📚┃suno-school

[teal raft]

We're building them a 'map', if you'd like

[elder dune]

Gotcha

[teal raft]

It's hyper important though

But like, for later

[elder dune]

I'll be sure and be more diligent about it, then, definitely

Anyway, thanks everyone for the help tonight. You've alleviated a great frustration for me, and now it's time for bed.

[teal raft]

Have a good night

[unique stone]

We were asked to do this for v3 alpha before as well, and since it resulted in what we have as v3 I think it's definitely worth putting the effort into highlighting what's bad 😄

[elder dune]

This is the version I'm gonna go with. A little editing at the end when it seems to go off the rails a bit, a little ambience, and it'll be just about perfect ..

https://suno.com/song/4e3b7d98-2a8d-4a1c-99cb-a13b294d3748

[craggy delta]

i see that v 3.5 even insturmental songs now have vocals, even within put lyrics. Weird bug

[wet minnow]

v3 did this too. hallucinations

[craggy delta]

yeah, still, its quite amazing for 80s rock and now PERFECT for 90s eurodance!

This is getting better and better, i cant' wait to see what Suno can do in 1 year from now

[unique stone]

I quite like using hallucinations as I usually don't want lyrics but I like vocals. Weirdly v3.5 doesn't want to continue my vocal hallucinations and instead throws a piano in there and/or tries to end the song 😄

[wet minnow]

i had some songs where 3.5 really really wants to end each part after like 20 seconds

[patent horizon]

so if i re-generate the full song in version 3.5, or 4 ... will it improve the quality

[wet minnow]

you cant regenerate. if you use the same prompts it will be a different song. and if you extend a 3.0 song the quality is most likely copied from the first part. for me it sounds that way.

[patent horizon]

what does "get full song do" i guess that's how yo would do it continue from the end, and remake the outro.. then get full song

[wet minnow]

it stitches together the parts you made into one song

[patent horizon]

I wish it had an option to like take all the parts you made, and then " remake them " sort of like a re-master, where it would improve it using the new engine

what about download the wav format do you notice a quality increase

i couldn't tell much of a difference between me downloading the wav file and comparing it to what i released on spotify already

[untold briar]

I wonder if you could regenerate hypothetically. Analogously like rerun a bark small output through just the later models, but it would be whatever hierarchical token thingy exists in Suno. It is 50/50 in Bark at giving the exact same voice when you do this.

[static swan]

so are we getting some vocal consistency now?

[vestal rapids]

does anyone have some tips at how to get v3.5 to generate a long track, it's nice that tracks can go up to 4 minutes but I am struggling even getting near that

[patent horizon]

write song stucture in your lyrics , intro, chorus, bridge, chrous, verse 1, verse 2, outro

[Verse 1] bunc hof stuff to sing, [Chorus] bunch o chorus stuff, [Verse 2] vbunch of verses [ Chorus] same chrous as before, [Instrumental] [Chorus] bunch of chorus stuff, [Outro]

[vestal rapids]

yeah, I have been doing that, the tracks still seem to only get to 3 minutes max

[patent horizon]

that;s good because prior id only get 1-2 minutes

[signal trout]

Really having a hard time getting it to not jump right in to lyrcis. And when it starts with lyrics, it always seems to start with hard to understand words, then become clearer, makes it real hard to make anything, when the first line keeps coming out garbled.

[hollow hare]

The best method to have an instrumental introduction is to make an instrumental track (using various tags instead of the switch so you can pad to 2 min and find the region you wang) then Extend it to have your lyrics.

I too struggle to make it do the intro. Either because I had too much lyrics present or the AI really did not felt like doing it for the genre I selected

[signal trout]

OH, that's a good idea. I might try that. I only just decided to subscribe. I was tired of only using the free points. So now I can actually sink time in to making the songs.

[hollow hare]

The best thing about this is that the instrumental tracks tends to be bangers. So even if you did not get the style you wanted, you will likely get tracks you will want to use for your future songs (or portions to clip in with DAWs)

[undone sleet]

does anyonne know how to make duets? I've tried labeling the style of the song as a duet and even labeled all of the lyrics on who is meant to sing them but the best that i can get is just two people singing together instead of them taking turns

[eager topaz]

I've tried so many different prompts in Instrumental mode and found very little that is good. I've reported so many clips, it's just a big time waster for me. Also, the 4 minutes is misleading as hardly any have been lengthy.

[hollow hare]

Getting duets... that is more luck than anything else. Try to use male and female sounding names in tags, that might help, but it is still hard to achieve.

A few people were able to do it but at an expense to their credits:
3Daizy's duet
MrTree's duet
Suno.Wiki entry

Again, getting it working on purpose is hard.

Alternative methods:

1. It is easier to make a duet by doing something like generating instrumental part 1 then Extend it twice (one extension where you specify for male vocals in the styles field, the other for females), or something similar for each alternating part, then edit them together in a DAW.

2. Even easier (and cheaper) is to make a full track then swap the voices gender using other AI tools like kits.ai (see #🗣️┃chill-chat message ) or your own voice with a RVC model (see #💡┃suggestions-web message ).

You can use a DAW to edit which voice track plays when.

[signal trout]

As weird as it sounds. Any time I use 80's metal as a tag, it seems to skip the first word, or line 90% of the time. Dropping it almost always has it work better.

[graceful condor]

Heyo! Does anyone know if it's possible to change the photo that appears on the link-form of the song? I love that we can now change the "Cover Photo", but is there a way to delete the other one entirely?

[signal trout]

I take it back, it just, drops the first line nearly every time even without it.

[alpine venture]

https://suno.com/song/11f2f666-82d0-4423-8b6e-f08077be5302

also https://suno.com/song/6c5e0803-fd92-4a2d-9592-7fa2df1f6423

a few prompts I have reliably hallucinated full rap songs, where the only distinct words are ones I cannot repeat here

not existing songs mind you, these are new, they're just in simlish

[graceful condor]

Do you know when this symbol will leave I've been here for about a week since the start of Monday

[wet minnow]

i think 7 days

[graceful condor]

Ok thank ☺️

[untold briar]

If you find yourself wishing you had more than 2500 characters, extend a song from 00:01 and generate two extensions each with 2500 characters in the same time frame.

[chilly leaf]

Also seem to end songs waaaay early

[graceful condor]

How do I check if I have the newest version of suno

[graceful condor]

oh

[graceful condor]

Thanks Mr. tree 🌲

@wet minnow How did you get early access? * looking in oral after image
Mr.Tree 🌳(@MrTree) sent*

[wet minnow]

i have the paid version of suno

[graceful condor]

Thank you Mr. tree

@wet minnow This you?

[wet minnow]

awww. thats super cool. thanks a lot 😄

[serene viper]

Is there a way that, by default, we can use v3.5? I sometimes forget and end up working with v3. LOl

Just curious

[untold briar]

I used a 3.0 prompt in 3.5 and I'm getting something more generic than the specific genre I asked for. What can I do?

1. Replace any spaces with dashes if you used any compound words. 'Acid House' -> 'Acid-House'
2. Consider reducing the number of very broad genre terms in your prompt. Repeating words is great, I'm specifically referring to repeating broad terms like country or metal. For example, if you're looking for a specific genre of 'metal', don't overuse the word 'metal' too many times in the style prompt. (I'm less certain about this one, let me know if it seems like it doesn't help.)
3. Use a longer style prompt in 3.5 than the style you used in 3.0. Adding unique style words is best, but simply repeating important or weaker style words also helps.
4. Also, add some of your style words in the lyrics. In the simplest case, just copy a few style words into [brackets] in front of your first verse. Better is to integrate styles and instruments you want throughout the song - but just doing that can be enough.
5. (Power Prompters): If you are using very unusual style prompts (not musical words, not real words at all) then you may be the exception and your 3.0 prompt may actually just mean something different in 3.5

[undone star]

using a word multiple times seems to have an effect. i did that in 3.0 for male voices - using ", male male male vocalist" which dramatically increased the male vocals. 😄 - same for genres.

[untold briar]

Oh I agree. I think I might be the first person to shout about the power of spamming back last summer. To clarify, in this case, I mean if you are prompting for like 'nu metal' or something, if you happen to use the regular word 'metal' 3 or 4 times in the prompt, it might push Suno to be more 'generic metal' and not what you want. So in that specific case, try to avoid amplying the power of metal (generic metal, not any specific genre) too much by not repeating it too manyh times.

I am 100% certain repeating words increasing the power of the words in the prompt. Very much like increasing the strength in Stable Diffusion. The part I am not certain about is, if you repeat 'metal' many times in the prompt, does it make it harder to get specifically 'nu metal' or some other more specific sub genre. I tried it a couple times it did seem to help, but could be rng.

[undone star]

when i build a prompt i often feel like i can lose the emphasize on let's say metal if i use a lot of descriptors that can be applied to other genres. So i might use "thrash-metal, aggressive dirty fat melodic 1980s melancholic thrash-metal, deep deep male male vocalist" (just an example - ignore otherwise :D) and that for sure puts a higher emphasize on the thrash-metal part. so using these compound words for genre (like bass-house is what i use i.e. or for -pop). I'm pretty certain, given it works in SD or Midjourney as well, creating fantasy compound words being able to lead to new genres as well. haven't tested that yet though. 🤔

[grizzled slate]

3.5 is Magic! Can I set this as my default?

[runic dock]

What exactly is supposed being improved in 3 5 in this announcement :

• Improved song structure and vocal flow

Who can give examples?

[untold briar]

I'm not sure exactly, but I'll take a guess:

Improved song structure: If a song can be 4 minutes long in 3.5, then a song could have better 'structure' because the beginning, middle, and end of a song can fit in 4 minutes. Before, with 2 minutes, when AI was composing the end of the song, it did not remember how the song started, because the song was longer than 2 minutes. Now it can keep the whole thing in memory, so it would make sense the music composition/structure would make more sense.

vocal flow: Not sure, but 3.5 vocals aren't so heavy in choirs and reverb. I wouldn't call that flow but I like it.

Is the search biased to showing you your own clips? I feel like I can't possibly be personally generating 80% of the songs with 'fm' in the style : https://suno.com/search?q=fm but I guess it's because I used the public flag just to remember clips and not intentionally to make them public, so now I have like a billion random mediocre clips that show up in search. 😬

[halcyon brook]

Even if you set it to v3.5, it will only create a 4-minute song about 10% of the time. How is everyone doing?

[runic dock]

How do we get .wav? I still see mp3

[undone star]

If the content of your Lyrics window isn't enough to fill a reasonable 4 minute song it will not creat a 4 minute song. It also depends on the tempo of the song. You should regard the 4 minutes as UP TO 4 minutes not 4 minutes each and every time.

[runic dock]

Make sens even it is a guess.

[undone star]

So far my experience with 3.5 is... a wash. I like the longer generation times, but somehow it still lost vocal cohesion in later parts regarding choruses or outros.

Also the same prompts now generate quite different results - needing to relearn and restructure your song prompts. Which is fine, but tideous.

Emphasize on certain keywords seems higher now - like. i.e. "swampy" in 3.0 which was always had a heavy impact no matter if its one of 20 keywords 😄

[halcyon brook]

Okay, thank you. How much lyrical content do you fill in? I have about 18 lines. So the song always ends in about two minutes.

[undone star]

4 minutes depending on the genre and tempo are like

Verse
Verse
Chorus
Verse
Bridge
Chorus
Bridge
Chorus
Outro

[unique stone]

It's on the "Download" menu

[halcyon brook]

Thank you, I'll try it! In my case, the BGM doesn't have lyrics, so I'll try to devise a little.

[undone star]

you can vary the style by adding interludes, risers, instruments etc to fill the void.

[runic dock]

I dont see it. So maybe it is an issue on mobile only or it is a bug on my side. Will try later on desktop.

[halcyon brook]

Will it work if I put a prompt in this gap?

[broken lotus]

3.5 is not trained on 4min songs....
it can make longer songs, but it wasn't trained to.

better 4min songs will probably be for v4, but they made 3.5 compatable.

[undone star]

it should - but you need to vary the input. i'm not sure its gonna read the same tag 3 times individually. So you can insert

[Intro Ambient Piano]

[Piano Riser]

[Ambient Piano Interlude]

[break]

[Melodic Piano Interlude]

[Beat Drop]

[Aggressive Piano Solo]

in that style.

[halcyon brook]

I see! I'll just have to wait for v4))

[halcyon brook]

Thank you for being so kind! I'll try this once)))

[broken lotus]

2 min Extend is working great!
the context memory is also longer (probably double?)

[undone star]

good luck with your prompts. and general rule of thumb: instead of breaking your head trying to make it fit into one prompt - just extend 😄

(JESUS WETCIRCUIT!)

[halcyon brook]

I think so, too. However, I think I will run out of pro credits😹

[halcyon brook]

Great, your prompts have expanded the range of my songs!

[undone star]

https://www.suno.wiki/ - expand your knowledge and after that just create, create, create and experiment. you gonna get better and develop your own style after a while.

[broken lotus]

Who has a comparison of wav vs mp3 waveforms...?
Is Bigger Better?

(Hoping it's a placebo. My hdd is already full, tbh)

[halcyon brook]

🤝

[unique stone]

Someone posted a spectral view of it earlier, but having plotted the spectrum of part of one of my recent (v3) songs it's got frequencies up to around 20KHz so there's definitely more detail in there. I couldn't really tell the difference with some brief testing though.

That said, I'm using an old (10+ years) Steelseries gaming headset so I should really listen with my better headphones on my music PC 😄

[young orchid]

With 3.5 it seems really hard to make music. It won't listen most of my prompts. But at least the sound quality is much cleaner.

[untold briar]

a few things to try #📚┃suno-school message

[unique stone]

Regarding this part of the recent announcement: "Please give us feedback by upvoting 👍 and downvoting 👎 to help us improve your experience. We especially appreciate downvotes on clips with bad audio quality. If it hurts your ears, let us know with a thumbs down!"

I'm assuming this is going to be guiding v3.5 training for the final release. With that in mind, I'd like to know: Are the team only looking at data for upvotes/downvotes for v3.5 generations or will they also be looking at upvotes/downvotes for v3?

[young orchid]

Thanks i will try

[broken lotus]

I think this data is most useful to Tony who seems to be the main chef cooking the models -- at least he's the one that asks questions and specifically wants the thumbsdown on bad generations.

He doesn't say what is different about the models (and everyone has an opinion and is using it differently, so subjective feedback isn't very helpful)...

I think he just wants an easy binary 👍 👎 to 'prove' some of the decisions he's making on the models. In the past he has said the downvotes are more meaningful (to him) – and specifically on 'errors'.

he has asked for feedback on v3, but almost no one uses downvotes in an ongoing way so not sure how much data can be discovered at this point.

I'd treat this v3.5 as another closed beta – smaller group of users who are already familiar with Suno) and assume the knowledge of what works will influence the creation of v4 (...roughly 4-5 weeks away...)

so Downvote, and Downvote often!

