📚┃suno-school

[alpine venture]

(example how theoretically on could inject inappropriate content ie something something redteam)

[oblique grove]

anyone know how to determine key or bpm?

[hollow hare]

I haven't worked with the uploaded audio yet but I assume that changing the 'extend from' time can help change the tune of the music.

I also worked with several songs with various genres fused. For these, I normally use the following as the style:
"X Y Z fusion " or "Fusion of X melody and Y rhythm and Z beats", they usually work out well.

Do not forget to use dashes when using compounded words like alternative-rock. See JFly's note about generic sounding genres in 3.5: #📚┃suno-school message

So maybe adding something like Gothic alternative-rock fusion or even without the fusion keyword would help get what you want (though I woukd add a few more keywords for moods and tempo as I briefly talked about at the bottom of #💬┃general-chat message )

[sudden pike]

I hope that one day we can copy and trim parts of a song and attach those parts to another song.

[hollow hare]

As in determine what was used in your song or try to get Suno to use a particular tempo?

[final hazel]

you cannot... (yet)

Only way is uploading a part that has a chords and bpm set

[oblique grove]

Both i guess, more or less if i was to try to add to it from my own midi.

[sudden pike]

[Tempo: 130-140 BPM]

[oblique grove]

Thank you all

[sudden pike]

https://youtu.be/gZzAeYWXFpk?si=oRMxGxlKwvWSdJzM

128 is the magic number
uses 140bpm track

[hollow hare]

As 3Daizy said, using an uploaded audio would be the best way for Suno to use a particular tempo. Usually <#> BPM do not work (though I have not tried that tag uset182 showed).

I had seen the use of up-/mid-/down-tempo and/or the Italian tempo keywords (e.g. Presto for 168~200 BPM, see #📚┃suno-school message ) influence the music... but it could just be a placebo effect as well.

Genres also influences the speed of the song like using fastcore (or w/e was the crazy speed song style Mr🌳 uses) sets a fast BPM.

As for analyzing a song BPM, you can use music analyzing AIs like https://sonoteller.ai or you can use DAWs that support such function like Beat Finder in Audacity

[final hazel]

https://cdn.discordapp.com/attachments/757388174854193202/1247140445281058847/linasanad-lina_10.gif

[naive quail]

You can also use this site to tap out the beat: https://www.beatsperminuteonline.com/

[abstract spade]

How to prevent a voice in music from breaking when singing high notes?

[sudden pike]

if its doesnt work ask chatgpt to convert it into unicode code

i created this agent https://chatgpt.com/g/g-UvWkJGit0-suno-commander-electronic-house sometimes help

[final hazel]

https://tenor.com/view/critical-role-dice-gif-15465652

[naive quail]

hahah. At first I thought she was casting a spell.

[hollow hare]

@oblique grove Forgot to reply to you in my post above #📚┃suno-school message

[oblique grove]

thank you!

[lilac garden]

Is there a trick to overlap the main singer voice over the chorus?

[full lava]

can the autoscroll for when a song is finished, when they are able to be extended, and when they are deleted be removed please? Its so annoying

[midnight swan]

Any way to edit the lyrics that are written after the song was generated? The lyrics written don’t match up with the song anymore after I’ve fixed some things with extend function

[oblique falcon]

I'm not really sure what to type to get good ska music.. I was using v3 and typed "fast ska" and I got a bunch of useful catchy songs.. however the more and moe I use it the less and less it geenrates anything I like

[primal socket]

Hi, everyone:

1. where is the technical white paper of Suno? Can anyone share a link ?
2. How can I see the terms and conditions of my premium subscription again?

[old edge]

is there a way to make a song with the old lyrics video, where the text isnt moving?

[polar thicket]

Hello, I have a question about the creation of songs here

Everytime I try to create a song, the AI always stretches it out to 4:00, even if the song has already ended

How to stop this from constantly happening?

[unique stone]

Extend the song, set "extend from" to be a time near where you want it to end. This will create a new part which should act like an ending.

In lyrics box, put:
[outro]
[end]

You may need to switch to v3 model as well, as this seems to follow this instruction better.

Afterwards, use "Get Whole Song" from the menu on the final part to make them join together.

[polar thicket]

Ok, thank you so much, I'll try it out

[still grove]

When I use “get whole song” from an extended uploaded clip, it has a big gap of silence between the original and the extended AI version. Any way to blend these better??

[rigid dew]

/chirp-beta

[naive quail]

try changing the start time of the extension

I'd go back a few seconds from the point where you want the extension to come in

[untold briar]

That probably means the silence was at the end up your uploaded clip, and you extended from after it. In the future you could avoid this by backup on 1 or 2 seconds on the Extend Time, but there's nothing you can do at the moment on Suno website to fix the existing clip.

BTW, the existence of something like 0.5s of silence after an Upload, or lack of it, is a pretty big difference in how Suno chooses to Extend uploads - this will change the outputs youget. And while generally extending from non silence is a better default, may be times you prefer the extensions that use silence.

[devout meteor]

I need help with making a beat drop on a rhyme making a drill song with lyrics is anyone here able to help??

