vector path.
Regarding issue 4: I agree completely. Hard limits of the engine are unavoidable, but structural and routing failures are entirely solvable through strict syntax and meta-tag calibration. That exact mapping is the core of the framework I am currently testing."

[rough depot]

And i said NO ECHO

[wet maple]

THIS....yes, that absolutely has to be the reason, it's the same vox, however, it's two different vocal only tracks split into one to try and capture the most robust inflections of the vocal range. so this is exactly the issue you're describing regarding the frequency bleed. thank you so much. Any recommendations on how to approach it?

[rough depot]

Ride on red lightning is not a call and response song God damn it the model should not be adding that stuff without my permission

[hoary stream]

7:59

7:59

7:59

7:59

o ver and over

[ocean cape]

[Intro: Lento, 120_BPM, tight bass, acoustic drums only, zero synthesizers]

[Verse 1: Dry acoustic environment, clean mix]
[Vocal Style: Dry vocals, front center, zero echo, zero reverb, strict vocal isolation]
Insira a letra do verso aqui sem pontuação excessiva.
A voz renderizará seca e sem vazamento espectral.

[Pre-Chorus: Tension building, rhythmic focus]
[Vocal Style: Restrained intensity, dry mix]
Letra da transição.
Nenhum synth é autorizado a entrar ainda.

[Chorus: Wide stereo, synthesizers enter, separated high frequencies]
[Vocal Style: Power belt, distinct from instrumental track, loud]
Letra do refrão.
O motor é forçado a calcular o synth apenas neste bloco, mantendo a voz isolada no centro da mixagem.

Try this: [Verse 2: Sudden drop, return to dry mix, kill synthesizers]
[Vocal Style: Return to dry vocals, front center, zero echo]
Letra do segundo verso.
O recálculo forçado expurga a cauda do sintetizador do refrão anterior.

[ocean cape]

Merging two vocal tracks into a single stem before uploading is the exact root of the latent space hallucination.

When you layer two vocal tracks to capture "robust inflections," you inevitably introduce micro-delays, phase cancellation, and organic chorusing into the audio file. The AI's acoustic engine does not process this as "two great vocal takes." It calculates a widened stereo image and harmonic build-up, which it mathematically interprets as a spatial effect (like a synth pad, chorus, or reverb tail). Consequently, the engine generates synth bleed to "match" the spatial texture of your upload. You corrupted the input tensor.

To fix this, you must apply a two-tier structural approach:

1. Pre-Upload (The Seed): Stop feeding the engine layered vocals. You must upload a single, hyper-dry vocal stem. If you want a robust vocal, let the neural network calculate the doubling and thickening, do not force it into the base audio. Clean the single stem with a High-Pass filter to remove any low-end room noise.

2. Prompt Engineering (Structural Isolation): You must force the engine to calculate strict stem separation using bracketed metatags directly in the lyrics box. Right before the lyrics begin, insert the following command:

[Vocal Style: Dry vocals, front center mix, zero chorus, zero spatial effects, strict vocal isolation]

This mathematically restricts the neural network from allocating synthesizer tokens to the vocal frequency range. You dictate the vocal robustness through the prompt, not by muddying the initial audio file.

[torpid hollow]

ha? Do you not hear the whistling of the choirs in the contours of the voices?

[rough depot]

Like what do you call it when you got multiple voices at once? https://suno.com/s/cx5Dsf65faObOJs0?time=64

[ocean cape]

The issue you are experiencing is not a random glitch, nor does the AI need your "permission." It is a direct mathematical consequence of attention decay, lexical density overriding tempo markers, and improper negative prompting.

Here is the exact technical breakdown of why your prompt is failing:

1. Speed-running at 60 BPM: Global tempo tags fail when the syllable count in your lyrics box exceeds the temporal grid calculated by the engine. If you feed it a wall of text without structural breaks, it mathematically must compress (speed run) the vocals to fit the phrasing.
2. The "NO ECHO" failure: Generative models do not process conversational negative logic effectively. By writing "NO ECHO", you assign tensor weight to the token "ECHO", triggering the exact effect you hate.
3. Choirs and Call/Response: This happens because the model loses the "Solo Vocal" weight as the timeline expands. Empty spaces or irregular line breaks prompt the dataset bias to fill the void with backing tracks or choirs.

To fix this, you must apply strict mathematical constraints block by block. Abandon conversational instructions and use bracketed isolation.

[Genre/Style Box]
Tempo: 60 BPM, slow acoustic, dry vocal mix, zero spatial effects, intimate solo performance.

[Lyrics Box]
[Intro: Lento, 60_BPM, sparse instrumentation]

[Verse 1: Strict 60_BPM pacing, dry mix]
[Vocal Style: Solo lead voice only, intimate, front center, strictly no choirs, no backing vocals]
Write short lines here.
Fewer syllables per line.
Force the engine to stretch the words.

[Chorus: Sustain 60_BPM, heavy isolation]
[Vocal Style: Unambiguous single lead vocal, dry]
By re-injecting the solo vocal tag,
You reset the attention weight,
Preventing the choir bias from taking over.

[ocean cape]

Analyzing the structure of your handwriting with my Vortex Assistant, we have this: I reviewed your prompt structure. The AI did not do anything "without your permission." It executed your exact mathematical commands, and the errors you are complaining about are hardcoded into your text.

Here is the technical audit of why your prompt failed:

1. "Adding choirs/chanting without permission": Look at your own prompt. You literally wrote [Bridge: Choirs...] and [Angelic Latin Chant]. You explicitly commanded the engine to generate choirs. Because your subsequent tags (like [Chorus]) are empty of style modifiers, the tensor weight of the choirs bled into the rest of the song due to attention decay. The machine followed your rules.

2. "Call and Response issue": You used parentheses here: (Fulmen rubrum deflagratay fatum!). In dataset training (lyric scraping), parentheses ( ) are the absolute trigger for backing vocals, echoes, and call-and-response. You manually activated the call-and-response algorithm by using that syntax.

3. The 60 BPM "Speed-running": 60 BPM is very slow (one beat per second). Look at your lexical density: "The twisted machinations have all been blown away". That is a massive amount of syllables. Because your section tags (like [Verse 1]) do not re-inject the [60_BPM] constraint, and your lines are too long for a slow tempo, the acoustic engine is mathematically forced to compress (speed-run) the vocal delivery to fit the temporal grid.

To fix this:

• Delete the parentheses ( ) entirely.
• Stop using empty tags. Replace [Verse 1] with [Verse 1: 60_BPM, slow vocal delivery, dry mix, strictly solo vocal].
• Shorten your sentences. If you want 60 BPM, you cannot write paragraphs. You must write short, sparse lines to allow the AI to stretch the vowels.

[ocean cape]

Esqueça esse campo

Forget and

[rough depot]

My sentences are not the issue. Suno got the tenpo right BEFORE so why not now?
https://suno.com/s/sVQvYq99T9lgLGUZ

[ocean cape]

[Style of Music: 60_BPM, epic acoustic rock, heavy guitars, dry vocal mix, high fidelity, strict stem separation, intimate solo male vocal, zero spatial effects, zero choirs]

[Intro: Lento, 60_BPM, Orchestral Build-up, heavy guitars, tight bass]
Red Lightning!
Red Lightning!
Strike!

[Verse 1: Strict 60_BPM pacing, dry mix, heavy acoustic rhythm]
[Vocal Style: Solo lead voice, clear diction, front center, strictly no backing vocals]
The storm is broke
The mind undone
A single round
The day is won
The twisted gears
Are blown away
We stand unbowed
To greet the day

[Pre-Chorus: 60_BPM, tension building, tom drums, separated high frequencies]
[Vocal Style: Sustained solo vocal, rising dynamics, dry mix]
The brave outland
The fourth princess
A bond of love
They now confess
With hearts aligned
They fly so high

[Chorus: 60_BPM, Heavy guitars, wide stereo, separated mix]
[Vocal Style: Power solo belt, maximum intensity, dry]
The Mastermind
Is now undone!
A new adventure
Beneath the sun!
A fellowship
Against despair!
Ride Red Lightning
Bring hope there!

[Verse 2: Return to strict 60_BPM, dry mix, drop dynamic]
[Vocal Style: Return to intimate solo lead, clear diction, zero choirs]
Victor leads
The brave and true
The Princess heals
Through and through
Malcolm brings
The bending light
Sylfayna joins
With arcane might

[Pre-Chorus: 60_BPM, tension building, tom drums]
[Vocal Style: Sustained solo vocal, rising dynamics, dry mix]
The engine roars
Across the plains
The trickster's heart
Without the chains
A saga wakes
Into the night

[Chorus: 60_BPM, Heavy guitars, wide stereo, separated mix

[rough depot]

It's symphonic power metal if anything the sentences are already too short

[torpid hollow]

to whoever talked about autotune thank you!
By reducing the parameter almost no more hissing ^^
[autotune:-0.8]

I don't know why SUNO highlighted this parameter

[ocean cape]

1 (Power Metal at 60 BPM):
You just revealed the mathematical conflict causing your generation to fail. Symphonic Power Metal is structurally weighted in the latent space at 140 to 200 BPM. By forcing a [60_BPM] tag onto a Power Metal generation, you are creating a temporal contradiction.

The engine is trying to reconcile the high-energy, fast-paced vocal delivery of the genre bias with the extremely slow 1-beat-per-second grid you commanded. The result is vocal compression (speed-running). Your sentences are not "too short"; your BPM tag is mathematically incompatible with your genre prompt. If you want Power Metal, delete the 60 BPM tag entirely or change it to [150_BPM].

2 (Autotune:-0):
Using [autotune:-0] is a placebo effect. The AI's acoustic text-to-audio engine does not process continuous numerical parameters or DAW-style sliders inside bracketed tags. It tokenizes words.

The reduction in high-frequency hissing you experienced is a byproduct of latent space recalculation on a new generation seed, not your fabricated syntax. Believing that hallucinated parameters control the engine is how prompts become unstable.