[halcyon brook]

I feel like the piano's high notes are no longer harsh!v3.5🫶

[unique stone]

Suno users: "I'm cooking"
Tony: "YOU'RE cooking? Hold my beer..." 😄

Alright. I'm going to solely focus on downvoting only poor audio quality with v3.5 (and tbh probably going to only use this version exclusively whilst I don't have any specific musical ideas I'm trying to throw credits at.)

I have other problems with it at the moment (always short instrumentals compared to v3) but I can at least get around that with extensions whereas audio degradation is a one-way trip.

[broken totem]

does anyone know how to generate drum loops and keep suno from mixing in instruments=

been trying this for a while, also with 3.5 but never really works 😦

[proven ivy]

is it possible to search through my own non-public songs? it's kinda hard navigating through 100+ pages through the library

[broken lotus]

for now only public songs are searchable.
Devs said the ability to search our own unlisted songs is coming (it sounded soon)

[runic dock]

I confirm that the menu with wav is not seen for mobile . This is a bug. It has not been announced as a limitation.

[broken lotus]

confirmed... just squeezing my browser window narrow shows the mobile page, and it does not have .wav...

[ocean venture]

Does the quality of Song generations improve when the song is fully loaded? So am i hearing a lower quality preview?

[broken lotus]

yes, I think this is correct.
the preview while it's still spinning is lower quality.

[ocean venture]

Then i need to make a suggestion in #💡┃suggestions-web

[vestal rapids]

does it matter if we delete a song after giving it a 👎 ?

[unkempt viper]

Is there any way to remake a song you previously generated because I used a very vague prompt but it made a really good song so im wondering if theres a way to make variations cause any time i use the same prompt it sounds nothing like it bc it doesnt really go into detail.

[broken lotus]

I think the trash is not a metric the devs look at (they never said)

[untold briar]

The closest you can do for now is Extend From some time in the song which has enough audio to keep the same style, but not so much you just get the same exact melody. If you can find the clips that you used, you could try using times relatively close to the starting times of the clips. That way the clip will have the sound of the song (mid song) but it won't have the memory of all the past verse and melodies, so it can diverge and go in a different direction, like a variation.

Also check if you need to tweak the style prompt #📚┃suno-school message but if you used a vague prompt, that may not matter (and the audio itself will be the biggest influence on the style)

[broken lotus]

does it have any music at the start, or does it go straight into lyrics?
you might be able to extend from the 1st seconds or an intro to get the same sound.

[unkempt viper]

Its using the instrumental feature, there are no lyrics but yeah thats a good idea, thanks!

[untold briar]

If it's an instrumental just the beginning of the clip probably works, that's more doable

[crimson terrace]

do we know if 3.5 is different in terms of vocals with another language ?

[broken lotus]

that's such a broad statement...
are YOU hearing something different?

[crimson terrace]

oh i don't have a pro plan so i can't test it, just wondering if the devs did anything in terms of multi language vocals as i mainly use suno with Arabic and would love to see it improve and become actually useable like English and French

[young orchid]

I compared the sounds and the results are as follows.
Waveform and spectrogram,(Left image) the upper one is in wav and the below one in mp3
Frequency and Phase analysis table(Right image) is as follows Left one waw, Right one mp3
TLDR:
Mp3 does not get the fine details due to its format, and it loses its details completely after 18khz. If the listener is over 35 years old or does not have good headphones, it is difficult to understand this differences after the 18khz.
Mp3 uses a higher bit width but it doesn't matter much. I expected them to use 16 bit but 32bit float was used.
Mp3 has a higher sound level,
The mp3 has a different soundstage than the original.

[broken lotus]

Ahhh....
give me some sample lyrics... I can try?

(but idk how much better you will find it)

[crimson terrace]

oh thank you so generous 🙏 , should I dm you instead ?

[broken lotus]

sure ok!

[visual palm]

Anyone else getting significantly shorter instrumentals on 3.5 for your first generation? Nearly all of the 20-30 I've done have been less than 2 minutes long, while on 3.0 they're 75% of the time 2 minutes long.

[unique stone]

All my 3.5 instrumentals are short. From new songs to extensions.

[broken lotus]

yes, mine too...

[visual palm]

Hopefully they can get that sorted out for the full release of 3.5, I don't mine if it's over a minute, since I can extend it. Just sucks when it generates 17 seconds or so and credits aren't refunded

[vestal rapids]

yes, often ridiculously short, it's definitely something the devs need to address

[ocean venture]

So wav is in frequenzes still not enough to Engineer in DAW am i seeing this right?

[broken lotus]

lol what?

[ocean venture]

I was hoping to Engineer the audio a bit in DAW especially when we are able to use Wave now 🙂

DAW - Digital Audio Workstation

[unique stone]

I don't mind the occasional short generation that I can bring back to life by extending it again but v3.5 just wants to pack up and go home after every couple of bars 😄

[broken lotus]

what would prevent it? LOL

[scarlet mica]

Is there a way to get a "duet" function? Like able to swap singers with each line?

[ocean venture]

Not "Prevent" it's more like... when you have such a high frequenz wide you are able to for example get rid of that blubbering sound suno allways does in the background

[young orchid]

imo at least 16 - 44.1 is enough for a start

[ocean venture]

But the frequenz widness doesn't allow for such "repairing" right now to be precise enough yet

[young orchid]

It doesn't change the fact that it's harder

[ocean venture]

https://tenor.com/view/cats-toby-the-cat-nod-yes-yes-yes-hooman-gif-17105827

[unique stone]

Admittedly I haven't looked into how the data is actually stored in MP3 files, but I don't think 16-bit vs 32-bit float means much considering that in WAV the bits determines how accurate sample amplitudes are. It might just be that MP3 always stores data as 32-bit floats but it won't be individual sample amplitudes.

I mean, I don't think I've ever encountered a "16-bit" MP3. The encoding bitrate is the only thing that seems to change.

[young orchid]

I mean, mp3 is already very lossy, so bit expanding doesn't make that much difference. 32 bit float is also a bit overkill. I think 24 bit wav would be a sufficient choice or maybe it is a limitation?

[zinc stirrup]

Hi, guys, I have a question pls. I am a pro subscriber, and I want more people to hear my music. Do I have the right to use a music distributor to release it on platforms like Apple Music or Spotify?

[broken lotus]

yes with paid subscription you have commercial license to sell/upload to streaming/use in professional projects....,

but one of the mods had an ai 'artist' banned from Apple Music this week... and some of the distributors are refusing to handle AI music.

This is outside of Suno's control, obviously...

[young orchid]

Can I monetize my songs on YouTube?

Yes, but only if you are a paying subscriber to Pro or Premier and, as always, subject to your compliance with Suno’s Terms of Service.

If you are a free user, then you can only use your songs for non-commercial uses, so monetization is not allowed. If you would like to use your songs commercially, you would need to subscribe to one of our paid plans, Pro or Premier. For more details, see https://app.suno.ai/account:

[errant relic]

how do you go about creating a song?

[unique stone]

I get what you're saying - I just think that MP3 is decoded to 32-bit float, but isn't really storing the audio samples like this internally.

[zinc stirrup]

oh, that's a bad news, I mean how did Apple Music think his song is AI created?

[broken lotus]

we are all on the website now:
https://suno.com/create

[broken lotus]

idk... She uses DistroKid as distributor. They had no problems with it.

[zinc stirrup]

Yes, I understood, normally the distributor accept it, but the stream platforms have rights to refuse it to release

[zinc stirrup]

But are there any legal issues? I mean if the streaming platforms think the AI content is not legal to release, do we run into legal problems?

I just want the creation to go to the broader platforms, because I really like this song

[young orchid]

Probably 32 bit float works because after all the machine is generating but 192kbit mp3 is scrapping.
In 32 bit, artifacts maybe occurring more frequently, resulting in a worse sound quality.
I think there won't be much of a problem as the overall sound quality has improved.

[zinc stirrup]

Even I could use Suno's name, but actually I worried they refuse if I use Suno's name

[broken lotus]

there are no actual legal issues, other than the ones that already exist:
(counterfeit, IP, copyright, artist's right to control their likeness).

These companies are part of the music industry, they are taking a stance to side with the music industry.

[unique stone]

I just saved a 16-bit audio in Audacity as MP3 and when I re-opened it, it was 32-bit float. So I am pretty sure the "bit depth" of the audio itself is irrelevant for MP3 and the software you are using is just converting it to the default format that it uses.

[young orchid]

For the time being, high-pitched sounds may be a giveaway in automatic capture

[zinc stirrup]

Understood sir, and the Suno give paid users rights like counterfeit, IP, copyright... so we don't have this problem, is it right?

[broken lotus]

I'm not a sir.

[zinc stirrup]

high-pitched sounds mean beyond human range. It's obviously not a human voice, right?

[zinc stirrup]

Oh, madam then

Sorry

[broken lotus]

Suno gives commercial license and all rights to the song.
Only government can give you copyright (Suno does not know if your lyrics are original)

[coarse scarab]

i need more control from tags on lyrics T.T
https://suno.com/song/7a2bc47a-0e4f-4fda-ba03-4cdfa89f6ece

[fluid robin]

okay now i am confused
i made a playlist yesterday but now it aint there did i forgot to save it after adding new songs ?

[zinc stirrup]

Oh, understood. Thank you, the legal things about copyright of music is too complicated to me...

[broken lotus]

I lost a playlist yesterday and got upset... it was on another page... (a new page I didn't realize had been created to hold the overflow)

[fluid robin]

thats sounds like a annoying problem mostly when u spend time creating the playlist and then boom gone

but u mention it was on a new page were exactly was it located ?

[young orchid]

Interesting. Same as source option is selected but mp3s open as 32bit float. I never paid attention xD
But yes you are right.

[final hazel]

Please keep in mind that this is an English channel. If you want to give feedback here be so kind to translate it first, so that it's not the other way around and everybody has to translate it themselves to understand you?

Por favor, tenha em mente que este é um canal em inglês. Se você quiser dar feedback aqui, seja gentil e traduza-o primeiro, para que não seja o contrário e todos tenham que traduzir por conta própria para entender você.

[rain mist]

suno keeps adding lyrics to my instrumental pieces today?

[final hazel]

This happens sometimes. Especially if you are trying to do it with a genre that normally has lyrics. Hallucinations are still common 😐

[rain mist]

it was just to force an intro, so i can still use it d: but yeahhhh weird nonsense lyrics xD

[final hazel]

lol.. What genre do you have in your "style of music"? And which model are you using?

[rain mist]

v3? soft indie pop, so yeah usually would have lyrics

[final hazel]

Have you tried starting it with [instrumental hook] without the instrumental option? That metatag normally does a decent job with a lot of genres.

[hollow hare]

(This looks like it was a post for a prompt to use in ChatGPT to help generate lyrics for those curious)

[rain mist]

oooo i will add this to my list, thank you :3

[hollow hare]

I found that if you are looking to make a pure instrumental song, use instrumental tags in the lyric box rather than using the instrumental switch.

It avoids hallucination and pads it out to the max time (as well as giving you control on how/what you want)

Examples:

• RayanAIR's Novo Lands which shows how flexible these tags are
• The generic catch-all tags I use to pad random instrumental pieces based on the wiki entry for the instrumental tags (#🪲┃bugs-web message)

[rose crane]

https://tenor.com/view/annoyed-favorite-gif-14675274

I don't want to make songs from any sound. I want my vocals to not sound like absolute trash near the end, SUNO!!!!!

[austere shard]

first impression is 3.5 seems to be better about that. it's a bit more auto-tuney, but less of that muddling chorus effect

[rose crane]

wait...there's a 3.5?

oooooooooo just discovered it

oh man i'm excited

[hollow hare]

For subscribed members atm #📢┃announcements message

[rose crane]

wow yeah it's pretty um......auto-tuney....

but...that also opens the door from some pretty wacky and interesting songs that auto tune would be good for

[hollow hare]

There are some tricks others have found for 3.5, see this post by @untold briar : #📚┃suno-school message

Oops sorry for the ping JF

[rose crane]

I'm a little lost on #4. Can you elaborate a bit more about what you mean, please?

[surreal widget]

Regarding 3.5 released today, am I the only one who have noticed a decrease in quality? I've given the same prompts on 3.0 and 3.5 but find 3.0 is not only more loyal to the prompt but the voices are more human like and rich.

[broken lotus]

if you follow stable diffusion, v3.5 is to v3 as Juggernaut is to SDXL.

It is the same model (over)trained to create a certain 'aesthetic', everything will be skewed towards that aesthetic....

whether that works for everyone... is kind of the point of keeping 3 and 3.5, it's not really a replacement. Might have different strengths and weaknesses, might use one to extend the other...

also it has some tech that's expected for v4 (longer context) but it wasn't trained to make 4min songs in 1 go..., so 3.5 is both an alternate model and a tech preview, but v3 will be staying too.

[prisma wraith]

I'm currently struggling to generating prolonged instrumentals with version 3.5, where in 3.0 they usually always went for the full 2 minutes with some abrupt cuts here and there.
I'm now having issues getting close to two minutes And the pieces I have tried to extend have all given me extremely short generations, any pointers as to what I might be doing wrong?

[low remnant]

Dunno. I'm generating full songs on 3.5. I DO think the quality is down a bit. But it follow instructions to an AMAZING degree. If v4 is going to listen like that, 🤤

[broken lotus]

yeah, it's excellent at the metatags, hates my style prompts... 😅

[surreal widget]

Question: Is suno learning from our thumbs up and downs?

[broken lotus]

yes, especially the thumbsdown.

[surreal widget]

Great.

[prisma wraith]

Real great at following meta-tags, however I have no idea how to tell the AI how long each specific part should be

[surreal widget]

And is it learning our personal preferences or is it learning only on the cumulative feedback?

[low remnant]

But I've been hitting it with stupid things I tried with v3.0, like duets and it changes voices on command. puts instrumentals in right where I ask. music swell, music fades.... stuff that was random before. i feel like I've got more control. I like it. I don't think I'll be making any songs with it to release, but definitely gives me hope for the future.

[broken lotus]

cumulative, from data reports as I understand it...

but it is info that will be used in v4 (which is rough eta 4-5 weeks away....)

[surreal widget]

Also, are the various versions learning from each thumbs up and down, or is it just that model that was used?

I.e. can we expect that 3.5 and 4 will need some ramp up time to learn from our feedback?

[surreal widget]

Thanks so much for the quick and informative resopnses.

[latent lintel]

Question: can i import more voice models ? or i can only use the ones in the library? Because in italian i can only see 9 voices

[broken lotus]

my understanding is that the models are 'baked' and don't evolve once they are made.

but Suno are always working on the next model, while learning from the current one.

Mods had about a day and a half to test 3.5... so it moved pretty fast from the oven to the website.

definitely use votes to guide the next model... it's all relevant.

[final hazel]

No. You can later split them and redo them with yours. But that is outside of Suno.

[latent lintel]

For example i like a model from eleven labs , how could i replicate it?