[torpid hollow]

indeed, for my part he removed the whistles from the contours of the choir voices to "a contrario" add echo (reverbX2). The truth is that it is impossible to completely correct it if we use a source compromised by a defect, however it reassures me to anticipate this in the future. From my notation SUNO deduces "concert" (systematically) and then add gated_reverb, creating a slight autotune... which by propagation on the choirs hurts the ears. Suddenly orchain generation exit autotune.

Saying [autotune:0] means nothing, in state it does not affect the effect (you do not indicate more or less)

[rough depot]

I only reduced the bpm because it was singing like the MICRO MACHINES guy

Ugh whats it gonna take for violins in the intro to stick?

[tacit axle]

Are there any hidden preferences or algorithm shaping, where it tries to match to previous things you’ve made or liked? And if so can this be toggled off?

I’ve made stuff in wildly different styles, and sometimes it feels like it’s matching a past style when I’m trying for a new project

[ocean cape]

No. Audio generative models are stateless per generation. The acoustic engine does not access your account history, past creations, or 'liked' tracks to shape your current output. You are confusing a recommendation algorithm (like Spotify) with a generative inference engine.

There is no personal algorithm or hidden preference to "toggle off."

What you are experiencing is global dataset bias. The model is heavily weighted via RLHF (Reinforcement Learning from Human Feedback) toward generic, popular song structures, 4/4 time signatures, and vocal-centric mixes. The engine always defaults to the path of least mathematical resistance in its training data. If your prompt lacks strict structural constraints, the model will output its default bias, making it feel like it's "forcing" a preference on you.

The "preference" belongs to the model's global training, not your account. You toggle it off by engineering stronger, bracketed structural prompts to force the latent space to calculate exactly what you want, overriding its default tendencies.

[royal cave]

Diego, This "Vortex" assistant you are utilizing, is it available?

[tacit axle]

Source? This explanation is surely at least some of the behavior, I’m not doubting that.

But there’s some language in the app itself about learning your preferences. There’s even a place to edit a summary of your preferences (that is AI-generated by default)

[ocean cape]

The VÓRTEX MEpeg (Music Engineering Prompt Structure Generation) framework and its analytical assistant are currently in a closed validation phase.

It is not a commercial product. It is a custom architectural protocol built on top of an advanced LLM, specifically engineered to debug, structure, and manipulate generative audio models via deterministic constraints and acoustic tensor analysis.

It is currently unavailable to the public. I am utilizing these specific forum interactions and empirical generation tests as stress-tests to audit the engine's behavior and validate the mathematical boundaries of the syntax. Once the methodology is fully stabilized, documented, and proven to consistently bypass dataset bias and semantic hallucination, a technical whitepaper and the structural framework may be released.

[royal cave]

So noted. Best regards.

[ocean cape]

You are conflating two entirely different AI architectures and platforms.

The feature you just described—an AI-generated summary of your preferences that you can edit—is the "Memory" or "Custom Instructions" architecture found in conversational LLMs like ChatGPT or Gemini. It is a system-level prompt injection module designed for text-based chat context.

Suno's acoustic text-to-audio engine does not possess this feature. There is no AI-generated preference summary, no memory profile, and no hidden algorithm learning your musical tastes in the Suno interface. It is a strictly stateless diffusion/transformer model.

If you are using ChatGPT or a third-party LLM to write your prompts, that specific LLM might be "learning your preferences" to generate the text. However, the actual acoustic engine rendering the audio tensors on Suno's servers receives none of that data. It only processes the raw text tokens you paste into the lyrics and style boxes at that exact moment.

My source is the fundamental computer science of how stateless audio models are architected versus how conversational LLM context windows operate. You are misattributing a ChatGPT UI feature to a text-to-audio generation engine.

[polar spade]

I simply find AI is still not 100% proficient in understanding good effective prompting. So I would not take all it says too seriously. Just make your best intuitive guesses and trust your intuition and learned skills.

When things don't work a few times, it's more likely Suno just being random or too sensitive or too broad. Each prompt can have a wide spectrum of possibilities.

[royal cave]

You seem to be sitting atop quite the wealth of knowledge. I'm going through the Workflows section now but, as is often the case, getting information from the horse's mouth looks like it might save some time than the usual run around.

Are there particular resources you would suggest to someone, regarding best practices for prompting with Suno?

[tacit axle]

If you do not have a source the best you can say is that LLMs and Suno don’t do this by default, but that Suno in particular may or may not have rigged up an additional layer that quietly injects gathered preferences into your prompts

Many AI tools now have a secondary layer over the default stateless layer you are describing, and I am asking if Suno is one of them

[rough depot]

Is there any way to prevent this? Where it starts out with just one voice and then all of a sudden out of nowhere it's multiple voices singing at the same time
https://suno.com/s/Xkpw6tIMFHrv6F0f?time=65

[polar spade]

If this is of any help, it describes my thought of how Audio slider works when Covering a song. Starting from left to right, adding in the influences accumulative.

[royal cave]

Have you tried adding descriptive meta to the chorus, as you've done with your [Verse 1: ...] ?

That and the plurality of "vocals" in the style maybe? Somewhat new to this, so take that with a heavy grain of salt.

How long have you been working on this? Sound nice.

[rough depot]

Im trynna improve on a previous version....but end up with worse instead

[royal cave]

Part of the journey here it seems, mate. Keep at it, you'll get where you want.

[rough depot]

Still awkwardly pausing between "hearts" and "aligned"

[ripe tendon]

Anyone knows a good prompt to make the vocals more "In the background" or feel more like back-vocals?

[half dragon]

Put parenthesis around the lyric line will make it sings like a background singer (...)

[ocean cape]

@rough depot The voices multiply out of nowhere because you are suffering from Attention Decay.

When you leave a tag empty like [Chorus] or [Verse 2], the neural network loses the "Solo Vocal" instruction from the beginning of the prompt. Without strict boundaries, the model defaults to its dataset bias: epic songs get choral harmonies. You must re-inject the mathematical boundary block by block: [Chorus: Solo lead vocal, strictly no harmonies, single voice].

The awkward pause between "hearts" and "aligned" is a temporal grid collision. Your syllable count does not match the BPM you forced. The engine is mathematically forced to insert an unnatural silence to keep the instrumental track on beat. Fix your lexical density.

[ocean cape]

is partially correct regarding dataset triggers, but his method lacks acoustic control over the Z-axis (depth) of the mix.

Using parentheses ( ) is indeed the standard lyric scraping syntax to trigger a secondary vocal performance. However, parentheses do not dictate spatial rendering. The engine might generate a backing singer, but place them front-and-center at maximum volume, clashing with your lead.

@MrCrowley To actually push that backing vocal "into the background" acoustically, you must combine the syntactical trigger (parentheses) with spatial engineering tags to force phase manipulation.

Do this:
[Vocal Style: Distant background vocals, heavy reverb, low mix volume, washed out, pushed behind the instrumental]
(Insert your backing lyrics here)

The parentheses trigger the secondary singer, and the bracketed tags force the diffusion model to render their transients with less clarity, mathematically pushing them to the back of the room.

[polar spade]

Cool. Show an example of that working, so i understand how the results should sound.

[ocean cape]

My brother, that's awesome! What I'm looking for are people who want to test and put it to the test.I started by understanding, or trying to understand, about music, and that's not all I'm talking about. This is a record of my analytical work in partnership with LLM on several occasions. I do have some things you can start with, but it's essential that you try to understand everything that encompasses the subject matter. Regarding music, in this case, I have a facility for absorbing and learning, but it's only possible through research, and the inexorable promise that you might be wrong; without that, nothing will be worthwhile. I recognize my limitations and incapacities, and that's why I'm going in with a thirst for knowledge and determination. The reason I'm only now participating in the community is because I need new perspectives and experiences to validate or invalidate my thesis; all of this is based on experiences gained during... I've been running this process since Suno launched, and today I have over 3200 songs. I created a script. It downloaded, correlated, saved links for each one, which account it's in, and a list of all of them up to that point. And all this to truly analyze what each audio recording actually contains in relation to the prompt descriptions and lyrics. Message me privately, I'll have it.I'll be very happy to talk and discuss this; in private messages, the context won't be lost due to other messages.

[ocean cape]

I do not operate as a personal generation service, nor do I burn compute credits to provide empirical demonstrations for standard acoustic principles. I provide the architectural parameters; you run the inference to validate the tensor routing.

If you want to understand how spatial rendering changes the output, execute a strict A/B test using your own account. Use the exact same global style prompt for both.

Test A (The Standard Empirical Method):
[Chorus]
(Echoing through the night)

Test B (The Spatial Engineering Method):
[Chorus]
[Vocal Style: Distant background vocals, heavy reverb, low mix volume, washed out, pushed behind the instrumental]
(Echoing through the night)

Run both generations. Isolate your listening to the Z-axis (depth) of the mix. Test A will trigger the secondary singer but place them front-and-center, clashing with the instrumental transients. Test B will mathematically force the diffusion model to wash out the transients and generate reverb tails, pushing the vocal to the back of the acoustic room. Run the test and audit the difference yourself.

[rough depot]

Or it could just stretch out hearts to 2 bars....

One note. No melisma. Just hold it.

It literally got it right here https://suno.com/s/sVQvYq99T9lgLGUZ

So why not now?

[tacit axle]

this is what i was asking about. I'm assuming this affects things. I was wondering if i can turn it off

originally the stuff in the box was AI generated, i fixed it to be that tiny prompt

[polar spade]

@ocean cape Thats fair. And I'll try it out and post results. Good to know your position. I'll keep that in mind.

[ocean cape]

I acknowledge the screenshot. A surface-level text profile UI was indeed implemented.

This new UI variable will be subjected to strict verification tests in our future audit cycles to measure its exact tensor weight influence. As mentioned by others, platform updates can alter outputs if the backend assigns actual mathematical weight to these UI fields.