[low remnant]

Ooooh, 3.5 actually extends 3.0 tracks and it doesn't have sudden style shifts like the other models seemed to do. Sorry, my first try at that. Glad I can keep my folders of intros 🙂

[final hazel]

In Suno you cannot replicate it. It's the luck of the dice. You can stear it to things like female, male, sultry etc. But not the exact sound of a voice known to you.

[hollow dove]

In 3.5, I'm getting vocals generated sometimes even though the slider is set to instrumental.

[low remnant]

LOL, apparently I use the word 'status' like ChatGPT uses the word 'neon'. I notice because 3.5 seems to love to pronounce it staetus instead of 3.0's stahtus. And I'm like "That's wrong!". And the three songs I've dropped in have all had 'status' in there somewhere.

[surreal widget]

OK. That makes sense. I'll do my best to keep voting.

[rocky blaze]

hello i just receveid 10k as ordered but i still had around 3k, they have not been added, please add them too.

[broken lotus]

hi, please email
[e-mail]
they will gt you sorted....

[rocky blaze]

thank you so much for your swift reply!

[wet minnow]

hey discord. do you have ideas what i could write as useful v3.5 TIPS in the wiki? please @ me the stuff you think that should get attention... maybe good stuff, tricks, tipps, things to look out for and not to do maybe?

[untold briar]

For example if you are getting generic metal instead of the aggressive metal you want, add some of the metal niche you are aiming for into the lyrics prompt:

[Verse, thrash-metal]

or

[Verse 1]
[thrash, extreme]

Lazier version: take key style words and dump them in first in the lyrics field in brackets. Sort of pretend it is a 'bonus' style prompt space to add emphasis. This is probably not ideal but even this lazy version sometimes helped give the style a bit of a kickstart.

[angry-rap, bagpipes, bagpipes, phonk]
[verse 1]
Start of Song

I don't have a good guess for the best text formatting, just the overall idea that adding some style descriptors into the lyrics has been helpful nudging the music a little more than just the style prompt. (More helpful than this was in 3.0.)

[indigo heron]

The page ignores my promps on the Phone

Then says crédits refunded next to the song

It only makes stock pop music

[random hazel]

Good day , where can I find v3 meta tag documentation? I rad faq, but no one

[wet minnow]

https://www.suno.wiki/faq/metatags/

[hollow hare]

Some helpful starting links: https://suno.wiki
https://youtu.be/lj8utBd_IUM

[random hazel]

Thanks 🙏

[random hazel]

Thanks 🙏

[hoary plover]

Thank you so much for the update!

I really appreciate the feedback on the nearly 5,000 tracks we've produced so far. The quality of the instrumentals with v3 has been absolutely phenomenal, and it's great to hear that v3 will be staying.

Regarding the .WAV files, do they have studio-quality sound?

Thanks

[warm coral]

how to get a song to end correctly?

also how do we improve the sound quality?

[broken lotus]

I think the layman would hear the wav as very similar to the mp3, for average person on normal audio, wav is not justified.

Sariserpito posted a waveform analysis : #📚┃suno-school message
the mp3 drops off at a high frequency while the wav does have information there...

[broken lotus]

the [end] prompt works.

there are more prompts like
[refrain]
[finale]
[fade out]

https://www.suno.wiki/faq/making-music/how-do-i-end-the-song/

[hoary plover]

thank you

[grizzled slate]

These seem to be one of the magic buttons we are all seeking. It does depend on the Style of Music. I've found some working in one Style of Music and not another. Frustrating I know!

I think it comes down to the fact that ai music generation is trained on what's out there in wonderland of musical data. I don't thnk there will be many [Angelic Voices] in Heavy Metal, if you know what I mean.

However, it would be good if thre was a section over on the Left that was for Tags that Work. Where we could add a Style of Music folder and peopel can add Tags that work in that particular Style. But who am I kidding... it's not going to happen.. is it!?

[unique stone]

Should I still downvote a song if it turns bad after e.g. 2 minutes?

[vestal rapids]

that's what we have been asked to do by the team

[unique stone]

Just not sure how bad it needs to be

Most songs seem to turn bad quite quickly but I have a handful that get bad a lot later

[grave wraith]

Not really complaining...more kinda curious. A lot of voices sound like Autotune (I mean actually it is kinda). I just wonder as the autotune effect comes from corrections... but why would the AI correct its own creation?

And the other thing I am a little bit annoyed of.... I get ever so often a feedback sound. Most times when it's powerful with full band playing. I don't know really how to explain that sound... it's like a tone is in an endles reverb.
Is that also an AI problem or is this part of a style? 😄

Like I said I am more curious about why this happens to avoid it. I am sure it will be improved with every update

[vestal rapids]

the biggest issue I have been facing with v3.5 is noise/audio artefacts after 2 minutes

[grizzled slate]

Yes! every vote good or bad helps

[unique stone]

Seems I'm going to be downvoting an awful lot based on what's been generated today.

[vestal rapids]

what specifically have you been facing?

[grave wraith]

As I can't describe the annyoing sound good... at 03:38 it begins and killing every ear in town

[unique stone]

I'm making instrumentals. Audio starts out pretty good but starts to degrade with metallic ringing after 1 minute in most cases. The song either ends soon after that, or it goes on longer and the audio continues sounding metallic. Sometimes it does this after 2 minutes though.

[vestal rapids]

this exact issue has been happening to me, it's very frustrating

[grave wraith]

Metallic ringing... that could be what I tried to describe 😄

[broken lotus]

However, it would be good if thre was a section over on the Left that was for Tags that Work.

I think this is that feature?

[unique stone]

The actual music v3.5 is generating is quite nice. The instrumental articulations are very good. But then it starts getting destroyed by the audio quality (pretty much the same issue people have had with v3 but much more pronounced.) I'm letting slight degradation slide in case it's just part of the generated music but if it continues beyond a few seconds I'm downvoting - which is currently a majority of my v3.5 gens.

[vestal rapids]

yes, this issue and short generations are what I am facing with v3.5

[willow oar]

is negative tags possible? like to remove a instrument

for a second part

[sonic heart]

Exactly this - I'm loving some of the songs, and really wishing they didn't turn halfway into white noise by the time I'm two minutes in.

[low remnant]

Well, 3.5 is I assume still very close to the 3.0 engine. So things like the ringing or reverb accumulating in a song would not surprise me. I had previously believed it was tied to number of continues but it just might be a thing for the 3 series engines.

[unique stone]

Yes. I'm not surprised by it, though it's happening a lot more than I'm used to. I'm just glad the downvotes will matter.

[graceful condor]

Suno is my favorite tool, and it would be great if Suno implemented "Audio inpainting" like Udio has. That feature would be very powerful and allow room for even more creativity.

Just my two cents. 😉

[hollow hare]

That featured has been planned for v4 for a while

[weary wharf]

Question... For version 3.5, is the time-frame changed as well? How much does the past current model sees?

[final hazel]

I believe double of what it was

[weary wharf]

That's great... I was creating multiple parts from different stages and have 3-4 pieces for a single melody I liked.
I'll try and see what it does 😄

Back to testing.

[weary wharf]

When extending from v3 with v3.5 I do not see the previous quality drops specifically in the vocals. With v3, it was always getting worse.
But it has a tendency to stop and finish the song without me giving an [Outro] or [End] or a [Fade Out].

[weary wharf]

4 minute song yep, but it feels like the new model rushes if I add more lyrics.

Thanks for the WAV download 👍

[elder dune]

Yeah I would like this, too. I tried to produce a guitar solo track - with ONLY guitar, nothing else, [think Eruption] and I explicitly said this in the prompt but got drums and bass anyway. I gave up after a few tries.

[surreal widget]

Agreed. It seems that minutes 3–4 in 3.5 still isn't that much different than the second extend in 3.0. I know keeping AI consistent the longer it goes is hard, but I think it's simply a matter of time. Midjourney made huge strides in less than a year. I'm sure Suno will improve this greatly within 6–12 months.

[livid arrow]

/crip

[bleak helm]

Sorry i must say that the Distortion still live in 3.5 early access. but that why it called 3.5 early access 🙂

oh shit is really bad, dont knw if suno team like the ott

🙂

dont get me wrong i love suno 🙂

[weary wharf]

If you get an output without distortion it stays like that sounding good, if it has it, continues with that.
With v3 the distortion came with an extend even if the original was sounding ok.

But the reverb is still there, de-reverb is a hard thing to do on mastering... 😿

[empty locust]

Hi everyone, I can no longer create sound, the button is created and gray and I can no longer test

[final hazel]

You need credits. You get 50 free per day, or you have to buy a plan.

[bleak helm]

ok that was answer i did not counted for, because, i did not talk about the reverb, but ok i dont want to arguing with you

you are probably right )

i just get some feedback 🙂

sorry i just stop talking if i offen any one

sorry my bad i did not read the answer correct, i see what you mean, my bad 🙂

[cerulean steeple]

Hello. Is there anyway to tell Suno the desired key of the song?

[weary wharf]

No. From experience defining minor or major in the style helps in some genres, but it depends on the genre.

[cerulean steeple]

You mean it can react to "minor/major", but most unlikely to something like "C#"?

[weary wharf]

Yes, defining the key does not work. You can write "major" but "C major" is not working.

[eager topaz]

Yes, most of my 3.5 gens are downvoted and/or reported

[cerulean steeple]

It's a pity. I was expecting that it will work in v3, tbh 😂 But thank you. Hopefully it will work in future versions

[runic dock]

Dead.

[unique stone]

Across 6 pages of v3.5 generations I've only upvoted 2 of them, and even one of those goes slightly bad during the last few seconds but is otherwise acceptable IMO. The rest are either poor quality from around 0:40-1:10 ish (so, downvoted) or are too short to be useful (I don't vote either way.)

[cursive jetty]

ive only managed 2 tracks out of 2k credits 🥺

[lilac garden]

how do you search your own songs when you reached a point where you have so many thousands ?

[final hazel]

Search for your own songs is coming soon

[azure crane]

The missing player in the Extend mode is a big miss in the last updates for the mobile version. Hope they will add it back cause that makes it a lot better to adjust the time for the extension.

[final comet]

Instrumental songs are frustratingly hard to make in V3.5, they keep ending abruptly, and even if you try to extend them the extensions end in less than 20 seconds... I like how much better v3.5 is at following along with custom lyrics but it is in my testing basically useless if no lyrics are provided.

What it does generate is pretty good though, too bad it refuses to actually make any songs for me...

[south kindle]

At least we can still make instrumentals on v3 and extend them with 3.5 if we want.

To be honest, that’s what I have settled into. Making instrumentals and then finding one that matches with how I think my lyrics will flow and then roll the dice until I get a vocal sound and cadence I like.

If it is spot on enough it’s almost like you can’t go wrong with the generations. Other times it’s like wtf. lol

Then I’ll try a different instrumental and see if it is a better fit.

One thing is to remember that you used a particular instrumental for a song so you don’t accidentally make a sister song out of different lyrics but the same beginning.

[final comet]

I enjoy making songs as background music completely lyric free, and while i haven't tried extending a v3 with v3.5 yet I suspect without lyrics I'm not going to get much out of it...

Gave it a try got back extensions that were over a minute long, too back I can't get extensions like that on the 35 second instrumentals 3.5 has been generating for me.

[charred merlin]

Playing around with prompt orders while making instrumental songs & finally getting what I'd expect to hear-- only to realize I forgot to swap from 3 to 3.5. Ahhhh, makes sense. 😛 (not reflective of my whole experience with 3.5, but it reminds me of when Stable Diffusion 2 came out and split the entire development community because it worked better, but differently)

[worn comet]

Why is it that suno's "vocabulary" for prompts is so fundamentally different (read: leaps and bounds better) than udio?

It's so clear that it doesn't work the same way. Udio makes these incredibly "believable," nuanced songs by let's be honest pretty much plagiarizing stuff from the genre/subgenre(s) that you ask for, whereas you can actually ask suno for specifics

[final hazel]

If you want to talk also about Udio, we like you to do that in #🗣️┃chill-chat

[worn comet]

well I know I'm asking for something unreasonable, which is whether people have insight into the fundamental way the prompts work

but it's such a huge sticking point for me

I can't help but speculate

[final hazel]

You can speculate in #🗣️┃chill-chat too?

Or don't get Udio in the mix here 🤷‍♀️

[worn comet]

is random speculation about suno's nuts and bolts not appropriate for this channel?

sorry, I'll move over.

[bronze helm]

Hi, just wanted to say that I'm enjoying using the new v3.5 model. Saw a lot of negative feedback in the channel, but honestly I'm impressed with the improved prompt adherence and the model follows custom lyrics way better.

Probably unpopular, but Bulgarian lyrics sound significantly better. v3 did a pretty good job, but v3.5 mispronounces words less often.

[lilac garden]

Some spanish genres i am using 3.5 is not getting right, while 3.0 is getting right. Wondering if its putting more weight to the lyric meaning than the style prompt.

[south kindle]

I, too, would love to get more info on what tags are recognized, even if it is a roll of the dice for it to take effect.

[unique stone]

Using v3.5 to start a song and then extending it with v2 seems to make it sound better whilst retaining the structure/style. (Just from very limited testing I've just done.)

[graceful cliff]

we don't augment your prompts -- you get what you input, pros and cons

[worn comet]

Yeah, that's exactly what I'm saying. But--well, but this falls under discussion about udio

What I'm saying is, suno has a surprising ability to catch things like... I think a while ago we found that "arpeggio" works

which is wild

it seems to be a fundamental difference

[graceful cliff]

I don't think it just works ....

The model pays attention to your input prompts, but how it pays attention, is not really controlled yet...we have some tricks here and there but it is not like it just follows your instructions 100%. These are things we need to improve on.

[worn comet]

Mm.

But, like, by comparison... specifics like "[genre], [(incongruous instrument) melody], [arpeggiated], [soaring]"

the fact that suno actually knows what that means seems to be a fundamental difference

yes, there can be "baggage" that comes with said instrument

yes, suno still doesn't 100% use that instrument unquestioningly, especially if there's really no kind of precedent for that instrument being outside its home genre

but the fact that it even tries is fascinating

[graceful cliff]

I think language understanding are just emerging behaviors of model sizes, as the competitors scale up, you will likely to see that happen in their models as well.

[worn comet]

Interesting. I assumed the opposite, because the difference is so vast. The others don't nail the prompts the way suno does at all.

Because what suno doesn't do is bring nuance to the instrument voices and compositions. You can hear when the competitor(s?) are getting really great details like that because they're stealing much more deeply from the genres

But if it's not as fundamental of a difference as you say... huh

I'm really curious to see how suno and the others get closer to each other and/or differentiate themselves from each other.

[teal raft]

Features

[graceful cliff]

yeah by the end of the day will be products

[teal raft]

Yep, which is why this whole social angle is so smart

so smart, that's harder to build and make last

[graceful cliff]

and users

[teal raft]

Yess

[unique stone]

I've gone off on a major tangent extending v3.5 stuff with v2 and v3 😅 it's giving some quite nice starting points that v2 seems to "clean" and then v3 is happy to extend.