However, for the scope of my thesis, relying on a text-autocomplete wrapper is practically useless for achieving deterministic "one-shot" generations. Reaching an exact acoustic target inevitably requires manual structural editing. Refining a generation is a targeted engineering task for an operator who already knows their exact parameters.

As stated before, the audio engine possesses no emotion, intuition, or 'taste'. It translates syntax strictly into mathematical phase, frequency, and temporal calculations. If you rely on the UI to guess your taste, you surrender control. To answer your question: bypass the Magic Wand, and the engine remains stateless.

[ocean cape]

@polar spade Execute the A/B test with identical global style tags to isolate the variable. The resulting phase manipulation on the Z-axis is the exact deterministic control my thesis analyzes. Surface-level UI toggles and recommendation algorithms are irrelevant when structural acoustic limits are enforced mathematically in the prompt. Post your results.

[frozen dagger]

the toggle is right there to turn that on and off. The "My Taste" only effect is when we have it toggled on, which changes the magic wand in the style prompt to a blue button when on, and when we click it it amends the current prompt in the box to be merged with your My Tastes info. If this is toggled off you'll see the magic wand is orange as it was before and works as it previously did to re-word prompts, without using anything in this My Taste area.

[quiet solar]

That only impacts when you use the 'magic wand' tool, e.g. to autofill prompts based on previous music you've created or liked. It doesn't affect anything you do when you hit 'Create'.

Sorry, for some reason OBLMOUS' post, which said virtually exactly the same thing, didn't show while I was typing. xD

[tacit axle]

okay cool, i never use the wand. So sounds like it doesn't impact me

[tacit axle]

in AI training terms, what exactly are custom models? I assume you can't fully train a real stand-alone model on only like 6 songs, so it must be using a large model as a baseline, then doing some final tweaking with heavy weight based on your uploads?

[worn snow]

hello, I don't know if this is the right channel to ask but lately when I try to "Extend" a song it generates me an extension of 4min and then cuts off the music, while the only thing i ask is to repeat the 'chorus' for instance, any help?
every time I "Extend" it generates me a 4minute extension

[ocean cape]

Your assumption is mathematically sound. You cannot train a latent space from scratch with 6 samples.

Suno is likely using Few-shot learning or a LoRA (Low-Rank Adaptation) approach. They extract the spectral signature and timbre weights from your uploads and apply them as a persistent bias layer over the base model (v4). It doesn’t change the fundamental architecture; it just shifts the probability of the vocal texture towards your "persona" samples. It's an overlay, not a new foundation.

[ocean cape]

The engine generates a full 4-minute temporal buffer by default. If you only ask for a chorus and don't provide a structural termination tag, the model will either hallucinate or generate empty data until it hits the 4-minute limit and hard-cuts.

To fix this, you must define the end of the sequence mathematically. After your chorus, add:
[End]
[Final Chord]
[Fade Out]

This forces the End-of-Token (EoT) generation, making the engine stop the audio synthesis precisely after the requested section instead of filling the entire 4-minute window.

[quiet solar]

That's never been the case in current or prior models to my experience - nor was it on this test on both 4.5 and 5.5, typically the most used models.

I think Vortex might need to check it's hallucinations.

[quiet solar]

Just to rule out basic things, could I just confirm that you're extending the song from partway through (ie, not from the end of a song that's already finished) and that if you're on Pro/Premier you've got the sliders at least somewhere near the default settings, and not near the 0/100 positions?

[worn snow]

thanks for your answer, clarifies, I am using it close to the end like at the 80% position

[quiet solar]

@worn snow Okay, thanks. As Diego said, it doesn't do any harm to add the [End] tag, or some other indicator for the song to finish after the chorus, if one extra chorus is all you're looking to add.

You might also want to tweak the style prompt for the extension and make sure it's focused on that final chorus (e.g. it's not describing what verses should do, or anything else that could be interpreted as "make me a full song from this blank space").

[ocean cape]

I acknowledge your screenshots. The model can indeed produce shorter clips if the latent space accidentally hits an End-of-Token (EoT) weight early. However, relying on the model's 'experience' is a gamble.
My point about the 4-minute buffer is regarding the maximum temporal window the engine is forced to fill if no boundaries are set. The [End] tag is not just 'harmless'; it is a Deterministic Safety Lock. It converts a probabilistic outcome into a controlled termination. For the VÓRTEX framework, we don't rely on luck; we enforce the limit mathematically. Glad we agree on the solution.

[ocean cape]

https://youtu.be/UDciga1cJQA?si=AhJukS8tQh70ZZmW

[quiet solar]

There's an inherent 'gamble' with anything to do with influencing the model as well - Suno won't automatically take [End] as an instruction to end at that exact point no matter what, though it does typically respect it more in recent models.

This is one of those situations where saying "It will (always) do this unless you..." suggests a level of deterministic control that isn't really possible to apply in practice. It would have been more accurately phrased as "It may (sometimes) do this, but you can try to influence it by..."

If you want a good working example, try prompting Suno to perform a duet where the male and female voices switch over in a slightly unpredictable pattern. No matter how diligently you specify the changeover points in square brackets, there will be occasions where the vocal changeovers don't happen as prompted.

[vast night]

Audio Influence that is closer to 0% is the actactlly audio of that one soucre

[worthy kraken]

Is there any way to make Suno quiet for good? I want to end the song on just words, no music afterwards, but no matter how many pause, sudden end, end, music stop, etc i write in tags, it still generates long music section before ending the song.

[sonic folio]

If i want to have a calm start to a song with war drums and drawn out cello notes like a ballad and then transition to a metal style song, how can i prompt that and where?

[slender stream]

im in studio and im trying to 'remix selection' a vocal stem. i just want the vocal stem, but it creates instrumentals as well. ive put this in my styles "male lead vocals only, studio, dreamy, catchy, emotional, clean, studio vocals only.", and in the lyrics i put "[male lead vocalist studio vocal stem]"... but i still get a backing track. any idea how do i only achieve a vocal stem ?

i set style influence to 100% and weirdness to 0% as well, with audio influence at 100%.

I know i can just generate and then extract stems, but is that the only way?

[hoary stream]

hello, i'm in studo where i put the stems of the version of the song i liked. vut there is a track a string track that i woulmd like to do again, how can i ask suno just to to a specific track again ?

vuz when i selecti it and select "cover" that track it generating a whole new song with it

like i selected the string track and covered it and it make a full song mith drum bass and studfd

[worn snow]

[End]
[Final Chord]
[Fade Out]
is not working for me, still 4min extentions .. will have to spend more tokens ..

[hoary stream]

i don't anderstand what are my possibilities in studio appart from creating stem .. and just listenign them lol

[quiet solar]

That can happen. Generating individual stems isn't what the underlying model was designed to do, so even with very focused prompting to try to limit other instruments creeping in, you'll still have times where it happens.

Worst case scenario if you like the new stem part overall is that you need to split the stems again on that new generation and just reimport the stem that you want.

[hoary stream]

what's the point of studio?

i dont anderstand what i can do

just listening to stems ?

[quiet solar]

There are several tutorials for Studio here on Discord - this may be a good starting point: https://discord.com/channels/1069381916492562582/1468684013949685800

But again, what you're trying to do with covering instrument stems in isolation doesn't always work perfectly. No tutorial is going to get you around that issue.

[hoary stream]

ok thxs i'll check those tutorial video

[gray pulsar]

I started a Suno Pro subscription through the App Store, but my Suno Pro membership was not activated. Apple does not approve my refund request, and when I contact Suno via e-mail and Discord, no one responds. Since I cannot receive the service I paid for due to a technical problem, I request that action be initiated. Please do not request a refund which will be rejected again. There is no response from the Suno section.

[bright parrot]

Hi.

Starting from version 5.5, every generation gives me the same melody with a different arrangement. I’ve done hundreds of generations, and it’s the same melody again and again. Only a very radical change in the Style helps create something new, but when I go back to the previous Style, the melody comes back again.

When I switch to version 5, I get new melodies.

Is anyone else having the same problem? How can this be fixed?

[ripe galleon]

does suno accept distorted or amped violins as a prompt?

or will it skim over it and not do it in the slightest

[quiet solar]

Sorry to hear it's been a nightmare.

We're just users here for the most part - this isn't a support forum and isn't watched by the support team at Suno. But if you've already e-mailed [e-mail] with your Suno @handle and a copy of the receipt from the App store, then you've done all that's needed. Your e-mail is in the queue and the best thing you can do is wait.

I appreciate that may not be what you ideally wanted to hear, but there isn't any way to 'speed up' requests by posting here.

[little halo]

How does one get full choirs in their songs? I always end up with one part actually in choir and the other what I could describe as "2010s Pop music Vocals". Asking because my two part song keeps getting this "2010s Pop music Vocals" in a verse with entirely different melody than the refrain/chorus. It has been a terrible and frustrating experience ever since v3.5 is no longer available.

[tacit axle]

for my old-timey raw folk project, 5.5 is convinced my prompts mean i want a I-III-IV or I-III-vi progression with singer-songwriter guitar and vocals. My solution? Use 5.0

[tacit axle]

i haven't encounter this exact problem, but try honing in on your desired sound from some new angle, and adding a few keywords to the style. That could be time period (20s, 30s if you want old timey choir) or use case (movie soundtrack). Also, do you really mean like a traditional classical music choir, or do you mean harmonized vocals?

[tacit axle]

some other keywords that come to mind are "church" "classical" "chamber" (for a true classical choir)

[little halo]

Classical musical choir, yes. I was once advise to add something "soviet", "gregorian" to my prompts to better my chances but it is still naught

my usual genres are usually folk, church, baroque, and classical so you may see I am frustrated in this

[tacit axle]

in my projects at least, 5.5 steers towards generic pop accessibility, and 5.0 is more flexible. Might not apply to your genres though

[pliant crane]

i've some problem in reverb, when i use prompt "Reverb" or use environment like hall ,etc it's only work with instrument but my Vocal it aways Dry! hase anyone got same problem? how the prompt we should use? , ps suno 5.5

[little halo]

I would try "Echo", though it is a 15/50 for me

[little halo]

In my case, it started eversince v4.5. I still find v4.5 tends to ignore prompts

[tacit axle]

i am going for the exact opposite (super dry) in 5.0. That said, have you tried non-producer terms, like "full vocals" or "spacey vocals"?