Is there a particular reason why v3.5 seems to suffer from quality degradation a lot quicker and more often than v3? And any idea why v2 simply doesn't suffer from this problem at all (as far as I can tell)?

[graceful cliff]

Has there been any generated music end to end > 2mins before?

[unique stone]

Ah I mean it becomes a problem from 40 seconds onward from what I can tell

I posted a bunch of examples in #🪲┃bugs-web today, with some explanation of what I was doing

[graceful cliff]

yeah we are all aware of these...

[unique stone]

I hope it's useful for me to do that?

I'm also downvoting v3.5 stuff but honestly I have about 6 pages of songs and only 2 of them I've upvoted for quality. Some are too short to tell (instrumentals seem to end very quickly)

(I'm not that bothered, just keen to actually smash the downvote button on everything that sounds awful so that this problem can eventually go away)

[graceful cliff]

thank you

and yes we are aware of this

it gives me headaches

[unique stone]

Any idea on what's up with the instrumentals being cut short? (especially when extending an old v2/v3 song)

[graceful cliff]

yep some ideas

what's your prompt

[unique stone]

About to go to sleep (it's uh... 4am here 😅 birds are chirping outside) but I posted some specific examples here: #🪲┃bugs-web message

An example style prompt was: "edm, brass section, brass, opera, edm"

In general though pretty much every song I tried extending with instrumental mode on would end fairly quickly. It seems to try to finish the current section of the song and then move on to the end stage of the song (percussion ends, instruments become less intense, etc. before everything stops)

I can send you some more examples tomorrow if that'd be useful?

(some are extended from songs I don't want to share publicly yet so I can DM them if interested)

[sudden spear]

Is there a way to control the voice strength in v3.5 alpha? I'm on part 3 of an original Indie song I'm creating with my own lyrics. Suno is wanting to fade the voice out or it makes it whisper or really soft. I want the voice to remain strong and clear to the end.

[graceful cliff]

nah I have enough

Thanks 😄

[graceful cliff]

you can play with your prompt...

[teal raft]

3.5 seems quite a bit more sensitive in the actual lyric box, slapping descriptors on stuff like [hot chorus]

[graceful cliff]

yep

[teal raft]

and that's just great. So many more avenues for customization

[sly parcel]

I've noticed a lot of my songs in 3.5 are great for the first minute or two but gradually get a little more distorted and noisy around the 1:30-2:30 mark until the end, unless the instrumentals fade out or transition strongly. Things will start very strong but then any distortion will compound over time towards getting a bit noisy and overpowering. Different instruments and the vocals can come through crisp and clean, but it's almost like there's a cycling/reverberating synth effect (or sometimes maybe like a bass? It's unpleasant on the ears, but I can't quite place it as a specific real thing) creeping in along the way. Should I just pick up from before it degrades and extend over the low quality portions? Is there something I should look into adding to my prompt to clean that up a bit?

[graceful cliff]

plz downvote such clips

that's the best you can do to help

[harsh fiber]

Very similar issues. I love the extended length of songs (and the ability to add more in the lyric box) but the overall sound quality makes me feel like I'm best off waiting for things to improve before generating too many more songs on 3.5. I am making sure to downvote the songs that aren't working.

[graceful cliff]

in general, please take downvote seriously

I can argue that ... partially we have these problems is because sth is broken in the feedback loop

[harsh fiber]

Is there a 3.75 in the works for when it moves away from "Early Access" and into the free account individuals? 😉

[graceful cliff]

there will be a 4

[harsh fiber]

Looking 4-ward to it already!

[kindred prawn]

Wheres my vocals. Downvote. https://suno.com/song/28477b3b-4d3a-42ff-a693-0dcf2d85fa43

[grizzled slate]

Do you find they are more usable in different Styles of Music?

[grizzled slate]

Hi and thanks for reply. I do like this new feature. I was referring more to lyrical tags and I should have made that clear. Have great day!

[undone star]

i really hope v4 will maintain v3's genre flexibility and enhance audio quality.

[fringe kelp]

it's the same with every song. i thought it was some kind of watermark because of the beta.

i have wasted almost 1000 credits and from the middle i always have this same effect

[south kindle]

I've just been extending from before it goes sideways to try and keep things on track.

[fringe kelp]

thanks for the tip so we'll probably have to put in more credits until v4

with this song everything gets out of control from 2:00 onwards
i get these kinds of glitches with almost all the songs i generate with v3.5
https://suno.com/song/f32e5534-cdde-479f-bb29-f9edec07bfd3

again from minute 2
https://suno.com/song/d6c300c2-bbbf-4ec7-b367-b57e92d6de23

[halcyon brook]

The sound quality of v3.5 is certainly excellent. But that doesn't listen to orders too much. Especially if we try to connect the songs in the middle, that will try to finish the whole performance, no intention of connecting. On the other hand, didn't want to finish v3. So I tried to mix these two bad things. Create a foundation for your performance with the high-quality V3.5, then switch to V3 at the parts you want to continue and recite the prompts. Once the songs are connected within 1 minute, I start writing the song again from the middle in v3.5. This solved everything in my case.

sorry my English is not good

[fathom relic]

hi, why out of 1000 credit it creates a few good songs ?

i know credit should be sold but it seem so hard to get good quality songs out of 10

1000

any idea

i use same promt, and i have to click over 1000 times to get a few songs.

any idea ? help to solve this puzzle

yes ai model but maybe it should be set when click 100 time at least some good beat and melody.

[grizzled slate]

I like and use Version 3.5 more often than not. However, I've noticed a distinct muddiness/quietness of the vocals. It seems that the frequency has not been set, and the parts of the vocals mix with the frequencies of the instruments. Could you please check and ensure the vocals are forward in the mix? I hope that makes sense!

[grizzled slate]

Hi and let me assure you, your English is great! Interesting idea of switching to V3. I'll give it a try. Many thanks and have a great day.

[halcyon brook]

thank you! There is one thing to note, as expected, the sound quality of v3 is poor (because it is compared with v3.5). Therefore, even if you switch to v3 and create a continuation of the song in one minute, I recommend switching to v3.5 at an early stage. In my case, it takes about 20 seconds to revert to v3.5.

[charred merlin]

Are you putting any work into creating a song by taking the good portions of a song & extending them to create what you want, or are you hoping for a fully-finished song to generate?

Can someone on the team elaborate on how downvoting & reporting bugs works in regards to "doesn't match style instructions"?

From what I gather, it's somewhat of a crapshoot as to how the prompts are interpreted-- as a test, I used "saxophone, dark, jazz" and made different tracks for comparison by adding the word "futuristic" from the first prompt word, down to the last.

The results were incredibly mixed;

• sometimes it would create a futuristic saxophone song, but "dark & jazz" are subjectively left out.
• sometimes it would create a dark futuristic sound, but "saxophone & jazz" were left out.
• sometimes it would create a jazz song with promiment saxophone (sometimes it was trumpet, but I'm not so much of a pedant to yell at the AI for an honest mistake 😛 ), but it sounded like something from the 1920's.

I understand that priority is given to "poor audio quality" bug reports and downvotes, and it's easy to understand how generations would be improved by our community interaction, but I'd love an explanation on how reports such as the ones explained above factor into RLHF, decisions moving forward, etc. Could these actions help to create a model that has a stronger CFG adherance to prompts (even if it would create something objectively "unpleasant")?

[wispy nebula]

Why has Suno become so trashy compared to just last month? Now, all the renderings have the unusable TIN CAN audio and the first second or two is always cut out. Seriously? What happened?

[muted quail]

where do I use bark?

[low remnant]

I think it's just general. Which I know is frustrating. Personally, if a song segment has 30+ seconds of good quality at the beginning and I suspect I could utilize "Extend From" to make something decent, it's not getting a down vote. It might even get an upvote.

Stuff you want to see more of, thumb up, stuff you want less of thumb down, it all averages out and the reason doesn't really matter. And I would not demand a perfect clip before giving it a happy vote, just enough to use in a song. I don't think we're there yet where it'll ever give 4 minutes of yummy goodness.

[wispy nebula]

Also, in addition to the endless TIN CAN "music", and the continuous toy piano motifs, I'm also getting a WEIRD wheezing and puffing integrated into the sound. Yeah, it is hilarious in a way, but spells the end of this site's usefulness. Especially by subscription. I got several that were solely a puffing sound, like someone blowing low notes through a comb covered with paper, LMAO.

And honestly, the exact same archipelago chords in a hundred renderings, even in completely different genres. Less melodies than last month. Maybe AI is running out of "ideas" for variations.

[broken lotus]

There is always the random number generator...

If I am getting variations on my prompt, I chalk that up to the model using my prompt.

If I am getting a completely different genre (a piano solo for instance) then I am sure the prompt failed completely... it's not even part-right.

of course it's a judgement call.

[weary wharf]

3.5 tends to pass style definitions and generate similar songs depending on the lyrics.

If you do not alter lyrics the outcome is less random then v3

Even without style definition I am getting similar songs, it does not matter if I write pop, rock, symphonic, etc...

[sterile finch]

The 3.5 early access is up for those who pay for a subscription

[hazy osprey]

I'm currently experimenting with special characters. I read this somewhere in this Channel.

What I've noticed so far:

. and: influence e.g. Strings, piano and drums.
. a tone
: Double tone

< and > the speed or volume

! bring the block to the foreground. clearer with instruments and with more intensity with speech.

~ makes a “wahwah” effect like a wave.

Are there any descriptions or does anyone have similar experiences?

https://suno.com/song/11c158c2-f275-4c14-8b29-9ae16fe17aa1

[cerulean steeple]

Hello. Is there any way to tell Suno which part of lyrics should sound like a verse/chorus?

[grave wraith]

https://www.suno.wiki/faq/metatags/verse-and-chorus/

[weary wharf]

You can use tags like [Verse], [Chorus], [Bridge]...
Also it's possible to differentiate like [Verse 1], [Verse 2], etc.

[weary wharf]

Check here for more explanations:
https://www.suno.wiki/faq/metatags/

[valid musk]

Can you get female vocals specifically somehow?

[weary wharf]

In the style write "female singer" or "female vocals", "female voice", etc...

[valid musk]

Thanks

[weary wharf]

You can also say soprano, rock female singer, etc... Not always working, depends on the genre you are going for.

[valid musk]

Also, in the past I would paste in lyrics I wanted and it would be cut off. It would also cut them short because the generation would stop at 2min. Now it seems I can paste in more lyrics. Does this mean it will generate using all lyrics I put in without having to stitch parts together now?

Up to 4mins max I see?

[weary wharf]

If you are paying user, yes it's possible. But with v3.5 only, not 3.
More lyrics tend to speed up v3.5, add slow song or slow vocals, something sinilar to style.

Another test result. v3.5 is better at generating duets. Still hard to do, but better then v3.
Specifying female/male or artist names is used in the tags.

Example:
https://suno.com/song/d5ad7c14-03b8-436e-bb47-516fdee4c627

[valid musk]

It also completely ignores the descriptive breaks too. Like [Instrumental intro] . It always just cuts straight to singing. no build up

[weary wharf]

Try

[Start]
...
[Intro] or [Break]
...

Intro is a really hard tag, Suno does not understand well.

[valid musk]

Ah

This is what I'm trying to go for. Just wanted a longer build up at the begining. So I would just add in that [Start] and the ... after?
https://suno.com/song/9e3c63ee-0549-4ab4-8700-300697f32443

[hazy osprey]

Try
[Instrumental-Intro] or
Intro
And add some free lines After this Tag. This works fine for me

You can also try:
[INTRO]
.
.
.
.

This will extend the Intro-Block 🙂

[weary wharf]

If I do not get an "intro", I tend to extend after first verse and use a [stop] or [pause] then add a [break]. I mix the given intro to the first verse.
The second verse mostly starts similar, so it's possible to get a nice intro like this. But you won't have a "complete" song in Suno library.

[valid musk]

I will have to keep tinkering :D The second generate was much better.

[weary wharf]

Be ready to burn some credits 😄

[valid musk]

Yeah, I am expecting this lol

Also while you are here. What discord bots play these songs we are creating?

[hazy osprey]

Just burned 1k in experimenting with specialcharacters today 🙂 But if those characters will really work … you have more control over your Build 💪🏼😎

So if someone tryed too 🙂 im curious about your experiments 🙂

[weary wharf]

Oh I burned a lot while testing tags... These are the ones that are working for me. Not all of them work, as I see that they depend on the genre a lot.

### Structural Tags
1. [intro] - Introduction of the song.
2. [verse] - Standard verse.
3. [verse 2], [verse 3], etc. - Subsequent verses.
4. [chorus] - Main, repeating theme of the song.
5. [pre-chorus] - Section leading into the chorus.
6. [post-chorus] - Section immediately following the chorus.
7. [hook] - Catchy and memorable musical phrase.
8. [bridge], [bridge 2], [second bridge] - Contrasting sections providing a departure from verses and choruses.
9. [break] - A pause or instrumental break.
10. [interlude] - A musical break between sections of the song.
11. [refrain] - Repeated line or set of lines.
12. [outro] - Closing section of the song.
13. [coda] - Concluding passage of the song.
14. [solo] - Focus on a solo performance by an instrument.
15. [instrumental] - Section featuring only instruments, no vocals.
16. [vamp] - Repetitive musical section, often used to extend the song.
17. [build-up] - Section that gradually increases in tension or intensity.
18. [drop] - Moment of peak intensity, often used in electronic music.
19. [fade-out] - Ending with gradually decreasing volume.
20. [tag] - Short, repeated phrase at the end of a song.
21. [counterpoint] - Section with interweaving different melodies.
22. [middle 8] or [B section] - Section providing contrast, usually in the middle of the song.
23. [modulation] - Key change within the song.

### Performance Tags
1. [powerful] - Strong, forceful vocals required.
2. [forte] - Loud or strong vocal delivery.
3. [belting] - Powerful vocal technique.
4. [emphatic] - Vocals delivered with strong emphasis.
5. [intense] - High emotional or dynamic intensity.
6. [climax] - Peak vocal or instrumental moment.
7. [passionate] - Deeply emotional vocal performance.
8. [soaring] - Vocals that rise dramatically in pitch and intensity.

[valid musk]

Amazing bit of info. Thanks

[weary wharf]

Performance tags work well alone, and not with other tags.

[hazy osprey]

Just posted it before
.
:
< >
!
~

Works greate

[vital tangle]

Is it me or version 3.5 doesn't respect to the type of vocals you ask for? For instance if I ask for a gospel choir I get single male voice. Any tips? Or is everyone having the same issue? And songs generated are not 4min long and cuts out the song (and the lyrics).

[hazy osprey]

This should be pinned 🙏🏼

[weary wharf]

Things like this also work. Like let's say your lyric is
Kumdan kaleler, yıkan.

You can do
Kumdan kaleler, yı-kan. <- Gets a nice attenuation for the last syllable.

Also extending vocals work too (only vowels)
Olmayan dualar, göğeee,

[vital tangle]

Thank you for that. Do so what do you do if you want a passionate chorus for instance? Do you put the two meta tags next to the other? Or is chorus on top and passionate before the lyric lines?

[weary wharf]

I first try [Passionate Chorus] and see if it works. If not I try one alone. Fron the lyrics Suno understand that it's a chorus mostly. Or I use one after another. First [Chorus] then [Passionate]

[hazy osprey]

I put (passionate) directly behind the tag
Chorus

😅 or like what the_builder just wrote 👍🏼

[weary wharf]

Ex

[Chorus]
Yalnızca aşka düşenin,
Düşlerine serilir yüreğin büyüttüğü güller
Dikeni battı mı,
Gönül, çok sevdiğini gül yapraklarıyla gömer

[Powerful Chorus]
Yalnızca aşka düşenin,
Düşlerine serilir yüreğin büyüttüğü güller
Dikeni battı mı,
Gönül, çok sevdiğini gül yapraklarıyla gömer

[Chorus]
Ne kalanlardan yarımım
Ne tamamlayandan azım
Aşk-ı ahiretten beri
Yedi cihanın ayazındayım

(Aaaaaa aa aa aaaaa) <-- These also work nicely with Suno
(Oooooo oo oo ooooo)

[lyric sonnet]

is that turkish?

[weary wharf]

Yep

[vital tangle]

Thanks a lot! And since you seem to be a pro, do you encounter issues in getting the type of vocals you want with version 3.5? For instance I'm trying to have gospel choir singing. In version 3 I can get it but not version 3.5. It's very strange. And it's not generating the full 4min either. So my song is always cut around 3min and cuts the end of my lyrics too. Any solution?

[weary wharf]

Suno has a tendency to repeate "repeating" syllables.

Sonunu, hazırlayan. is mostly read like Sonununu, hazırlayan. with an additional nu.

Sonu-nu, hazırlayan. works.

[weary wharf]

I'm currently getting similar vocals from 3.5. I actually got a 4 minute one. The problem is, adding more lyrics makes the model read them faster.
I start with 3.0 and extend with 3.5.

[broken lotus]

Agree!

and DONE!

[vital tangle]

And how do you get suno 3.5 to sing black gospel style all through the song? I only get one or two voices or a duet 🙂‍↕️

[weary wharf]

Do you have an example I can test?

[cerulean steeple]

Is it better to extend from a previous part or the whole song? Your experience.

[weary wharf]

Suno migth not know that genre well. I'm getting pop songs mostly.

I think this hits the spot for black gospel
https://suno.com/song/06ec6504-c1c6-4129-b2a9-08a0beb81254

Too many distortions 😄

[charred merlin]

3.5 seems to incorporate poppiness into most genres, but it just understands lyrical flow too damned well to be upset at...

[stone parrot]

I'll be honest, am not happy with Suno. Why would anyone censor and control people? Especially if it's a paid account. Baffles me how you can't use curse words or make flirty song with sexuallity expressed. Yet this is somehow supposed to take on the record industry in the future but no freedom? It's not like the music I make is for the radio and if it was I understand censoring. Sad. Funny thing is, no single curse word in the song and still denied. Just wow. They need to ease up on their restrictions.

Not sure if this is the board to voice my frusteration.

[charred merlin]

Can't count the times I shook my head at v3 like, "dude, that's clearly meant to be in the next verse, why are you squeezing that line in there?", while v3.5 handles flow better than I would

[unique stone]

You can. I've made an opera song consisting of just the F word a bunch of times with no issues, and a song about someone being excessively over-adventurous in the bedroom. Neither of these triggered any form of moderation.

Yet sometimes people make songs about e.g. computer games involving killing and the moderation blocks this. You can just get around it with a little creativity e.g. mis-spelling words so that the moderation doesn't understand the context.

At the end of the day though, Suno's ToS does state what kind of content the service shouldn't be used for making and that's what the moderation is meant to be in place for.

[weary wharf]

Try using , and . in the lyrics and at the end of the lyrics.

For shit you can use similar pronunciations. Like sheet 😛
Ask ChatGPT for these, as moderation uses ChatGPT, you can find ways to pass.

[charred merlin]

"I eight a person, I hope that I can eet a person again", as sung by Taylor Swift

...waiting for Suno lawyers to kick down my door for breach of service.

[low remnant]

It's never blocked me on shit. I got moderated for blood once, but gallons of blud are fine. I didn't get moderated on the F word, but the singer didn't pronounce it. but she was fine with fah'k. 😛

[charred merlin]

https://tenor.com/view/pedro-monkey-puppet-meme-awkward-gif-15268759

[vital tangle]

3.5 is also not as good at orchestral music as v3. It sounds as played with a synth from the 90s instead of real instruments.

[weary wharf]

From what I understand, model training is using mostly "safe" songs where explicit words are silent.
So f word becomes a silence. Or a beep...

[low remnant]

3.5 sounds generally more "video gamey" to me regardless of genre.

[grave wraith]

It doesn't take your advice every time... I made just 4 tries with Cyberpunk as style and got some poppy songs, even one hard rock

[low remnant]

Well, that's kinda my sweet spot. I tend to somewhere between pop, alt pop, bubble grunge, alt rock, rock. So I can't say much about if it respects other things lol.

[vital tangle]

Ok i got the worst result. If you like Vivaldi, you might, well...puke 🤮

https://suno.com/song/939a2c61-bdac-4478-adc7-694ca6e28448

[vital tangle]

Exactly

[grizzled slate]

Thank you for sharing!

[mystic flume]

Alone like, only one per song? Or just not near each other?

[bright pike]

The biggest issue I am having is the longer the song especially in symphonic metal songs causes the voice on females to go very "vocaloid" sounding adding more and more autotune and shrillness till it's almost lost in the high pitched tones and the words are lost.

[weary wharf]

You can use as much as you like. Too many tags is a problem as well though. Ai starts to embed them to the lyrics.

[mystic flume]

I haven't had any issues getting curse words pronounced. So far the only time the vocals skip a word is when the lyric meter doesn't match the song tempo (which means I need to rephrase the lyrics and re-generate that section.

[surreal widget]

I've tried to use "a cappella" many times but still get about half of them that use instruments. Any suggestions for resolving this?

[hollow hare]

Are you using simple mode?

[charred merlin]

Custom style, randomize lyrics

[hollow hare]

I used a cappella many times in custom mode and never had an issue with it, albeit I did it as a Fusion with another style to generate something a bit more unique

[surreal widget]

Custom mode with my own lyrics.

[mystic flume]

Not sure if this will be helpful, but I have had very consistent results with v3.5 interpreting my instructions for this song I've been working on:

Lyrics here to
start things off!

[Lead In - Electric guitar and drums]

[Intro]
First verse lyrics here

[Break]
[Interlude]
[Instrumental: Guitar]

[Verse]
Second Verse lyrics

[Hook]
Catchy refrain here

[Chorus]
Chorus obviously

[Interlude]
[Instrumental: Drum solo]

[Verse]
Third verse lyrics

[Verse]
Fourth verse lyrics

[Hook]

[Chorus]

[Bridge]
Fifth verse lyrics as bridge
worked really well leading into
a key change

[Key Change]
[Hook]

[Chorus]

[Break]
[Interlude]
[Instrumental: Keyboard solo]

[Outro]
Hook and chorus repeated
using different wording and phrasing
worked really well when labeled
as "outro".

[End]

[surreal widget]

I periodically get piano or keyboards on them.

[charred merlin]

If you've marked a cappella in the custom styles and your own custom lyrics and it doesn't respect those instructions, report it as a bug for "doesn't follow style instructions" and "doesn't follow lyrics", then downvote

[hollow hare]

Usually Suno uses piano when it doesn't understand what to do something.

Can you show us one song you had issues with so we can provide a better feedback?

[charred merlin]

afaik, the more reports they have on when the model "doesn't listen", the better they can tune it

[surreal widget]

Another challenging prompt is trying to get songs that are meant to be more tribal/african/middle eastern flavor with just vocals and frame drum (custom mode, 3.5)

[hollow hare]

I had more luck with this in a test, just had to find the right style

https://suno.com/song/2975b826-54ca-4425-8c8d-30a3cfcc225c

https://suno.com/song/9967a89d-eff3-486b-9b0f-fba875cd6229

Just avoid using catchy because it will make it pop-y
https://suno.com/song/fac7443a-029c-4744-a2a6-b2a5d0a2549c

Mind you though, they were made in 3.0.
If you did not see Johntahon Fly's notes about 3.5, check out #📚┃suno-school message

[charred merlin]

Completely unrelated from the current conversations, but I just want to take a moment to acknowledge how hilarious it is that people complain about something like this not working perfectly as they've imagined.

Makes me think of the Louis CK bit where someone complains about not being able to use WiFi on an airplane-- "do you not realize how amazing this is? That you got to fly through the air like a bird? How quickly some people feel entitled to something that they didn't know existed 30 minutes ago."

Like dude, a robot is making songs for you. This was literally science fiction 10 years ago. Just enjoy it for a bit.

"I can't believe I had to spend 2000 credits to make one song I'm satisfied with."

And how many credits would it have taken to put yourself through music classes, develop music appreciation & theory, book studio time, and mix & master your own work?

[vestal rapids]

but I want it all and I want it now .... seems to be the approach of many

[surreal widget]

I've started downvoting more, but never reported a bug. Is there general guidance for when we should report a bug and when a simple thumbs down is enough?

[charred merlin]

I was asking about that earlier, didn't quite get a clear answer but it's probably not a straightforward answer.

Thumbs down if you don't like it, bug report if it blatantly breaks either of the rules (maybe being lax for instances where it leaves out a prompt or two but gathered the overall theme)

[surreal widget]

OK. That seems to make sense.

[charred merlin]

I would imagine bug reports are more significant in early-access for 3.5

[surreal widget]

Thanks

[grave wraith]

Mine actually too...that's why I took my pop song and used another style to see if it makes a difference and...well...yeah...it sounded even more like autotune but not really like a different genre

[surreal widget]

I'm still working on the songs lyrics, so I'm not yet comfortable sharing, but I'll make some new ones with AI lyrics and will share.

[willow wren]

Made a Test with V3.5 today. Great that it is possible to get longer than 2 Minutes First but we Need larger Textbox than. But it is great that the System improves.

[unique stone]

Specifically if you're testing v3.5, downvote if audio quality turns bad. DON'T upvote if it has bad audio quality (even if you like some of the song.) For other versions I don't think it does anything significant.

For anything else, bug report.

[low remnant]

I thought it was weird myself. That was the only time it ever happened to me. But literally, just dropped in the homonym and it worked with no other rewording/phrasing.

[weary wharf]

I love it when Suno disregards sad, sadness, melancholic, slow in the style, and gives me a summer pop song with catchy rythms for my depressing lyrics 😄

[surreal widget]

It’s just trying to cheer you up, like the good friend it is!

[weary wharf]

Ahahaha. I don't need that, I want to stay sad and depressed 😄

[elder dune]

Found a slight bug. When I change the song image, then download the video, the video still has the old image.

Otherwise 3.5 is giving me some absolute bangers now that I've got a better handle on the syntax

[weary wharf]

As the video needs a re-generation, I think it is not possible to change that now.

[elder dune]

That fair - it's not a big deal just thought someone would want to know.

As a software guy myself, it seemed like using the new icon would be expected, but now that you mention having to regen the video, it makes sense why that would be ... Difficult

[opaque canopy]

i just wish premier plan had the free 50 credit per day because im out of credit since 1 week and the refill is in 15 days lol im a premier user but cant even try v3.5 yet because im out of credit ...

this model is not optimal

[vestal rapids]

just buy credits

[opaque canopy]

yeah sure

[low remnant]

I am a little surprised, I think for v3a they gave all the paying subs 500 credits to try it out. But I guess they feel there's no shortage of people testing at this time.

[vestal rapids]

simple supply and demand, there is much demand

[graceful condor]

guys

[shadow dragon]

you delete unused credits? WHAT? i want my money back

[graceful condor]

how can I do those type of drum

https://open.spotify.com/track/3fEkeE0XSS4aUAuxZohaPV?si=5dJXA5VnQ7uJygS8sJ4YPQ

what is the prompt

[low remnant]

Good luck. [e-mail]. But it's exactly how the terms say subs are handled.

[shadow dragon]

Purchased top up credits do not expire, but require an active subscription to use.

how do u purchase credits and not take a monthly year plan

?

[sturdy flax]

Spoken like a true artist 🤘

[shadow dragon]

oh I see now. ok thanks

[coral plover]

try something like syncopated drums distorded disjuction...

i once asked in the prompt: give me some completely broken bullshit and i got something close to what you got

wait

[graceful condor]

I've tested various prompts, from breakcore drum, glitch, and more, but nothing like that pops up

[coral plover]

https://suno.com/song/28d95267-ac7b-4301-96d7-dd01912d76e5

it was this. the beginning is somewhat similar but with vocals. it returned electronic experimental disjointed as genres so i would suggest

experimental syncopated disjointed distorted drums (as instrumental of course)

maybe surreal? there is a lot haha.

yeee glitch could be a thing.

or something like free jazz. not sure how to make drums out of it. robotic free jazz and then [mixed drum solo] in the lyrics?

https://www.youtube.com/watch?v=hXf7wtX4vWk&ab_channel=JazzatLincolnCenter'sJAZZACADEMY

thats free jazz haha. not mine though

[merry sandal]

Bruh you're here XD

[runic path]

Will Suno ever implement the possibility of downloading the stems, therefore the multitrack of the songs? what do you think?

[low remnant]

When they don't sound like poo, yes.

[runic path]

in my opinion it could be the best update and the real game changer in music production and music industry. I'm a music producer, I think if I had suno multitrack I could improve and bring the songs created with suno to another level with the addition of professional arrangements, mixing and mastering. another music producer like me wants it?

[charred merlin]

audacity + openvino ai tools for 4-track isolation (drums, bass, vocals, other instruments)-- just requires intel architecture

[frozen pivot]

This has been suggested over one thousand times now in the suggestions channel. But for now, Spectral Layers, RipX, Izotope RX, etc. can unstem the songs

[graceful condor]

who r u

[runic path]

yes, but these extractors are all nice on paper, but in reality they extract stems full of artifacts. The solution is not extraction but having the original tracks/stems that make up the song. one by one

[merry sandal]

Zeph

[runic path]

or the midi too

i think is possible to realize but i also think it requeire some work yet

[elder dune]

Just wanted to thank you for this. I've saved off your template and it's giving me a lot of good stuff to go on, syntax-wise. 🤘

I almost feel like I'm able to 'program' my songs now, and rely less on random luck, which is awesome

[clever bloom]

If you're trying to mix up genres, what is the best way to input that into a prompt?

For example, vaporwave music featuring the Spanish Guitar

[runic path]

I tried to do it for a long time but only a few times he understood my song idea.. I burned a lot of credits doing it

[clever bloom]

Same. I managed to get a few done today, but when through a few credits

[runic path]

I stopped creating too elaborate prompts, I understood that the simpler the prompt, the more effective it is, this will allow you to create higher quality songs. unfortunately in this case it is difficult to experiment

[hollow hare]

I love using 'Fusion of X Y Z' but I think 'X Y Z fusion' will work as well (not sure about X/Y/Z fusion).