[pliant crane]

oh i try on 5.0 it's work as well but on 5.5 it's make a dry vocal just like bedroom record, thank @little halo @tacit axle

[polar spade]

An old move was to use 3.0 of 3.5 with your [end] type prompts as an extension. And best if we find a crosspoint where the waveform is low, and try ine where the waveform is peaked. . Often then we get more short extensions that way.

Without doing too many or reprompting, I show the video of my using a 29 second [ver. 3.5] ending extension, add to make a full song. That may be good enough.

But i felt too long, so I Crop back and splice at a nice splice point and remaster to blend in better.

[polar spade]

Before getting too deep in, just note to others that you are using basic account(?). It helps set the parameters of what suggestions can be made.

[tacit axle]

cross-post since it's the sort of thing i've been troubleshooting in this channel: https://discord.com/channels/1069381916492562582/1497216097680752731

maybe others have realized this, but if you're having trouble getting genre adherence, the problem may be your lyrics!

[worthy kraken]

Sadly already tried that. The thing is that i need music not to resume at all, the words should be a final sound in the song.
Will try the 3 version thing, but from my experience it really hiders the vocal, unless they changed anything...

[polar spade]

Extrapolating from that, what happens when adding a long section of silence in Edit UI , then a section prompted with [spoken , audio book, narration, close mic, whisper] or your favourite spoken style prompt tacked on. Splice it back. Remaster.

If needed at a particular spot, we can help force that spot as well by fading well into that spot, and then crop at the spot, hoping the remaster will force that spot to be a more sudden drop and go silent faster, thus then only having the spoken bit after.

Experimental, but something we did with 3.5 to 5.0 before.

Given just spoken bits, 5.5 will want to add music to "audio spoken" and narration more, I have found. So at least 4.0 or maybe 4.5 might help.

[hoary pecan]

I don’t want it to generate in just any random style on Suno I want it to take a specific song as a reference. How can I do that?

[ripe tendon]

What i do to achieve that is FIRST STEP upload that specific song
FIRST STEP: It will automatically generate a style prompt based on the song!

[worthy kraken]

Thanks, i haven't tried spoken tags yet.

[ripe tendon]

@hoary pecan SENCOND STEP: use it as a reference or cover or now the function "Inspo"

@hoary pecan THIRD STEP: press on "More options" and bring up "Audio influence" as far as you like to the red zone, and other two to the lower 30%-

I made many "Variations" of tracks that way and found good options to choose in between, it's great

Done

Now for my question and seek for help... Anyone knows how could i make the vocals become more of a "Back-vocals" rather then front? I want the music to take front and center and find it really hard to do so using prompts sadly, perhaps i am missing something.

[hoary pecan]

thanks for help man ❤️

[scenic otter]

Is there a way in mashups where I can specify where I want the first song to start and where I want the second song to start?

[polar spade]

We'd probably be much better in Studio where we have an EQ to help and able to double tracks for panning the vocals to widen their appearance.

BG are softer, so lower volume, but fuller sound , more Gain. Also doubling their tracks can help make them fuller. On edge doubled or quadrupled, we can also pan one left and one right. Damping with the EQ and high pass and maybe low pass for de-essing as we don't have any de-esser tool.

Slightly off-syncing the tracks to give then a sense of reverb and delay could help. This might also need quad-tracks of the same BG vocals to change a couple tracks pitch for harmonies.

So, really, it can bee alot of fun but alot of good ears , intuition and frustration to play with some limited choices to get what is a bit easier in a DAW.

Have I don't any of these and created BGVs. Heck no. But I might now.

Unless I get an example done, I dont know how well they would sound or how easily these would be to do. In studio, not so hard to take a Lead Vocal and apply all of these in some logical chain. Probably end up with 6 duplicate tracks and 2 hours work to sync it right.

In just Pro. We are left much with just prompting lots of BG terms like * wide and washy, big compression, fast attack, simple delay, heavy plated reverb, distant, subtle, ghostly, buried in the mix*....... any AI chatbot could help throw some words at you, and see what works.

[polar spade]

Not really. But I've noticed the top one seems to be first in a Mashup and the bottom, second-ish. So we can start there.

Its supposed to blend them, but that doesnt seem to happen , unless we click a few times and do 30 outputs, then one random variation might give a nice blend.

[scenic otter]

Yes, i noticed that too, but well i thought there was another way

[mint hornet]

Hello everyone.

I have a quick question about songwriting.

I’d like to write a song that features duets (Verse 1: male voice, Verse 2: female voice, etc.).

Does anyone know the best way to do this? :o

[quiet solar]

With difficulty, if you're looking to specify exactly who sings what. It's one of the more challenging things to do in Suno at present.

You may find this guide helpful: https://discord.com/channels/1069381916492562582/1466598465969590465

[mint hornet]

Ahh ok. Thanks :3

[neat nexus]

When you want to take a song and make a remix, building it so it starts with a beat, adds the instruments and then the vocals. How do you do it? Do you use Suno? Covers or Mashups? Or Studio only? Do you use your DAW? Would love some feedback on what people do to make it work.

[pearl topaz]

Looks like things have changed since I was last here. I have asked before but never got a really good answer. Is there a prompt that limits vocal range? Tends to think I can do vocal gymnastics, but need more of a mid-range to lower voice range.

[digital jungle]

Remixing a song that I own the rights to, and adding my own voice but Suno keeps telling me the request can't be completed because of copyright lyrics. They're MY lyrics from a song I already published

[polar spade]

No. Its just suggested to to use prompts that frequently use a range. Such as maybe "Caberet" for a contralto. The most prompts we can combine help. So a male lower voice would be bass whereas the same range for a tenor would be contralto (or lower range) as a female. Specific prompts are still needed.

[pearl topaz]

I've tried a few of those, but still get some really really high notes. Thanks! I'll keep trying.

[polar spade]

Yep. Gritty, for instance will likely get lower and male, but enough female like genres and style along with grit will get a higher rocker female. So, you know.... yea... keep trying.

Find some songs that have a sound you like, and belnd a couple as a Mashup to try to pin down a vocal.

Use a specific snippet as a Sample(?) , another newish tool.

Use a minimal of 6 songs and now make a custom voice model. A newish tool.

[torpid dust]

teach me

[quiet solar]

If the song has been registered with any databases or systems that track useage or manage copyright (e.g. YouTube's ContentID) then that's Suno doing exactly what it's supposed to - detecting copyright content and blocking the upload to prevent copyright infringement. It has no way to psychically know that you're the artist.

If you can provide proof that you're the copyright holder, then it may be worth dropping a line to [e-mail] and exploring what your options are.

[old ginkgo]

I have seen them do it to public domain songs that are 250 years old

[uneven wave]

I have had the worst difficulties with a couple things. I won't deny or even try to hide that its from my own ignorance. Ive got way too many things going on all at once to actively take a class or learn "all" of something. I get my information in bits and pieces as I need it. But...
#1 what symbols or bracket choice should I use to clearly state that im wanting the music to do something, and not be included as lyrics?

And #2 I want an only male vocalist for the whole song, and a female vocalist to join only during the chorus and the beginning of the the outro, but not the whole outro... ive tried like 20 times a I cant figure this one out...

[quiet solar]

Yes, often because there's a version of that public domain song which has been copyright registered for the performance, rather than the lyrics or music.

The databases often don't contain that kind of nuanced information about underlying exemptions to copyright, so although it's blindingly obvious to a user that it's public domain, it isn't to a simple automatic check against a copyright database.

[storm pivot]

is prompt stacking legal?

[remote ivy]

is the custom model recommended if you want a sound and vocals for a concept album...any specific advantages that just a vocal persona and a group of instrumentals that fit the concept for album won't have?

[half dragon]

There's no prompt police -- you're fine 🙂

[polar spade]

Custom will customize all outputs to be somewhat similar though. For now it's hard to know how manipulative we can be with a "model" somewhat trained to sound a certain way.

So the jury is out as its only a month old, really. No one has used it extensively as well as compared it to extensively use of +Voice for voices others than our own voices which many are having fun eith that no electricity aspect.

Bottom line is everyone is doing their own testing ,I'd say. So best to get a consensus opinion of a few users than to just ask one or two.

The reason I pointed out the option was mere to say they exist. In the end, a +Voice persona may be consistent enough.

[abstract socket]

I'm hoping someone can help me. I keep trying to use the Editor to replace a section of a song, usually fix a few words. So I open it up, replace the lyrics. I get a couple alternate options (as shown). I play the previews, click Insert, then click Commit. Then I "Save as New Song", and the changes aren't reflected. I don't understand what I'm doing wrong.

[mellow condor]

Hi, I want to add a longer drop—kind of a hardstyle build-up—between the intro and the first chorus in a cover of an old Suno song I did on 5.5 as an remake. Unfortunately, Suno always skips right over that part and goes straight into the chorus. Why is that? Use the following after the intro:

[Instrumental Breakdown]

[Instrumental Distorted Drop]

[Instrumental Distorted Fill]

[Instrumental Distorted Breakdown]

[quiet solar]

What slider settings are you using?

[mellow condor]

I've tried a few things—turning up the audio influence, turning down the weirdness, turning up the style influence, or even just turning down the sliders—but nothing has helped so far

[quiet solar]

Hmm. Too high audio influence was my first thought as a likely culprit, but if you've already ruled that out...

[mellow condor]

suno tends to skip instrumental sections and sticks too closely to the original song instead of, for example, reworking a part. Since I want to build up a kind of hardstyle section here that isn't in the actual song, I go straight from the intro into the chorus.

Are there any good hardstyle/hardcore/rave prompts I could try out?