I recently played around with having different styles for the rhythm and melody.

In your want, I would have played around with something like "Spanish Guitar melody with Vaporwave fusion" or something along that line

[runic path]

i did that but doesn't work fine

[clever bloom]

So you don't use the custom "Style of Music" box?

[hollow hare]

Oh, that was a style prompt in custom mode

[clever bloom]

I see

[hollow hare]

Taking that simple prompt I gave, here are the results as an instrumental (still generating)

https://suno.com/song/99f7aa86-6ee2-48cb-9fa3-8acb71f6ca39
https://suno.com/song/ef7d6668-234d-4638-b134-3434aacb0abc

The first one... it's so lovely

Second one is more vaporwave style while the first was more like a lullaby

[brave solstice]

is there any good way to create a catchy german schlager song? With V3 i got really good results with "german party schlager", but with V3.5 its not that good, to slow and "boring".

[clever bloom]

Ahhh. I never put anything into the lyrics section. I just used the style of music box

[runic path]

currently, after 2 months of creating tons of songs with the premium account and tried tons of prompts I realized that he will only focus on one key feature... in most cases the first feature you list, in some cases it focuses only on the second. I happened to ask for a female voice and he gave me the male one for example.. so it doesn't make much sense to fill the prompt.. I currently work very effectively by using very few words related to the specific genre in the prompt, for example: "punk" or "edm"... everything else is terribly frustrating... I prefer to do this and have some demos created before choosing to extend the one that comes closest to my idea. If I consider it useful for an "important" production I redo the production from scratch and have the piece sung again, maintaining the SUNO initial idea. In this second phase I experiment with the arrangements to bring that starting idea towards my vision.

[clever bloom]

https://suno.com/song/ccb75d13-6fc9-44d1-b978-1aaf7f1ecffe

That's what I got

[hollow hare]

Love it

Do not be afraid to add more keywords to the style.
I usually like doing three sets (each set separated with a comma which gives all an equal weight):

genre, mood, tempo (not as # BPM though)

[clever bloom]

I see.

[hollow hare]

Ah I see. There is actually a lot you can do with the instrumental tags too:

I found that if you are looking to make a pure instrumental song, use instrumental tags in the lyric box rather than using the instrumental switch.

It avoids hallucination and pads it out to the max time (as well as giving you control on how/what you want)

Examples:

• RayanAIR's Novo Lands which shows how flexible these tags are
• The generic catch-all tags I use to pad random instrumental pieces based on the wiki entry for the instrumental tags (#🪲┃bugs-web message)

I really should do what Rayan did

[clever bloom]

I'll have to experiment with those then.

This is a city pop fusion I did earlier

https://suno.com/song/91d1175a-13dc-423c-8614-fc6bd75defa0

I'm definitely going to try out some of the instrumental tags. I was having a rough go of it making a song with lyrics and thought making a pure instrumental would help my mind

[twin sandal]

Hello, i make a song with suno.ai but i dont can deploy it over Routenote or Tunecore. It failed because they say that are different vocal Samples in the song and dont accept the License from Suno.
Have anyone deploy a Song? Can you send me which partner you use? Regards Sascha

[unreal needle]

#أجْـــرَادَة مَـــالْــحَــــــة

فِــي كُـنْـتـِـي سَـارْحَـــة
فِـي جْـنَـانْ الصَالْـحَـــة
آشْ كْلِيتِي آشْ شْرَبْتِـي
غِيرْ التْـفَـاحْ وْالنَّـفـَـاحْ
وْالحَـكْـمَـة بإِيــدِيـــكْ
يَاالْقَـاضِي يَا بُـو مْفْتَـاحْ

Video

[runic path]

distrokid

I'm publishing tons of songs created with SUNO on spotify

[twin sandal]

Very nice, thanks

[final hazel]

ما بْقَاش مُمْكِن تَدِير الموسيقى فْ ديسكورد. ولكن تْقَدْر تَدِير الموسيقى فْ الموقع ديالهم: https://suno.com
وْ زِيدْ: خْصْنَا نْتَكَلْمو غِيرْ بالإنجليزية فْ هاد القناة.

[twin sandal]

Which Language is it?

[final hazel]

Moroccan Arabic (Darija)

[final hazel]

I personally use Ditto Music.

[twin sandal]

Ok Thanks

i test it. Thanks for you response

[magic axle]

and spleeter, https://github.com/deezer/spleeter

[frozen pivot]

Another great choice. Seems like the father of all the others.

[cedar grove]

Hi! I will simply stop using 3.5. It didn't get a single result right, and all my credits were wasted trying to make the model better. I will let the free users start using it next week and hope they downvote it correctly.

[charred merlin]

Yo, are we discussing track isolation? Has anyone else tried Audacity + OpenVino for comparisons with anything else? I'm seeing "Spleeter" for the first time, might look into that, but curious if anyone has comparative notes between different methods?

I try to add a couple shifts when using OpenVino to get different perspectives on the track isolation, but it still "warbles" a bit in some spots where notably loud segments are playing in other tracks

[sharp kraken]

https://suno.com/song/fd937148-7f7a-4158-8346-cf13bf197f21

I split with rx10 it's cool i think its nice cause you setthe seperation value so you canease up on simpler tracks

[weary wharf]

I use Logic Pro's own stem splitter. Has 4 channels max (vocals, drums, bass, others) but works pretty good.
If I need a more-channel seperation, I have RipX for that. But is not as good as Logic Pro when it comes to vocals. Too much bleeding.

[charred merlin]

Do you find the "warbling" happens on the tracks adjacent to a loud section of vocals or horns with Logic Pro's splitter? (Not familiar with technical jargon if there's a better way to describe the phenomena)

[weary wharf]

I don't know if I understood. Do you mean loud vocals appearing in different stems?
If so with Logic Pro never had that. But on RipX the high pitch (and specially belting) vocals sometimes appear in Guitar stem.
In Ripx you can "move" those parts to the correct place actually, hard to do, but a real good feature.

[charred merlin]

Sometimes I get bits of tracks bleeding across in that manner, often background horns will pop in between 'other instruments' and ' vocals'.

I'm referring to a temporary warping or garbling that occurs, for instance, in a guitar track once isolated from a dominant vocal section. During sections where the vocals are particularly punchy, the guitar can have "audio artefacts" that can make it sound like it's coming from within a well, or behind a wall.

If we're able to upload .mp3s in this server I can share a section if I'm not being clear

[weary wharf]

Aaah. Got it. Yep that happens. It feels like there is a background voice that raises the gain, and feels like someone added a reverb in that section.
It happens on all stem generators. It's all AI based, and models are not 100% accurate.

Someone needs to access WB's or Sony's database and re-train a model with better stems and final non-mastered output.

As it's a technical discussion you can uplaod an audio file.

[nova ledge]

I'm not sure (obviously) exactly how Suno trains on user actions or utilizes reinforcement learning from human feedback (RLHF), presumably it's with the rating system and general user actions. There seemed to be a sweet spot in time where I was getting cool results in V3 more frequently, but that went away to the point where 50+ generations in a row won't yield anything that is compelling. V3.5 has been all misses so far, imho. I think there is an issue with 'accounting for taste'... we're not very good at discerning/agreeing upon what constitutes good music, so Suno's 'taste' gets worse or more 'average' with more user feedback not better...

[rustic gale]

V3 alpha. was best one so far. Now it degraded a lot

[weary wharf]

I was talking about these... High pitch piano in guitar stem... So frustrating 😄

[rustic gale]

This was my thoughts as well. To generate different tracks and not 1 master.

[weary wharf]

If I need stems, it means I'm re-arranging the track with my own instruments. I do not care about the sound quality that much. It's just for that I can hear the notes so I can rewrite the track.

Not using for mixing..

[rustic gale]

Well in any case transcription would be much simpler any way.

[weary wharf]

You mean getting as midi? With Suno it's hard. Too much reverb gives a lot of "false" notes.

[rustic gale]

nope i mean transcribing songs (aka putting them into notes instrument by instrument). Even though it is possible with anything, and suno output as well ofc, it still would be much simpler with split tracks.

[weary wharf]

Yep that's what I meant with re-arranging 😄

[elder oyster]

Been getting lots of distortion lately. Burned through quite a bit of credits to get a couple songs decent enough. Anyone else have issues with this? Is there a fix or suggestion that may help? Thanks in advance

[vestal rapids]

which model?

[weary wharf]

There is no solution. Try to change the lyrics a little bit. If you are getting the distortion in the extended parts, change the seconds, go 1 or 2 secs before or after.

[elder oyster]

3.5. I agree with a previous comment that there was a sweet spot with 3, but it now occurs in both

[elder oyster]

It happens with original generations as well with no extensions.

[weary wharf]

Then only was is to change the lyrics or tags a little to get some some alterations.

[elder oyster]

Bummer. But thanks for suggestion. Fingers crossed v4 will be better in this area

[weary wharf]

🤞

[rustic gale]

Well even in their promo video v4 sounds bad.... I hope it will get fixed as well.

[neat field]

taking notes here, i may test too eventually

[graceful cliff]

that's not v4 promo, that's voice input promo

[rustic gale]

oh i was sure that will be in v4. My bad. 🙂
Any way, unfortunately still it sounded not good.
We all love suno and hope it will get better 🙂

[magic axle]

I couldnt the module installed in my arch distro for some reason. But I have a good RVC ui that does stems as well as spleeter. I like to take stuff in xtts and rvc to train other voices.

[graceful cliff]

It will, help us with feedbacks plz. Downvote on noisy clips and bad clips.

[rustic gale]

Yep. Do it anytime i get rly bad quality 🙂

[sacred cobalt]

In general, I recommend this idea to the developers:

it makes sense for AI to analyze what exactly people like and why, finding “hidden patterns” of music - and in the future it should strive to generate not even like people, but perfect music from Paradise (or the singing of the Gods),

(you won’t believe me, of course, but I remember a number of worlds from posthumous experiences with such music, and I’ll tell you that everything you’ve ever heard on Earth - in comparison with it, it’s just like rattling bones in a cave, in the cave ages)

P.S.2:

Well, we are waiting for devices to appear en masse that allow us to read feelings, emotions, mood and the degree of pleasure experienced by a person - in real time,

Then, on millions of people, it will be possible to self-teach musical AI (like AlphaZero) based on the real response of their feelings, checking in real time the greater or lesser pleasure a person experiences from the smallest (and not so small) changes made in samples, etc. (ultimately, up to the control of every bit in a byte of a music file, so that it brings only maximum pleasure)

when this becomes possible, you will really be blown away by how incredibly awesome it is (so much so that it is now fundamentally impossible for you to imagine and even remotely imagine it)"

P.S.3:

You can, of course, not have all this yet, try to analyze trends - but there is such a problem that there is basically some kind of bullshit (except for rare exceptions), I have a playlist (which I generated 40% myself) and even more interesting

here it is by the way,
https://suno.com/playlist/cc9c484d-92f6-400f-8fe6-29eb5d9485b1

(it might make sense to enter a rating from 1 to 10 stars for songs, which any registered user can set - and then average it over all users who rated the song, displaying the result),

in fact, even one assessment (for each new generation, including unsuccessful ones) from its author will allow, on a large array of such data, to feed it to the AI so that it searches for (and in the future predicts) what brings greater musical pleasure to people

P.S.4:

and let's go over the concept again,

1. “if” Heaven (paradise worlds etc) exists somewhere, then no matter how wonderful the music sounds in it, it can always be reproduced on Earth on any digital audio system (if you get the corresponding sequence of bits in the bytes of a .wav file)

2. the bitrate of modern music is limited, so the total number of possible, for example, three-minute compositions is not infinite, but imagine that we could extract (from this multitude of them) only the most awesome ones? (this is perfect music from Paradise)

P.S.5:

Well, the point is that we need to somehow try to expand the classical (i.e. modern) approaches (the so-called “human”) to creating music, gradually expanding the permissible combinations of what we can hear from SUNO,

especially after, using them, you finally teach SUNO AI to generate (without failures) music at the level of the best of people, you will still have to develop somewhere further after that, striving for even greater heights 😄

[hardy bane]

One note based on my recent testing as well, YMMV -

I've found some success by using tags like [Repeat Pre-Chorus], [Repeat Chorus], [Repeat Verse 1]
which saves on characters, and effort to copy/paste/update multiple copies of the same lyrics

🙂

[weary wharf]

You can also do [Chorus x2]

[hardy bane]

nice

seems to be even better if I specify [Chorus 1] and then later [Repeat Chorus 1]

e.g.
[Verse 1]
Lyrics
[Chorus 1]
Lyrics
[Verse 2]
Lyrics
[Repeat Chorus 1]

I've had less success when I've just done [Chorus] at the end 🤷

[weary wharf]

What do you mean? Different chorus?

[sharp kraken]

If you have rx11 or 10 you can run spectral repair after the seperation and also adding harmonics with something like kelvin makes it sound really good

[weary wharf]

Oh I've never had a good outcome from rx11...

[weary wharf]

Do you have a process that's working well? Which kelvin app/plugin you are using?

[hardy bane]

(what's rx11)

[weary wharf]

https://www.izotope.com/en/products/rx.html

There is a trial version and good youtube tutorials from izotope...

[hardy bane]

nice, thanks!

[sharp kraken]

Kelvin 2 in fl studio

We get wavs now which changes a lot

but with mp3 the process should go dx10 - resample wav 32 float

dx10 stem seperations (conditionally Deverb after stems)

dx10 Spectral Repair.

Import WAVS to FL Studio

If i have a "busy Stem" I use another stem seperation.

Then double everything and play the original + Pitch modulation / Original + timbremorph

after timbremorphing all 4 tracks congrats you just wrote 50% of the audio according to us copyright law

Just make sure you can tell the difference

^MY VERSION

SUNO

A little samesy till i start playing my synths

but they are totally divergent within a minute bc my weird ass synths

synths add some glue too suno is really minimal with layering

I also like to redo some of the lowend

Mostly Made with Morph 3, UhE Bazille and good ole MMFM - Ozone master with my p42 and some kelvin on the low end

Its stupid loud - technically too loud for youtube

[remote flume]

Wow, this v3.5 is pretty good

[meager inlet]

Uhhhh not sure if this would be the correct chat for this issue or if it is even Suno or my computer doing it but last time I downloaded a WAV file for a song it downloaded twice. I thought that was just my bad but just now I downloaded another but this time I noticed it wanted to download a second time again so I hit cancel on the download and so it tried over and over and over again. I had to cancel it every time and it wouldn't stop until I accepted the second download. Has this happened to anyone else?

[stuck kernel]

Converting a mp3 to wav 32 float will do absolutely nothing to change sound quality. There’s no “resampling” going on, really. It will still be the same mp3 in terms of sound quality, only taking up much more disk space.

[stuck kernel]

What?! US Copyright law? How? Please explain..

[sharp kraken]

Um it's the Spectral restoration that rebuilds the mp3 into a wave

but you encode it first

Im the one playing suno on F1's my guy

you have to change a peice of music by 20% in order to claim it as a new peice of music

theres lots of people explaining it on youtube if you need more info than that im not a lawyer and Joshuasca and PSYAI do not offer any legal advice regarding the authorship of AI music

You can look up spectral restoration as well, you can use the prompt [Spectral Manipulation] or include it in your style prompt to try it out

essentially it uses FTT to "reimagine" the original audio before compression using a complex timbre simulations in Matlab

[frosty crest]

Quick question, if you sub to Suno, and then after a year you cancel your subscription, do you still retain your rights to the songs you made within the year you were subbed for?

This was not covered in the FAQ nor could I find it in terms of service, and if it was perhaps the terminology used was beyond me

[thin flint]

With 3.5, since we get four-minute generations now, how much can be put into lyrics in a single generation?

[unique stone]

Yes - anything you create whilst subscribed remains yours to do what you want with it even after your subscription ends.
Similarly, anything you create outside of a subscription isn't yours even after you subscribe.

[unborn latch]

Hello! I tried to sign up for monthly payment, but I subscribe the year payment by mistake. Just wanted to ask if it's possible to get a partial refund?

I noticed it after using 60 credits...

[unique stone]

You'll need to contact [e-mail] and explain it like you have here. They might take a few days to get back to you though.

[frosty crest]

Thanks! That particular scenario isn't wholly clarified in the FAQ so I appreciate your response. I'll send an email to the support team too just to make it entirely clear and in writing lol.

[unborn latch]

They did reply that they received my email. and they said it takes time to answer. anyway thanks!

[unique stone]

This bit from the FAQ (my highlight in bold):
"If you are a paying subscriber to Suno, then you own the songs you generate while subscribed to Pro or Premier, subject to your compliance with Suno’s Terms of Service."

I guess it's possible to read this differently (you own songs whilst subscribed) - it's quite a common question in this discord server so if you search around you'll get some further confirmation 🙂 (as with billing questions it might take time for them to reply to an e-mail to the support team but if you want to make sure, go ahead.)

[frosty crest]

Yeah that part tripped me up because I've worked in legal for a while. The stipulation "while subscribed" could mean either you own the tracks made during a subscription period even after it ends OR you own the songs while you have an ongoing subscription.

In legal, I think they could swing the argument to their advantage either way unless the exact scenario is clarified explicitly.

[unique stone]

The actual Terms of Service might make it clearer

"Subject to your compliance with the terms of this Agreement, if you are a user who has subscribed to the paid tier of the Service, Suno hereby assigns to you all of its right, title and interest in and to any Output owned by Suno and generated from Submissions made by you through the Service during the term of your paid-tier subscription."

[frosty crest]

Yeah I saw that in the IPR section and it's kind of the same situation where the stipulation is not clarified

[chilly leaf]

God damn it's annoying that both V3,5 and V3 now randomly end songs without me specifying it, zzz

[kindred prawn]

Can you do this in Reaper ?

[sharp kraken]

Nah but rx integrates into reaper

as far as I know only rx10/11 does spectral repair

trust me i looked for something cheaper

if you have matlabs you could build a timbre simulation - but thats like - software engineer not simple rea scripts

[kindred prawn]

Maybe something to worry about in the future then. Its not cheap, especially when you are kinda just doing this for fun atm.

[sharp kraken]

yeah just get the waves for 10 bucks a month

[wispy nebula]

HUH? You mean there was a time when AI actually "listened"? I can't even get BING to listen. Lucky.

[chilly leaf]

V3 at release did listen, as in it (almost) never ended prematurely unless you specified [end]

Or rather, it's not that it did listen, more that it never ended prematurely in the first place

But it now does

[untold briar]

Ending prematurely is probably something tweakable, if I had to guess, without model changes. Always better to err on the side of going to long IMO, ever since we got the Extend From custom time feature.

[chilly leaf]

Most likely yes, considering that there was a "bug" some time ago where V3 would do the same thing

[untold briar]

I have only really had a problem with it in instrumentals in 3.5. 3.5 usually finishes lyrics.

[chilly leaf]

Talking about instrumentals as well, but others have been having lots of issues with extending songs now

[lone oak]

Thanks to 3.5V for saving a song that succeeded in the first half but failed in the second half. I wonder if 4V can force a half-minute intro to be added?

[lone oak]

It's strange that 3.5V can't expand the song from a specified time point. Has this happened to you?

[untold briar]

It is possible that you set a Extend From time point, and then later selected model version 3.5. If you do this, the Extend From time is reset back to the end of the clip. This behavior drives me nuts and I waste a lot of credits because of that.

[lone oak]

haha, this bug is really as tragic as V3

[untold briar]

It's awful, because looking at the generic clips it's not always clear it reset the time unless you Get Whole Song. So you just wonder why they sound strange or maybe not even notice until many generations and credits are spent.

[sharp kraken]

Yeah instrumental is broken but you can switch models too

[sharp kraken]

Oh man hell yes - the entire ui resets way too often. You cant undo an ai lyric generation either. I have lost over 20 hrs of my life

the whole ui looks nice but is functionally jank and you need a txt editor to make a suno on

otherwise suno will just delete your work= especially the big 3.5 guys

[torn brook]

Yep, learned to select all and cut before the ai lyric generation conpletes

[sharp kraken]

cntrl a+c yeah ive owned fl studio for 24 years im quick on hit keys, but i use laptop

so it happens alot

(touchpad)

[untold briar]

I am amazed at how many people create complex prompts on mobile, without using a text editor open in a different wnidow for that stuff

[sharp kraken]

Its got a really wonky scale in edge i dont think it was designed for a pc

i use bing and its so squashed with copilot

yet the only thing that seems to know how it works is bing LOL

bing is really good at suno,

until you get it to make a track

and 50% of it is hallucinations

and Im pretty sure theres 2 prompting formats? I see alot of people using some really weird syntax thats like a grocery list

Prompts for Bark and CHirp different? who know - I have the source code but havent really picked through it

[untold briar]

Grocery list is the default. If you mean words, separated, by, commas. But it's not a hard rule.

[sharp kraken]

i use the []'s

like lots of them

and they do a lot.

spectral manipulation reset genre all kinds of stuff

[untold briar]

Bark is mostly not related, except in a historical sense. [tags] in bark were just an accident. And in Suno music, Chirp, Suno decided to intentionally make tags work like [verse] and [chorus] and later also generally. But in bark is purely just 'oh look it sometimes works'.

[sharp kraken]

I have a text file that expands on the PDF

you can change the entire composition by putting sdj before verse

thus dark verse, dark brutal verse, and brutal v erse

[untold briar]

I've burned enough credits on tests for awhile, but one thing I forget to test is the | character in 3.5, in the lyrics, between words.

[untold briar]

sdj? is that short for something or do you mean literally those 3 characters?

[sharp kraken]

hunh

[lone oak]

Only the emotion and singing style in the lyrics occasionally take effect, and the intro marks are often ignored.🤣

[sharp kraken]

adj

typo adjective

Nah you just go Dark twisted melody

brutal harmony

off-kilter rythym

constantly increases tempo

[untold briar]

Ahh, you mean like:

[angry, build]
[verse 2]

Yeah 3.5 benefits from lots of extra brackets in the lyrics

[sharp kraken]

dude i have thouands of unique commands

I write experimental darkpsy IDM

[untold briar]

You can even [go wild and write whole sentences] which won't be followed literally, but will influence style in useful ways

[sharp kraken]

nah that can really screw it up

i dont reccomend it

they all need to be in this order

[ADJ NOUN VERB TARGET]

[untold briar]

I forget that not everyone is cool with hallucinated words in their songs. If you don't mind them, you can push things a lot further.

[sharp kraken]

So [Funky Beat Drop] but not [Drop funky beat]

definitly not it rap it sounds rough

i like it in my folk stuff

Ive had a few funny ones tho

[lone oak]

Use more punctuation and spaces and you will get better singing results

[sharp kraken]

definitly

all my fast rap is written phonetically

ahl mah fass rap roet like dat

[untold briar]

My feeling is that in general, more diversity in the prompt = better singing. Punctuation and spaces is one way to add that. Some of the reason I believe this is that my very favorite voices have almost all came from lyric prompts with tons on pure noise, like morse code: https://suno.com/song/f193f729-cd82-48e9-8454-39e48235e3cf which is taking that to to the extreme.

That isn't practical when you actually want to use real lyrics. But what you can do is prompt like a madman for Clip 1, just to find a great voice, and continue it.

[sharp kraken]

it goes like 2x faster

what are these verses

is that just like a vocal break or is it being generated by the dots n slashes?

[lone oak]

Use "~" to lengthen the tone, use "-" to speed up the speech, and use spaces with emotional descriptions to connect "!" to get emotional reinforcement.

[untold briar]

The dots and dashes are not being understood, probably any similar noisy-looking punctuation works. If you put words like 'vocals' in the style prompt, Suno will try to sing, even if there are no real words in the lyrics. In fact even with blank lyrics: https://suno.com/song/6c3d296e-ed43-459e-b9b3-a69c798a2c70

[sharp kraken]

Yeah i do that like "o o o OoOo ooo o e EEE eEe e e e ee"

but with letters

it seemed like it was controlling it in your guitar scratch one

[untold briar]

Oh it does influence it, like you might find the length of the line matches the length of the music roughly. I just mean the morse code is not actually being interpreted as lyrics, or as musical notes. Rather it's a semi-random pattern

[lone oak]

In addition, SUNO's ballads don't feel like ballads. I think blues rock is more interesting. Let's try it with the "broken voice" Style of Music tag.

[sharp kraken]

like this https://suno.com/song/077e4219-ee47-4952-80c3-3d80e32111c9

[untold briar]

It's been different in every model version. For example, in V0, the speed of the song was influenced most strongly by the amount of lyrics in the prompt. Lots of lyrics = fast singer and the opposite. In V1, I think line breaks were the biggest faster. So like

Hello
How are
You

Would be about the same speed as

Hellow, How are you
This is whole new line
And one more line

because it sort of paced by the lines.

[sharp kraken]

girls seem to do it better

[untold briar]

Line breaks still do that now, but it's just one of many factors

[sharp kraken]

to do rapid-fir drills i actually have to break almost all the words into single or double syllable phonestics

[untold briar]

Maybe it's more of a factor in 3.5, actually, than 3.0. This seems to help 3.5 make longer or shorter instrumentals, padding them out like

[instrumental]
...
@@@
...

[sharp kraken]

like it can do UNTO but not OBLIVION you need Oh-Bliv-EON

on fast paced tracks at least

[untold briar]

Especially if that [instrumental] tag is the song intro, before the first line of lyrics. In that specific case it can help a lot to pad it out like that.

[sharp kraken]

I dont mix that with vocals

i use it for my psytrance

[untold briar]

Are there real life genres of music that have lyrics written like Oh-Bliv-EON? There must be something in the training data (rap battles?) so Suno understands that as fast?

I can't remember if other versions were sped up with dashes. Is it also faster in 3.0? Don't think I tried that.

[sharp kraken]

no i think it has todo with the vocal structure of rap

like a lot of it is triplets these days

just triplets doing triplets with someone singing stacatto triplets

so 3 syllabels fills up a triplet

[untold briar]

I followed you, keep up the experiments, good stuff!

[sharp kraken]

IF (1) I (2) trainwreck (3 +4) so now it just litterally trainwrecks the triplet flow

[lone oak]

The intervals between lines and paragraphs of lyrics are often confusing. The gap between [Verse 1] and [Verse 2] is often simply ignored. The interval between [Verse] and [Chorus] is effective. Currently I don’t know how to make suno understand that there are gaps between sentences and between [Verse 1] and [Verse 2].

[sharp kraken]

but if its train-wreck the ai can continue triplets

gap doesnt matter only command cue

you can also overload the command cue and carry stuff that was supposed to happen in the last frame into the next one

thats why i like to make all words as small as they can be. "The The The" "the the the" "da da da" "da-da-da"

[untold briar]

I am not sure of the answer either in 3.5. You might try what I suggested for instrumentals:

[verse 1]
Last line of lyrics
.
.
[verse 2]

Or something in brackets, not sure what word.

[verse 1]
Last line of lyrics

[something here?]
[verse 2]

[sharp kraken]

all those will go progressively faster

i put periods on all the blank lines

[lone oak]

If possible, please tell me how to describe it?

[untold briar]

If all else fails and you don't mind adding filler vocals

[verse 1]
Last line of lyrics

(Ahhhhhh, ohhh oh)

[verse 2]

[sharp kraken]

if a tag breaks and it hits a period it stops

[lone oak]

It would be better if suno allowed me to describe how many seconds the intro is, or how many seconds the interlude is!

[sharp kraken]

I dont put text next to brackets either

[lone oak]

break and fade out can take effect, but often it’s time to climax immediately

In V3, I tried putting (Ahhhhhh, ohhh oh) before Verse to add an intro, but I found that suno skipped it directly. (Ahhhhhh, ohhh oh) You have to put [...lead-in verse] before it to make sure it takes effect

[sharp kraken]

cool to remix?

[untold briar]

This is not easy. For an intro, many people create an instrumental Clip 1 (no lyrics) and then Extend From a custom time with the lyrics in Clip 2. This gives more control of intro time.

I have not tested, but using lines of punctuation to length could be an idea.

[instrumental]
@@@
@@@
@@@

I picked @@@ randomly, no idea if that is good.

[untold briar]

Sure

[sharp kraken]

i dont even know what fugue is

[untold briar]

The "pop" one is actually a V2 song (only the last clip is 3). So it may be strange when extended.

[lone oak]

Then I'll try again, and I'll have to be trained by AI again.

[sharp kraken]

im gonna remix thenprompt

you dont use cap?

[untold briar]

To save credits, use search https://suno.com/search sometimes can find examples, and not need to use your own credits to test. Especially good for style words.

cap?

[lone oak]

This song has too many Style of Music tags, maybe four or less will provide better control.

[untold briar]

If the song sounds too 'generic' in 3.5, I recommend using many style tags. I think in those cases it is helpful. This may not be necessary for some styles.

[sharp kraken]

I know I was learning cross-genre when i made that

[untold briar]

Overall I find many style words are weaker in 3.5. Often I have to repeat the word. phonk, phonk, phonk instead of phonk or it doesn't seem to work. Sometimes even more repeating.

[sharp kraken]

my cross-genre stuff got much higher weight on the cross

with 3.5

the jungle drill stuff is fantastic - drill tends to sit at a halftime of oldschool 90's tech-step

and I really love to mix orchestral jazz and freak folk

and the freak folk is totally different

you can just switch 3.5-3.0 and instantly tell

thats the secret to make 180 plus music is cross-genre

Ive made over an hour of music 200+ so far

AFTER daw-work BEFORE DAW - this is what i do so ya know those wav downloads are clutch

[weary wharf]

Thanks for the process. The songs I generate all have lyrics, my main problem is with the vocals. Rest I have east west sub + kontakt so instrumenst is not a problem.
I am trying to find a way to recover the broken vocals.