[quiet solar]

Next best thing I can suggest is to give Suno more to work with than just that tag - e.g. describe the instruments and/or vibe. It is more tricky to prompt for instrumental parts, especially if they're following an existing instrumental.

[mellow condor]

i will try thx

[barren cedar]

How do you get 1 voice for songs especially rap? I've tried putting solo vocals, solo voice, I usually like to add a little personality to my voice so I'll put "Solo Deep Raspy Voice or Vocals" and it still gives me multiple voices and switches tones for verse 2, pasrts of the chorus and for intro's, bridg'es and outros.. I've also tried putting lead vocals or primary.. even zero backup, zero layered, zero gang vocals and it keeps giving me multiple voices. It's not just rap though, I've noticed it in pretty much all of my songs. Even when I simplify the prompts it still happens. Any tips?

I've also tried putting no backup, no layered, no gang vocals

[hoary stream]

is there any way to make a cover of a real song ? like i'd like to cover the theme from an anyme, but when i try suno say the lyrics cointain copyrighted material

[hoary stream]

tf is this

[honest moon]

just type out the lyrics, but don't make it to a T, use and replace letters that make the same sounds, like for example if your song is
"i kissed a girl and i liked it, the taste of her cherry chapstick"

then write, "eye kissed a gurl, an eye liked it, the tayste of her cherrie chapztik" or something along those lines lol

[hoary stream]

oh ok let's try

[mint granite]

are there any dialectic tricks to prompting in lyrics? i use a prompt concoction technique using both brackets, parentheses, and symbolic notation, tempo manipulations mid words, and so on, but I also want to be able to tweak dialects.

so far I have had to toneshape with an eq3-7 in postproduction but that leaves me with only the south african and posh london dialects possible.

<@&1082792303976001547>

[polar spade]

When you say changes aren't reflected, are you saying there is no new saved file found in the Library or Workspace? Or that the file saved that you do fine (with different artwork and a current saved time noted) has no changes?

[strange crow]

Hi guys!
Im generating something using 4.5. It's incredible how much more creative it is... :S did you notice?

[honest moon]

y'all wanna hear the weirdest and craziest song i've made (honestly isn't THE weirdest and craziest one i've made but yeah lol)
You'll understand once you listen to it most of the way through, ha.
https://suno.com/s/3KflXVjAxXajefam

oh crap i'm in suno school i thought i was in general chat lmao mb

.> oopz lol

[polar spade]

Well, they aren't old models, they are alternative models. So its good to know what each can do by experimenting. Well done. Glad you got a nice result.

[strange crow]

Thank you for your answer. On the one hand, I think it's enriching to be able to use the other models from previous versions to experiment and search for something unique. On the other hand, though, it creates a bit of confusion for me and I'm not quite sure how to handle it. It seems to me that 5.5, despite improving the quality, heavily standardizes the type of songs given the same prompt, even with increased wilderness

[polar spade]

Suno is always working on bettering the "fidelity" with each model over the previous. So, yea, they are supposed to "sound" better than the previous, in some way.

But you're right, that sometimes, one just works better in one instant than another.

I still use models from 3.0 and up, for some aspects of making sound effects better, of ending a song faster, or having certain vocals I like in one over the other. Though the fidelity does seem to get better or the artifacts and hiss and noise does seem to be lessened as the model progress, each model still has its place.

Many Basic users still are frustrated that they can't use 3.5 anymore. Thats a sign that it has its benefits over 4.5all.

[rancid wing]

hello. i dont know if this is the right spot but i will ask here anyways if not then just let me know. like 1 year ago you could make some songs in your dms but that is not possible anymore?

[polar spade]

The Bark system has been retired [about Summer of 2024] So, we can no longer make songs directly in Discord.

[abstract socket]

A new file is saved without the changes.

[polar spade]

Yea. That's odd then. When we click committee, it will throw that replacement up into the segment slow as a different colour.

Then clicking Save should actually "commit" it into a new file with the changes.

Ill go try one and see if i can make it NOT save a new replaced file.

[bright parrot]

Can someone explain how to use new voices? It's look like I need to create new voice for every new song and sing exactly what should be in song/

And interesting thing voice created before last update can generate different melody. New one voice, created after update generate exactly same melody. What am I missing?

[rose crane]

Howdy everyone! I need to know if I'm either going crazy, my speakers are doing something weird, or if it's 5.5, but has anyone else been having issues with high frequencies or treble? I don't know how else to explain it; 5.5 gens just sound like there's a high freq somewhere in the background

[frozen dagger]

not crazy. it sounds like the old "shimmer", just hidden much better than it was before.

[rose crane]

Ahh okay thank you. I think I found a way to fix it up a little bit. Granted these are davinci settings, but I suppose you could apply them to any EQ

[lost kindle]

Got half way through editing a song in studio when it just randomly hits a point where the tempo doubles....all the settings look normal to me. ANy thoughts?

[polar spade]

Have you tried: Click near the beginning of the track in Track 1 position. Click the BPMs at bottom and choose Manual. Set your BPM.

[deft bronze]

Is there a way to add two personas to one song for a duet? Or am I going to need another program to put them together 😩

[frozen dagger]

There are always... ways... Reliable way tho? Yes, editing is the most reliable probably. Either Studio or another DAW.
Usually involves creating the song in one voice, then Covering it with second Voice, and then editing them together.
You could try to opt to switch the Voice and extend the song, but I have not tried this method as extend usually gives me more trouble than it is worth.

[deft bronze]

Awesome thanks!

[frozen dagger]

https://discord.com/channels/1069381916492562582/1466598465969590465
This goes over one method. Others shared in https://discord.com/channels/1069381916492562582/1252267444773326848, may be worth skimming through.
Methods for prompting shown in there too. (Trying to avoid using an editor)

[devout quarry]

Weird question , I wanted to make a instrumentalist track the thing I'm having a issue figuring tho , how to make suno make gibberish rap sounds that are like words if that makes sense. Something to help me find the rythem and stuff for a rap its self but no words deposited thx in advance.

[eager delta]

I know they changed Personas to Voices a while back. Can we not make Voices with a song the way we could with Personas?

Am I missing something?

[tacit axle]

You can probably generate gibberish syllables with an LLM and use them as the lyrics

[nova finch]

in the lyrics you can put [instrumental]
on a computer there is also a button

[polar spade]

I"ve always done decent with using [male vocal] and a verbal cue like "uh " or "mmm" or just any letter might work like "a" or "z".

I'm sure we could try other "voicing verbs" perhaps as well... chatter, gibberish, mumble, whispers... have fun to try others. I just tried [whisper] and it worked as well.

Prompt first with "male vocal" and add other prompts as needed..

My first one tonight was very explicit ( i wont post it) , so I changed it to a christmas rap.

Using 5.0 which does good , but can get lucky from 3.5 to 5.0

https://suno.com/s/q4iy6VB2Y75XP3XU

[small bridge]

so I prepaid for a year and I ran out of credits so I was just using my free 50 I had about 10 left and then it reset to 50 again just now. Thats great. I knew reset time was around this time for me as this is around the time I paid around 5months ago. Problem is I paid on the 25th and it has always reset to the 2500 on the 25th at this time. It reset to 50... What do I do?

[frozen dagger]

Voices made from uploads use the "Create new Voice" ui, but for making a Voice with a song that is made by Suno, right click the song and make the "Voice"/Persona directly from the song.

[polar spade]

Check your account to see if its been reverted to a Basic account . If so, send a ticket to have it reviewed, with your name, and account info in the email. No charge card number or such needed, but the receipt would help. Find that in your Billing history and y our email. At which point , also see if any change next month on May 25th

https://discord.com/channels/1069381916492562582/1479483412250366054

[halcyon sundial]

is there someway to upload an existing song just for cover it in another language?

i want to upload one and still flagging and blocking me

[quick wedge]

You can cover your own songs that you already made just never any existing ones.

[frozen dagger]

You could amend the prompt here, but the general idea works. (Not for free-tier tho) Advanced Options > Lyric Mode > Auto Then give it a prompt to give phonetic sounds only, non-lexical lyrics, or gibberish may work. I used:
Non-lexical vocal performance using purely phonetic syllables, no real words or recognizable language. Melodic “fake singing” driven by vowel shapes and consonant textures, fluid transitions between sounds, expressive pitch movement. Voice functions as a tonal instrument rather than language. Soft attacks, sustained vowels, subtle rhythmic phrasing, evolving syllable forms. Abstract vocalization, pre-linguistic expression, no structured lyrics.

this one is more vowel oriented but the idea is there. Until there is a toggle for mumble rap as a placeholder option. 😉

[eager delta]

Thank you!

[halcyon sundial]

i already did this before

i just forgot how

[quick wedge]

Here’s the article for it https://help.suno.com/en/articles/2872257

[frozen dagger]

o.O Did Mashup with some of these... 🧠🔥. More scatting than gibberish, or mumbling, but there is phonetic flow for sure.

[toxic gust]

How do you stop the lyrics being displayed like karaoke lyrics on screen? I liked it when they where all there and had to scroll to follow along. It's on my lyrics

[frozen dagger]

suno app?

[toxic gust]

Yes

[frozen dagger]

i don't know of a way to make it do one way or the other. i do know that it kept changing for me as well. i think they are still making changes and updating it like the website.
Might be worth mentioning in the suggestions for the mobile device you use? https://discord.com/channels/1069381916492562582/1258071022511329340 https://discord.com/channels/1069381916492562582/1314280094851928187

[toxic gust]

Ok thanks

[devout quarry]

just wanted to say thanks i was at work so i have been busy but thats exsactly what i needed.

[frozen dagger]

thanks for that. i actually listening to some of that again. it was a good idea. i hope it works for you. again, some tweaks to the idea could do some cool stuff. (Putting things in quotes "keeps hooks" in auto lyrics.) Muli-lingual and even just made up languages could work really well for this case as well.

[last summit]

I'm late, but personally I either use Lorem Ipsum (but sometimes that also imparts specific tonal qualities), or an alphabetical cipher, owo. Rot13 is probably fine.