[bleak bobcat]

Any trick for the prompt to generate only instrumental music? Because even if it is checked in the option, it gives you voices, even with the bird covers. Today the 5 generations are all wrong. It's the way I've found because within the lyrics you usually skip them, and I see I'm not the only one.

[sly mulch]

/how

[weary wharf]

Try using the custom mode, in the lyrics just add [Instrumental], and write your style.

[bleak bobcat]

Thanks. I'll try this. Waiting for the public new version 🙂

[lone oak]

https://suno.com/song/7133267d-c3a1-4ddb-b454-3c7536ddc3b0

https://suno.com/song/aeed658b-0ae3-46f9-85cf-b8da297beba0

[shy sand]

Hello guys, there is no possibility that when I make an instrumental in the custom way I can indicate the measures in the intro, verses, chorus, etc., measures or time so that it doesn't make it too random.

ty

[valid musk]

I made a new song and its ruined by the start lol. Why is there some random scream. Can I get rid of this somehow and keep everything else the same?
https://suno.com/song/a7a67e59-a295-44b4-8bb7-06b801dbfa56

[broken lotus]

the feature to 'in-fill' or 'in-paint' (re-generate over parts of the clip) is expected for the next model, v4
(coming probably in 4-5 weeks)

[valid musk]

ah thanks for the information. I will wait till then.

[agile anchor]

bro how do i make clear voice, i mean, i want it to be just 1 voice (no harmonies)

[valid musk]

This is weird, all the generates are getting this weird scream at the start lol

[shy sand]

edit with audicity, is easy

[deft quest]

I'm having hard time making the AI understand that I want to swap with one man singing and sometimes others men singing. Is there a way ? Same for a song without instruments

[broken lotus]

I got this to work with
'male solo' and 'mens choir'

https://suno.com/song/3c56620f-c8f1-4204-8cd6-9489e8b22b81

(it was v3 so it didn't follow my tags perfectly, but it got the idea and went with it)

[sharp kraken]

VOCAL BREAK

[lone oak]

Can you describe it in detail?

[sharp kraken]

Once you have both vocals in the track, you just [Vocal Break] and tey usually switch

or you go Duet and they sing at the same time

[fierce bone]

I have a DAW program where i might be able to! Send me a DM and i can try to RIP the track and just edit out the scream for you

[sharp kraken]

duet has to be in your style

[broken lotus]

The Devs have said it is an expected feature of v4...
I havn't seen it working, but in theory you'd 'lock' some of the clip and then re-generate it.

the 'locked' parts of the clip stay the same, and new parts are generated where you left 'unlocked'

so if there is a mistake word, lock the song all except the wrong bit, and re-generate.... the new clip will sound the same, except for the part that was free to re-render

[sharp kraken]

udio has that

[broken lotus]

yeah but unfortunately you're stuck with udio's generator model.

[sharp kraken]

Hahaha

I couldnt believe people were hyping that

Im a full time like drive around the country in a van music producer for my whole life udio really struggles

and its really obvious - like any taste test

I still tried them out but i was onto this chirp stuff i actually enjoy making my main project tracks with suno

so much so that I UPSCALE MP3!?

hahaha im glad i can fix that now cause even with rx 11 they have a lot of garbage on the low end and from now till the fall it's F1's everywhere i go

Funktions - the speakers I know, I KNOW how to cheat it, but im glad i dont have to

[lone oak]

It’s hard for me to have high expectations for suno’s sound quality

🤣

Sometimes the distortion of Suno is quite similar, much like the tapes I listened to tens of thousands of times 30 years ago.

[sharp kraken]

they still need alot of harmonics added in. I'd love to get dynamic multi-layering and spectral manipulation as a PERMANENt prompt

give ya 15$ rn 🙂

[dire island]

are we still able to generate music with chirp??

[lone oak]

So now suno is still a scratch card game, just treat it as a game

[final hazel]

No, chirp is retired. You can now make music on https://suno.com

[sharp kraken]

God udio has more problems than Distortion

i have a really nice extra bedroom chirp can come live over heree

[dire island]

do you know the prices for suno?

[lone oak]

That's a really outrageous way to play.

I’ll try it after V4 is released. The bugs in the current version no longer allow me to recharge it, especially the failure of the language model and prelude. It’s really frustrating.

[idle light]

Q: Is it possible to have a song generated without drums when typing for example "drums-free" in the music style? It seems that whenever i tried "drums-free" with "instrumental" and "folk" it still generated music with drums.

[primal juniper]

Sound quality made me to drop my sub

[valid musk]

No matter what I do, I cannot get the song to have an intro that doesn't start with instant lyrics lol. Just wasting credits here

[Instrumental Intro: Piano] does this not work or something?

even tried [Lead In - piano] but its just instantly starting with lyrics

[idle light]

try

[Piano intro]

or

`[Piano intro]

[Piano]`

for intro i use

`[Instrumental intro]

[Instrumental]`

[valid musk]

I tried some of them already. But still instant lyrics. I will now give the [Piano intro] a try

ok this is frustrating me so much lol. Its just ignoring it completely and starting with lyrics

[Verse 1]
荒れ果てた道を進む
彼の目には決意の光
試練が彼を待ち受ける
彼の心は折れない

[Chorus]
暗闇の中で彼は輝く
燃える魂を持って
困難を乗り越えて
真の英雄となる

That is the start. Yet its always starting with someone breathing in and singing instantly

[idle light]

maybe try adding [Piano] in between or swith piano with instrumental in both brackets

`[Piano intro]

[Piano]

[Verse 1]
荒れ果てた道を進む
彼の目には決意の光
試練が彼を待ち受ける
彼の心は折れない`

unfortunately its like a lottery sometimes

[valid musk]

What do you mean instrumental in both brackets? like [ ] or ()?

Ive used them both in different variations now and it always ignores them and goes straight to lyrics lol

@idle light Take this for example, this is what it does with every single generation; https://suno.com/song/0251f590-cbbe-4454-b69b-acf01b08c2ca

[idle light]

like

`[Instrumental intro]

[Instrumental]

[Verse 1]
荒れ果てた道を進む
彼の目には決意の光
試練が彼を待ち受ける
彼の心は折れない`

[idle light]

aah i see

[valid musk]

200 credits of the same thing xD its mad

[idle light]

oof

[valid musk]

Maybe its a bug

[idle light]

or try shorten the input to

`[Piano intro]

[Piano]

[Piano]

[Verse 1]
荒れ果てた道を進む
彼の目には決意の光
試練が彼を待ち受ける
彼の心は折れない

[Chorus]
暗闇の中で彼は輝く
燃える魂を持って
困難を乗り越えて
真の英雄となる

[Piano]`

and extend from there, idk

[valid musk]

ahh it has finally got it right. Finally... https://suno.com/song/f45aeeaa-da9f-4060-9ed9-5b4e4c89bc15

220th credit for the win lmao

[charred merlin]

Ahh, I've tried to get only drums in the same manner, and it seems that trying to be that specific with instruments can be a slot machine-- maybe finding parts that start off the way you want and experimenting with different extension points could be effective, but I'm not a style-prompt master by any means.

[foggy jasper]

having a problem trying to attain classic dubstep i.e. Digital Mystiks, Benga, Caspa

hell even Liquid Stranger

it keeps making modern brostep or riddim

https://youtu.be/8LGRyiLEznE

[valid musk]

ok v3.5 has huge issues making a long intro, its completely ignoring it every generation and then after about 200 credits it seems to create one

[graceful condor]

It has huge issues understanding the prompt

[sharp kraken]

Good Morning from the Western Reaches: Heres my fresh morning vibehttps://suno.com/song/84e412d6-f3b3-4766-8f69-eb746c053bd7

[alpine venture]

https://suno.com/song/2e296017-107e-4395-a7b7-73386ac4476f

for those wondering about 4 minute gens

this is a direct gen, no edits

it's just needs an [end] slapped on

[valid musk]

Well, I think I'm done wasting credits on this lol. v3.5 is completely useless at listening to instrumental commands for intros.

[surreal widget]

This will be so amazing. There are many times were there are only a couple of 2–6 second spots where I want to change.

[surreal widget]

Yeah. I'm really struggling to get good intros. It would be nice if we can also get it to add extensions PRIOR to a song.

[manic saffron]

So I have the pro version how do I make sure that my song is copyright protected is it automaticly or do I need to go somewhere and copyright it?

[lone oak]

Instrumental cues are less effective in lyrics than adding "slow" after the musical style.

[sharp kraken]

MARCHING BAND! WOW

i never used that before its amazing

I dont think its a genre

[lone oak]

Suno itself has no way to guarantee that it owns the copyright, so users have no way to obtain the copyright. What we can currently know is that the copyright of the lyrics exists. In addition, according to Suno's explanation, paying users have the right to use it.

[sharp kraken]

its more like live (with horns)

Or you can just remix it

[lone oak]

Symphonic rock？

[sharp kraken]

Abysmal Orchestral Jazz, https://suno.com/song/b330dab8-4289-41c9-97ff-fbf6199bbb89

Marching Band can be added on top of the genre

[lone oak]

AI excels at weird stitches

🤣

[sharp kraken]

Abysmal Orchestral Jazz, Glitch-hop Cross-Genre, Fusion, Marching Band

Abysmal Jazz is its own thing in suno

[lone oak]

https://suno.com/song/b26f13fd-69d0-4b70-a575-2757d9dea505

[sharp kraken]

That BPM tag works like that?

[lone oak]

These labels are correct

I think

[sharp kraken]

sweet

i play like 180-250

psycore hitek

Im a friend of Gil

[lone oak]

Every time I use the J POP or J Rock tags, suno always sings in a Japanese accent. suno seems to have confused the style and language.

[sharp kraken]

Its really nice track, very polished.

I spent 5000 sunos trying to find psycore

there is no psycore

and hitek gives you techno

[lone oak]

Dark music style + deep broken voice may be possible

[sharp kraken]

the best way to get it is "Master of Puppets Festival,

[lone oak]

I think it’s not easy for suno to distinguish between rock genres, or rock that just explodes to a climax with no foreshadowing. The core problem should be the lyrics, because I find that some lyrics can easily be matched to a suitable melody. For some lyrics, it is difficult to get a good melody even with 2,000 points.

[sharp kraken]

yes its true

i always style in vocals and just make it instrumental - it usually puts vocale

then start on the first word and you can take the voice

start on the 3rd word and you get the rhythm

start before any word - all new

[lone oak]

Therefore, the tags in Style of Music should be as few as possible, otherwise even if it takes effect, you will not know which tag actually took effect.

🤣

Last month, a user only generated one dialect no matter what language tag he filled in. However, I tried to generate dialect songs but never succeeded. This is really funny.

[broken lotus]

if you want ACTUAL legit copyright that stands up in (US) court, there is a link to (US) Copyright Office submission guidelines for works that used generative AI... pinned to this channel.
(spoiler: they charge a rather high fee)

As we said earlier, you probably want a distributor (DistroKid, Ditto, Soundrop, etc) who can fingerprint the song and provide the fingerprint to companies like YouTube who will block the upload.

[sharp kraken]

true but not always

The right combination can make weird stuff

sometimes you need the whole thing

Like i cant get an orchestra to play live with modulars unless I use EVERYTHING

[jolly tundra]

[intro][male voice][No performance]Will you sing a cappella first and then go into the main song?

[rocky star]

I'm having difficulty getting an instrumental intro in before the lyrics kicks in, its strange because I didn't have problems with gens before the update? I'm wondering if others are having this issue with v3?

[vestal rapids]

when you find a solution let me know

[rocky star]

ok so its a known problem, thanks. I wont generate more then

[proper storm]

Having a bit of an issue with extends. As part winds out and i select extend it wants to place it at the end of part one which causes a "dead spot" in the song. If I back it up to insert it earlier it chops the music and starts the new extend. Am I missing a meta tag somewhere in the lyrics that can prevent that?

[warm sparrow]

Do something for vocals and electric guitar. After spent a lot of Time, Suno is under Udio for theses 2 things

[low remnant]

Have you tried generating a first segment of the song with the instrumental toggle? Then pick a good point in that clip and continue from that point with instrumental turned off and your lyrics there? That's how I usually do it.

[rocky star]

No I haven't tried that because usually there is some instrumental for at least 5-10 seconds before the lyrics begin. Right now all my gens are having lyrics at the first 2-3 seconds, so I don't know how to fix this, I don't think you can join a track at the beginning.

[low remnant]

You do it the other order. Generate a song in the genre you want, but mark it instrumental. Find one that sounds good. Then hit "extend", choose a point at like 10 seconds to continue the segment from after the hook is established, and put your lyrics in from that point.

[rocky star]

hmm.. here's an example,

https://suno.com/song/6d35ba08-0b1f-4692-8fc3-3f61bc585054

I suppose I could create an instrumental in the extend and try to loop it to the beginning, I have no experience with music editing

[neat field]

adding math anthem instrumental experiments on v3 free credits today

experiment 1
https://suno.com/song/d01a8f40-d083-4e7a-8178-fa120a07a305
experiment 2
https://suno.com/song/d6ece9ce-4b1c-40fa-a30d-4b2ec45e0810

seems to have limited repetition well enough

[low remnant]

Ok so by the time you have the song, like you do there isn't anything you can do. But what I do is I generate these instrumental intros and keep them in a folder called intros. Here is one. https://suno.com/song/7792346a-70eb-4224-87d0-19a55367070d

When I make a new song, I continue one of the intros from a nice point and put words in. You can't track a segment on before an existing one in Suno, yet. But you can plan around how it works.

[rocky star]

Yeah, I get the workflow, I will do that in the future, shame that was one of those that i liked lol

Thanks for the suggestion

[low remnant]

LOL story of my life

Good luck

[weary wharf]

I mark them as "don't delete" 😄

[vestal rapids]

how do you create folders in Suno? I must be daft!

[prisma wraith]

Library -> Playlists -> Create Playlist

[vestal rapids]

oh ok, but not actual folders in Library

[clever mirage]

why i cant create music ? - when i want to create

[vestal rapids]

do you have credits?

[prisma wraith]

No, not actual folders. but you can make 'em public, so you can get to them through your profile page

[clever mirage]

nah

how to get >:>

[vestal rapids]

ok thanks 🙂

[clever mirage]

credits

[vestal rapids]

subscribe to a plan

[clever mirage]

oh ok

[sharp kraken]

anybody know how to use Double-time?

[low remnant]

Clear at this point but I misspoke. I'd love honest to God folders but it's a playlist

[vestal rapids]

thanks, I thought I was going senile, well I probably am but you know 🙂

[weary wharf]

We mostly use playlists.

[vestal rapids]

I'd love real folders

[weary wharf]

I'd love a "workspace per project" instead of seeing all songs in as a list.

[prisma wraith]

And if you want that old-school cool kinda feel, you could burn your tracks onto cd's and put them in one of them massive CD folders thingamabobs