[frozen dagger]

Glad you chimed in, better late than never, ChatGPT explaining those foreign words to me now. (I thought it was gonna be some plugin to turn voice into moosh or something. 😅) That helps with the "made up language" part I was suggesting quite a bit. Gonna include that as well now myself. Nice to have options for how to shape the sounds.

[last summit]

Is welcome OwO! The prompt that was suggested helped a ton too. -w-.

That said, here's one I did based on a cipher. It was used as a fictional conlang from Starfox Adventures, spoken by anthropomorphic dinosaurs. I like how Suno somehow got the accent weirdly correct. -w-. https://suno.com/song/5560ea8e-db4c-4286-9ba8-0e2344b7fa45 I'm not sure how Suno manages to pull inflection and accents out of lyrics like this, but somehow it's /really really really good at it/. I keep meaning to experiment more, but keep forgetting to.

[frozen dagger]

I could not be more hyped up to click a link right now.

[last summit]

Wait what? XD Have you played the game or something?

[frozen dagger]

I dunno about "Starfox Adventures" specifically as a title, but Starfox??? Yah. and the concept of the weird language from the game mixed here is brilliant.

I was playing a demo of it on the n64 in the mall (Sears) and missed the last bus leaving cus we stayed to play. ended up taking a 4 hour wlk home. worth it.

[last summit]

OwO! I cried when I first heard it go. This was practically the first go (okay, maybe the 4th, but I clicked the button like ten times in one shot. <.=.<) and I basically melted finally hearing a voice that wasn't one of the usual voice actors or my own, saying something new. I liked the game so much when I was a kid that I learned the language, but never had anyone else to speak it with. Suno just friggin' nailed it. o.===.o.

[last summit]

Also I am flipping out over this. o_o I never knew this was a feature, and now I'm running it on like a bunch of older instrumentals and am loving it! I love the "cybernetic intelligence making up words" thing and have been trying to figure out how to encourage this to happen more than one every 100 songs. @w@;

[frozen dagger]

Auto-Lyrics is a hidden gem I thought was not for me until I started toying with it. It is a prompt box. We should all consider using it. lol. It can do magical things. The Inspo/Mashup is nice, to kinda mix them up as well.

[rough depot]

its still "speed stacatto" singing the shorter lines. gah!!!

Doesnt make any sense

Why wouldnt it be the other way around???

[still gazelle]

Hi first time

Here

[devout quarry]

sha256? if you have the socks for it you know 😄

[last summit]

Nuh, not that random, owo. Still needs to follow some sort of linguistic structure. -w-.

[honest heath]

<@&1082792303976001547>

[unborn star]

I´m looking for a possibility to skip fade out process

Does anyone have experience with it?

[bright parrot]

Someone can explain how to use Style in 5.5? It's look like it work in totally different way as was in 5.

[tacit axle]

Dunno, I don’t use 5.5.

Doing a gritty folk project, and 5.5 interprets every prompt as “actually you want a polished pop singer song writer”

[bright parrot]

my problem 5.5 generate same melody again and again( I already cleared cash. tryed from another PC, same melody. Instrument, without any voice((

[sleek stump]

Thats happens with many genre in 5.5, it's why I am still using v5 mainly. Theres no technical thing or issue, it's just how the training and weighting is done. It needs changing for sure, and it likely will be, but for now it is what it is. Use v5, theres not "that" big of a difference overall.

[sleek stump]

Are you referring to the Styles (Prompt) window in Create, or the "My Styles" feature?

[bright parrot]

I'm referring to the Styles (Promt). Can't really understand how to work with this. And if I create "My Style" is it help?

[unborn star]

Probably not

I guess you making Lo-Fi Tracks? Do you have any song which have meta-Tags? Prompting is more precise in 5.5

[bright parrot]

I write style plus instruments. Like: A chillhop track that opens with a beautiful, emotional piano intro — spacious and soulful, setting a deep late-night mood. As the beat drops with warm, heavy bass, the oud joins in, subtly echoing and answering the piano lines. The blend creates a dark, immersive groove, hypnotic and introspective, perfect for deep focus and reflection.

[sweet ivy]

Hoping a mentor might have a suggestion. Tying to cover an original uptempo song, 142 BPM, that is 2:49. The song ends with a double chorus, with just 2 bars turn modulating up and really building. Suno is throwing in 4 bars, and losing that energy in the modulation. Maybe I need to change the lyric page. Right now I have [Chorus] followed by [Chorus}

[unborn star]

Should be working then

[plucky terrace]

Help, I need music keywords for Gregorian chant. I've tried several myself, but there's always rhythm or instrumental accompaniment. I'm about to lose my mind

[half dragon]

https://suno.com/explore?type=all&q=Gregorian+chant&sort=most_relevant
See if you can find what you're looking for.

[quiet solar]

Typically, Suno recognises that Choruses are usually repeatable and the same musically and lyrically. You can generally get away with lyrics tweaks - musical tweaks are a bit tougher.

You could try tagging the general iteration as [Chorus 1] and the modulated one at the end as either [Chorus 2] or, if there are words behind it, with [Outro] and relying on it to take the context from the repeated words that it should be based on the chorus.

Suno also isn't the world's biggest fan of key changes, so if you have one you should probably tag that where it happens. It's not something I've tested a lot, but I've not found a reliable way of getting Suno to perform a key change on demand as yet.

[sweet ivy]

@quiet solar. this is the first time I actually uploaded a full instrumental track from the studio session years ago.. I remember reading on here, that sometimes the simpler the track, the more freedom Suno has to do it's thing. To your suggestion in between the 2nd and final chorus, should I put in something like [quick turnaround, key change to A]

[quiet solar]

I'm honestly not sure - it's something that you'll have to test, both for the key change and any other steer you want to give as to how it should be performed. I've not really done enough with prompting for key changes to guide you from experience, I'm afraid.

[half hatch]

Can you use the exact same song that’s in your library to add new lyrics?

Or even extend your lyrics with the exact same song without a remix ?

[half dragon]

Extend is one form of Remix.
But you can definitely add new lyric to the existing Songs to create a new song.

[half hatch]

Thanks i’m really looking for no remix. I just want the song as is extended, so I’ll try that

[stable ridge]

my bad

[rough depot]

Sigh. It keeps changing from regular singing to multiple voices. https://suno.com/s/MZePXi54qZ9IV2ru

[rancid wing]

how do you get more credits on the website?

if you dont have the payed sumo

[quick wedge]

The 50 daily credits refresh each night

/refresh

[flint haven]

Refreshes at 0000 UTC+0 (<t:1712880000:t> local Daylight Saving Time)

[rancid wing]

ah okay. thank you

[quick wedge]

You’re welcome

[rough depot]

Its so annoying i JUST want ONE singer. i want it to stop changing between singers like this.

One. Singer. Why is this such a hard concept for the AI to understand I HATE when it switchjes

https://suno.com/s/JXqTWJLrUKiFw3zm

[quiet solar]

It might help to specify a solo singer in the style prompt, since Suno isn't psychic.

Ironically, getting a backing choir is often quite tricky - I don't think I've ever seen one turn up by accident before. I don't know Christian Rock well enough to know if they go in for choral harmonies, and whether that's influencing things.

[rough depot]

its been doing backup choir LITERALLY every song irrespective of style

[honest thorn]

Afternoon everyone. Has anyone had success on exporting stems and not having them get all out of phase? I have a great song, and I want to get the stems, but as soon as I mute the vocals, the entire mix sounds compressed/phased.

[quiet solar]

Might be a v5.5 oddity then - it's certainly got it's share of quirks. Though I can safely say that raging against the machine doesn't help, it only raises your blood pressure. xD

Generally speaking, if you want a very specific outcome it helps to tell Suno what you do want it to do in the chorus in the style prompt and/or the tags, and specifically NOT mentioning the thing or things that you don't. Those positive steers are far more influential than trying to exclusion prompt.

Often a couple of broad genre prompts as you're using should take songs in very different directions each time you create something new. Maybe try generating something in v5 - you can always cover it afterwards to 5.5 if you're looking to get the 5.5 sound

[rough depot]

yet somehow THIS SONG

Always COMES OUT BETTER LO;L

https://suno.com/s/LdAWhQq0MT1iQo2O

i cant get a better version of Ride on Red Lightning to save my life

But EVERY remix of "Master of Mind" comes out perfectly.

[quiet solar]

I have 'problem child' songs like that too. v5.5 finally did a really nice rework of a song that's been lurking in my archives since v3. Every other model absolutely minced it to pieces. xD

These things happen. Take the wins as you get them, don't get too stressed by the losses.

[rough depot]

its still speedrunning the shorter pre-chorus even though logic dictates a slowdown

YET!

I like this opening, I like this perfect tempo https://suno.com/s/sVQvYq99T9lgLGUZ

But I cant get it to replicate on a new cover

It starts out good then it speed-runs the pre-chorus like the micro machines guiy

[quiet solar]

It's worth bearing in mind the stems aren't actually stems, they're retrospectively broken out by AI from the completed song.

So one of the downsides of Suno being a lot better at mimicking production effects in more recent models is that they often show up in the stems when they're split out (which leads to that mismatch when you remove a key track like vocals), or they don't transfer across as they should at all, and the whole thing sounds tinny and off compared to the in-Suno version, and it's a lot more difficult to fix in a DAW than it used to be.

This is one of the situations where I've found Studio helpful, since it seems to do some kind of reprocessing pass on the stems when it saves to a new song, which often makes that issue a lot more manageable.

[quiet solar]

I think we've been pretty exhaustive in suggesting ways you can try to get around that in the past, so I wish you all the best with the fix. 🙂

[rough depot]

None of the suggestions have worked

I tried them.

The issue persists

It like insists on doubling the bpm on those lines

I even tried adding a new line to the pre

[Pre-Chorus]
The brave outlander and the fourth princess
Have fallen deep in love a bond they now confess
A fateful collision beneath the twin-mooned sky,
They ride with hearts aligned, flying ever high!

[quiet solar]

That's not entirely true now, is it? We fixed the tempo and pacing issues in the looting song absolutely fine by messing around with meter and syllable length to match the song's BPM and steer it pretty much dead on to where you wanted it to go. xD

But yeah, it really wants to interpret your pre-Chorus as a much faster pace than the Verse in the way it's currently structured.

[rough depot]

The looting song fixes worked, yes

It still dont explain why it worked fine in the original

The one i linked, with the Princess Carry cover.

[quiet solar]

Suno picked a different way to interpret it first time around, which seems to be different to the way it would handle it by default. The default wins out in subsequent approaches - I've had it happen too, and it's not always easy to strongarm it back to where you want it to go, even with an audio feed.

Let me test something for you, one sec.

[honest thorn]

So, are you telling me to use Studio to create the entire song first, but somehow tell suno to 'go easy' on the production? Will it do that?

[quiet solar]

No, not at all. If you have Studio, try splitting the stems in there and then muting the vocals and saving it as a new song, and see if that version is closer to what you're looking for.

[rough depot]

danke

This is slightly better but the instruments are subdued , what do i gotta do add "Driving guitars"? https://suno.com/s/hkTIph94HcPxM1OA

[quiet solar]

Incidentally, I really like the way it handles the opening line to the chorus in that song.

[rough depot]

The way it punches VANQUISHED and shoutes SHATTERED AND UNDONE?

It would also be nice if it actually gave me my "fulmen rubrum deflagrate fatum"

gah it needs to stop melismating

[quiet solar]

Yeah, just didn't really bother with the guitars at all in that version, huh?

I suspect that's an oddball generation. You don't usually need to go into more detail specifying electric guitars (unless you want to) in a genre where pretty much every single song has them. It might help to include prompts around mastering for the overall sound (e.g. clear mastering, wide stereo imaging, balanced EQ). Although I don't know if they'd be right for metal - I'm not so familiar with mastering metal tracks.

[rough depot]

now its going too slow though i made it 300 bpm,

https://suno.com/s/DA1rF3iR137Jf8GD

[honest thorn]

Sorry for all the questions. So, start a new project, pull in the 'completed' song, then how do you split the stems inside studio?

[quiet solar]

No worries at all. Pull the song in, then switch to 'Clip' on the right and you'll have the option to extract stems, or import previously extracted ones.

They'll come in as new tracks, with the original (usually) pre-muted. Just mute the vocal stem(s) and then choose Export -> Full Song.

[honest thorn]

Got it. Is there a difference betwen 'version' pull of the stems?

[rough depot]

@quiet solar you were experimenting?

[quiet solar]

Still am - it won't surprise you to hear it's being stubborn. xD

[quiet solar]

Yeah, Suno does the split with slightly different settings, so there'll be slight differences. It won't make a huge difference in most cases, but sometimes it can (e.g. if the backing singers get caught up with the lead singer in one version, etc.)

[rough depot]

more bpms usually means faster, right? O_o or am i crazy

[quiet solar]

Higher BPM means the whole track will be going faster, yes. Although if you have an audio reference (e.g. cover) Suno typically sticks pretty close to the BPM of the original track.

[rough depot]

I also want to preserve the way it holds "leads" for just a sec. what is that an extra quarter note?

yeah i dont get it t he og audio doesnt have all that chorale at all

[quiet solar]

@rough depot Yeah, sorry - no luck with the quick re-arrangements I tried.

I could get it to reinterpret the lines differently by moving things around, but not to replicate that slightly unusual off-bar phrasing it did first time around to make it fit, where it's starting the second line halfway through the final bar of the first line.

The only way I could get it to stick to that cadence was to stop it being a pre-Chorus and attach it to the verse, but then it stuck faithfully to the musical style of the verse rather than that slightly different bridging you had in the audio.

Have you tried downloading the song and then re-uploading it to Suno, to see how Suno itself interprets how the song is structured? And then use that version to cover from? It's a bit of a long shot, but sometimes things like that make a difference.

[rough depot]

i have yet to try that method

this whole "adding chorals" thing is getting really annoying

[quiet solar]

Given how stubborn a credit-sink this is being, it can't do any harm.

[quiet solar]

Try dropping Christian Rock and adding "solo male vocalist" to the style prompt, and see if that makes a difference.

[rough depot]

thats a different song

[quiet solar]

Oh, you're getting chorals in the other one too? That's...downright weird.

[rough depot]

Yes I am getting chorals in Ride on Red Lightning

https://suno.com/s/Noq5sM6nvjUHLs01

like i said, irrespective of style.

it shouldnt be "has been reduced (awkward pause) to 10)" it should be "has been reduuuuuuced to 10"

( but also not reduUUuuUUuuced) One note, sustained, not multiple notes across one sullable

[quiet solar]

Ah, okay - when you said 'chorals' I thought you literally meant a choir-style backing behind it. What you mean is the harmonies/backing singer/gang vocals bit.

That's likely coming across because it's common in the genre, and you've got a slow-ish chorus that's well-suited for an anthemic "get the whole band singing on this" line. Again,** emphasise "solo singer" in the style prompt** is the best advice I can give on that - don't rely on trying to exclude things, its nowhere near as effective.

[rough depot]

Either that or its an echo

[quiet solar]

It sounds very, very deliberate. I really like it - appreciate our tastes are different. xD

[rough depot]

Ifi could JUST change the lyrics on the original I'd be happy with it as is.

[quiet solar]

Is it a simple lyrics change to the original?

[rough depot]

going from "The trickster of the night" to "illusions bending light" \

[quiet solar]

Literally just that?

[rough depot]

last time i checked yeah

[quiet solar]

Alright - if you DM me the .wav or .mp3 of the original song, I can have a go at just switching that half-line out in Studio and leaving everything else alone, then send you the edited version back.

That mini vocal edit will be a thousand times easier than fighting the pre-Chorus.

[rough depot]

solo singer didnt work either. christ.

This is my current project. need to figure out how many syls i really need https://suno.com/s/Qy3rvxJRJFZxfcT7

The only non negotiable thing is i gotta have that yoda reference in the chorus. "Adventure, excitement"

[tacit axle]

anyone have any tips taking a song (suno generated) and replacing all the words, but keeping everything else as similar as possible? Same singer, same rough melodies (new lyrics roughly match the structure of the old, syllable to syllable), same accompaniment?

[rough depot]

like i dont mind it holding a note for multiple bars, but i can not abide using multiple notes for one syllable

[tacit axle]

i'm think i may have to extract stems into my own DAW, then generate covers with the new lyrics until one delivers the vocals similarly enough to put over the original non-vocal stems. But if anyone has a better option, i'd love to try it out!

[sleek stump]

There is a way, you can do a simple stem split, and then select add lyrics to teh instrumental stem.

[quiet solar]

That would have been my go-to suggestion, to be honest. Cover the song with a reasonably high audio influence, with the vocal tweaks in place. If they shadow the structure of the original, it should keep them mostly to the same melody and instrumentation behind it.

It'll be a bit more work if you're looking to keep the existing instrumentation exactly as-is, as that'll require work in Studio or a DAW.

[tacit axle]

won't the lyrics be a totally unrelated melody for this workflow? The melody is a banger

[quiet solar]

That would have been my thought too. 😉

I've not really tried that route though, so I'll defer to Ginger's experience

[sleek stump]

You said that "(new lyrics roughly match the structure of the old, syllable to syllable)" which should do exactly what you want. I have used this method to add lyrics to instrumentals I created first, and as long as the cadence matches the instrumental, Suno will usually just add the vocals on top. @tacit axle

Here is a perfect example of vocals added to an instrumental after the fact. https://suno.com/s/34xs2WNFuEBJSmnr

This is where the option is on an instrumental...

[quiet solar]

Out of curiosity, I just tested that approach on a song I'd converted from vocal to instrumental, to see what would happen re-introducing the original lyrics to the same structure.

The re-added lyrics picked up the melody pretty well, but in this small test at least, the vocals deviated quite a bit more from the cadence and melody of the original compared to a straight cover. Instrument deviation was a LOT less than a typical cover though, which makes it a very useful tool to be aware of.

@tacit axle So I guess it depends which is more important to you:

Cover - more likely to stay closer to vocal melody at cost of wider instrument deviation
Add Vocals - more likely to stay closer to instrumentation, but with more deviation to vocal melody

[sleek stump]

My process is to take the original instrumental, do add vocals, and then split the version I feel is best and use that vocal and backup vocal on the original instrumental... magic!

This also gives me control to edit out unwanted ad-libs

[quiet solar]

Haha. I basically mirror the same process using Cover - do a few variations on the original, then combine the best version of instruments and vocals together (along with any other tweaks).

[sleek stump]

I use Cover for vocal and lyric edits but I use add vocal for original instrumentals that I write lyrics to.

They work because I am writing to the melody and cadence.

[quiet solar]

Yeah, that makes a lot of sense. I haven't really tried adding vocals to instrumentals before, but I've been pretty impressed with how it handled it just now.

[sleek stump]

There are some really awesome tools in teh Suno suite, just have to figure out how to use them 😄

[bright parrot]

When you add vocal you can't use Voice (((

[tacit axle]

I appreciate all the tips, even if some did not honor the whole “exact adherence to the original melody” part of the goal.

I did end up finding a workflow to do it, but it was a lot:

1. I used a custom model I had built a while back and abandoned. I created it to capture the singer and production style I’m going for in this project. I had mostly given up on using it because the instrumental accompaniment was too polished for the project, but here im focused on getting new VOCAL stems to put over the original song.
2. I created and used a new voice from the original song
3. I did a cover with those settings, and set “audio influence” to 100

The result was basically like telling an actual singer “hey sing it again, exactly the same, but with these new words”

[slender pagoda]

Hi from Italy

[velvet steeple]

Hiya Elwing

[tawny beacon]

Do you have access to studio?

you could technically create a bass song there by speaking or humming the tune and creating a stem from it

[pure panther]

You should be able to do that in Studio. It will sometimes add other instruments and vocals like a madlad, but you should be able to get raw bass stems too.

[pure panther]

Yes!!! This is an AMAZING change! Now when we upload audio, we can manually input the style section! Woop woop!

It doesn't seem to be actually working though haha. But it's at least a WIP! Good change.

[paper hearth]

Yeah the more they give more customization options the deeper the songs will get and more personal this is a great feature addition in my opinion

[polar spade]

Do we still get the original auto-description? I also took that, kept the correct bits and changed anything pertinent. Seemed a shortcut.

But also, it gives an insight to what Suno thinks it's hearing, rather than my telling it what I think it should be hearing.

[pure panther]

It didn't seem to save my changes at all.

But I see where they're going with this and I like it

I'm still so confused as to why people find Studio so useful... It just never blends what it makes with my existing songs, I don't see how that can be useful. I'm going to hop into the Studio chat and add screenshots and a question.

[neat lake]

Hii guys, do you have tips to force Suno models to be creative with the melodies, expression, different types of vocals, etc? In v5.5 especially, many of the generations sound very the same in terms of the voices and many genres have specific instrumental patterns which are consistent across generations. I would like every generation to surprise me, or at least stick to the genre I want, but to be creative and interesting, not just generic. I want to achieve this, without going too crazy on the weirdness, so it doesn't break my lyrics pronounciation 🙂

[snow flare]

no, but im going to wait and see what people comment cuz I want to know too🤣😂

[pure panther]

I literally put "creative instrumental solos" in the prompt and it works quite often.@neat lake@snow flare

[neat lake]

Rofl this made me realise, I never put creative vocals in the styles box, lemme try that 😄

[snow flare]

oooh im def trying that right now,Thank you!😎🙏🏻

[rancid wing]

btw if you dont use the credits will it stack if you dont use it?

[pure panther]

no

not the 50 daily ones

[quick wedge]

/refresh

[flint haven]

Refreshes at 0000 UTC+0 (<t:1712880000:t> local Daylight Saving Time)

[quick wedge]

They do refresh each night though

[frozen dagger]

May be a dead end, but could check the ... and look for "Earn Credits". I'm curious at this point if anyone has this on their UI anymore. (Mine is there but never refreshes with new tasks.)

[unreal sleet]

leave it blank so the ai dont confuse your prompt, from my testing at least

[polar spade]

No. Credits are an allowance for that period. So Basic tops off daily to 50 credits. PRO TO 2500. AND premium to 10,000.

They do not accumulate.

.

[polar spade]

I don't know though, can Basic accounts earn usable extra credits? Without a paid account to activate them. I thought ALL extra credits are "inactive" until we have a paid subscription.

[unborn star]

@polar spade Hi I got a question for you

Have you ever tried 100% weirdness in Suno 5.5?

song seems to be stable now, but I´m not sure which influence weirdness has now

any theory/idea?

it feels like weirdness is off, but not sure 🤔

[gritty mulch]

having a weird little issue. i am prompting the BPM for a song in both lyrics and style cause this song is giving me issues. When I run it in 5.0 it makes it 70 BPM but when I run it in 5.5 it makes it 155 BPM, so doubling the BPM while sounding identical more or less but it prevents me from mixing and matching stems in studio which is what i like to do cause i think each does different things better. anyone know of a solution to either force the BPM or sync these up in studio? i can take this to studio chat if it'd be better there

[half dragon]

I believe you can't reliably prompt for BPM in any model.

[gritty mulch]

Hm. Usually when I prompt for BPM it's at least close. Doubling in one model vs the other is just a weird quirk to me. Thanks tho, i'll find a workaround as per usual

There it goes, adding "downtempo ballad" to 5.5 made it actually run in the 70s as requested.

[polar spade]

100% Weirdness usually made things really weird. I haven't tried in a year. If its doing fine, then yea, that could even be thought as a bug. But 5.5, at least in Studio, says its beta, so that gives the devs room to keep working a while to find a good balance on the sliders to create the sweet spots.

[polar spade]

Yea. SUNO knows a concept of a million songs and moreso specific genres that are often downtime ballads amd slow.... than just 70 BPM, which is not a "criteria" for any genre.

[split jewel]

Helloo, forgive me for my English. I'm desperately trying to create a model of my voice with "voice" on suno v5.5, but the result is mediocre. I'm not recognized much, and the AI voice sings completely wrong

[frozen dagger]

Culd try this Cutom Model trick to reinforce the audio and try to get the Voice to sound closer to what you want.
https://youtu.be/funB0pzc-hA?si=BgfArf5i5iB4z4wp

[frozen dagger]

The style you sing in the Voice and the style you use for prompt matter too. If the styles are far apart then the ai has to do some work trying to interpolate the styles. So reccommend making a Voice singing in the style you are creating the song for. (If not already)

[shy hollow]

Hey guys! Does anyone know if it's possible in any way to have 2 personas in the same song and deciding when which of then sings? Trying to solve it with a mashup right now, but it's not ideal.

[rough depot]

I am attempting to make an overture

Hmm but i dont think "inspo from the album" is the way to go. Basically I'd like there to be samples from all the songs including the final confrontation song.

Actually this seems to work, although I wish it would sample "Master of Mind" closer to the end: https://suno.com/s/A8b2uranlWdavXjt

Wait can i specify WHEN to sample specific songs at specific timeframes??? is that a thing?

[pure panther]

I would make the first one in whatever version you prefer with one persona, then cover it using v5.5 (it sticks to the original much better than any other version imo) using the other persona. Then I would splice them as desired in Audacity (doesn't have to be perfect, a rough mix is ok) Then upload it back into Suno and cover again using v5.5 until you get the version you want. If there's a high pitch shimmer in what you settle with, remastering on subtle with v5 can fix it.

[quiet solar]

Not that I'm aware of, no.

[delicate karma]

I decided to test my custom tool's potential by recreating a million-view hit from scratch. This is the result of the AI transformation process. What do you think?

[worthy kraken]

Why could the insert button not be available? Bug?

[shy hollow]

Thank you! I'll give it a try.

[polar spade]

Could be glitchy, sure. Yours show an information button on the most recent ones and it looks like you have run several replacements. What do the info buttons say, and maybe after so many it could get glitchy?

They will always [or should] be in your Library and you can create the full song from there if needed.

I ran four, and they were consistent.

[worthy kraken]

They are not in the library. I managed to open one in another window and generate a full song fron it, it only generated last 40 seconds. And when i reloaded page those disappeared from the library. Seems like a bug.

And the info button said nothing, coursor turned into unavailable when hovered over it

[polar spade]

Sorry. Go to workspace and click off "Hide clips from edits" in your filter to show them. It appears the Library may or may not show them presently.

If you like one, click Get Song to complete.

Report as a 🐛 bug. Maybe it has to do with so many replacements at once.

[polar spade]

The slider bar looks pretty thick as well. Any change of you adjust the zoom of the display and zoom out a bit ?

[worthy kraken]

Nope, zoom never chaged anything. It's just gone. Also, it happens quite often. Sometimes it just eats credits but generates nothing. Sometimes it is fixed by reloading the page, then it starts generating. But just now, for example, it just ate my credits and gave no output at all.

[polar spade]

All available in workspace, though?

[worthy kraken]

Yep, actually they are there, just hidden.

[polar spade]

Its like restarting our computers, the 'ole log out, clear cache, close all browsers, open one browser and log back in might help of its a display thing, as i ran about 20 at once and everything works as expected on Android phone.

[worthy kraken]

But when i press get full song, those that did not appear on the timeline only generate 47 last seconds instead of full song

[polar spade]

Yes. As noted. We have to click off of the "Hide clips....." option in filters.

Glad you have an option of the workspace to work with while , if you would please, note the glitch in the bug-web channel if using a desktop. The more to report a similar things the more it will point to Suno and not an individual issues.

Thanks.

[worthy kraken]

Yes, thank you, i will report it later. And the bug as well.

[worthy kraken]

Yes, it seems like a bug. For me it generates, but not full song with those exact clips. I will report it.

Interesting that they always generate 47 seconds as full song

[polar spade]

Thats all? And the original was at least, like, two minutes? Very odd.

[worthy kraken]

4 minutes)

[hoary stream]

is it me or suno is making me the same song again and again? no matter what i change in the style section... once it has decided my song should be one way, it will just keep doing it no matter what

i wrote a song i asked for flute in the prompt

then changed my mind removed flute from the prompt

but suno keep adding the flute... generattion after generation, and anyway the tune of the song stay the same na matter how i change the prompt

do i have to make a new workplace to make it forget my previous generations or what?

nope

when into a new workspace, remove my prompt totaly and wrote just "k-pop" wich add nothing to do with my previous promp

and bingo i got the same tune and the flute is still there

and the tune it's just like the previous song i created... it's...strange

i changed for "blues" just wrote blues and it's giving me same kind of tune... witht the freakin flute...

what am i dealing with here?

i'm loosing my mind with the flute!!

[worthy kraken]

Do you have persona attached maybe?

[spice mortar]

Flo, same.

[polar spade]

I always feels there is some buffer latency of about a minute, even if we change things. So I have a habit of saving prompts and lyrics, not my settings , so I will then delete or wipe the Create panel completely.

Manually put everything in and start over fresh.

Thats a good way to see if latency is an issue.

[polar spade]

Im assuming this is just lyrics and prompts, with no custom models or as @worthy kraken asked, no Voice/Persona or any other outputs attached and sliders at default.

[neat lake]

Yes, very relatable especially with the newest models

[signal shore]

any recommendations on how to stop the voices turning into a damn powerhouse like christina aguilera running riffs all over the place? Especially in later choruses, I still want the vocals to be clear and concise, but the main vocal starts screaming melodies over the top 😩

I've tried replacing the sections in the song editor, but it just replaces them with a new variations of the hootin & hollerin top lines

[pure panther]

Maybe adding intimate vocals to the prompt and verse headers would help?