[slow rampart]

Hello teachers! I would like to ask how to improve the consistency of the melody in Extend songs.

[still grove]

Thanks so much. I can re upload without silence

[untold briar]

You can also Extend From the already upload clip, but choose an earlier time, that may be just as good.

[still grove]

Worked perfectly thanks

[untold briar]

tiny bridge clip example at time 2:06, 2:07, 2:12 used to shift the music into 'live concert with crowd' mode, instead of changing genres.
https://suno.com/song/607b0628-c6f6-4c06-816a-f2072f05b7df

I was hoping the song would slowly become more like a concert but it wasn't getting there by 2 min so I forced it. Also not a clean example (ideally the crowd clip wouldn't have a random vocal sample) but still a good demo of the workflow of using 1 or 2 second clips to force a change. (Audio Upload is from Suno's Rebecca.)

Also example of keep-a-living the original Upload Clip, so the song can end by calling back to the original "do do do" Upload even at time 4:00. Unfortunately this is very fiddly in current Suno. Note that first 2 minutes of the song is a blizzard of small clips, but the last two minutes are 1 clip.

It has to be done in 1 clip like that after 2:12, so Suno remembers the original upload at start of clip (2 minutes before), and at time 4:00. The only way to use more than 1 clip in the last minutes and accomplish the same thing would be to call back to the original humming earlier to refresh the memory (at time 3:00 maybe).

[wicked knoll]

If we move from free to pro - and then search our library a few months later, how can we tell which songs were made with PRO (so can use commercially), and which free? Do we need to label them differently or is there some other way to differentiate?

...and if you extend a song on free subscription, whilst in Pro - is the result a PRO song that can be used commercially?

[charred merlin]

I was asking this over a week ago and didn’t get a conclusive answer, if anything it raised more questions.

I’d be really grateful if you or a member of the team could @ me if we ever get an answer one way or the other?

[wicked knoll]

I'd imagine there is some sort of #tag on a Pro file... and if the song is extended whilst in PRO then you'd think that it then becomes a PRO song too... but would like confirmation of this

[unique stone]

The only way I know is to look at when I received an invoice email for my subscription and then check the dates of the songs to make sure they were within the month my subscription was active.

Regarding extending pro with free: as far as I'm aware the whole thing would need to be done in pro. Similar applies if you started a free song and then extended it with pro. There's a watermark in the generated audio. It might contain information about whether it's a free or paid generation but only the Suno staff are able to read the watermark.

[wicked knoll]

Official confirmation would be really helpful on this.

[subtle python]

I tried the upload audio feature 3 times, 2 out of 3 times, it did nothing but have the sound effect at the start of the song and nothing else. How do I get it to integrate properly?

[charred merlin]

@unique stone (tagged because hamsters are adorable)

Here’s another one that might make you consider the question more carefully, if you’re anything like me;

If you put a lot of effort into a song, make it public so people can hear it, and somebody extends it to add a really quick simple solo at the end of it— do you have any “ownership” of that other users generation, are they required to leave an attribution to the fact that it’s made from your creation, or is it wholly theirs to do with as they please? (If the latter is the case, if they own the track and edit out any changes that make it distinct from your song, is it simply a “similar creation” that Suno excuses themselves from dealing with in their ToS because algorithmically created music always has the potential to create similar results?)

Forgive the unformatted wall of text, sent from mobile

[wicked knoll]

good point. Ethically that situation is wrong... but I'm not sure they'd (SUNO) would attribute the ownership to you in their TOS... but it should go to the owner of the original (stem?) imho.... and surely could be decided by some sort of original watermark (where extended is original watermark+)?

[charred merlin]

Even one “surely” or “should” introduces a remarkable amount of ambiguity.

I haven’t been able to make clear heads or tails of it, but I’m a tech enthusiast, and not a lawyer. I’m sure the ToS has changed since I did a deep dive, but from what I can tell it sort of seems like as far as their obligations are considered, they’ve promised both parties “general commercial terms” to both of their creations in my example.

Whether “general” means “in clear cut cases, and otherwise it’s amongst you to sort out yourselves through DMCA and in (small claims) court, someone smarter than me will figure out. Or offer to explain it to me if I pay them for their time and legal services 😛

[daring lava]

@charred merlin Hope I’m not being too repetitive… Seeing you so proactively helping people and trying to figure out things is a joy! Truly appreciate what you do! 🫡

[shrewd lantern]

How can I prevent AI from singing verses twice?

[daring lava]

As you draft out the lyric portion, try including [end] or [ending] where you would like the song to finish

[charred merlin]

Well heck, I’m genuinely warmed by your kind words. 🙂

Thanks for making me feel so welcome! ❤️

[shrewd lantern]

i mean AI does it in the middle of the song

[daring lava]

Hmmm… if you’d like to share the prompt and lyrics you are using, we could try to refine it

[unique stone]

I tend not to make anything I really care about public, because I wouldn't want someone else extending off it and claiming it as "theirs".

As far as the Terms of Service goes, I'm pretty sure that just covers Suno's interest in the songs. Anything generated outside of a subscription essentially isn't yours. So extending even songs started during a paid sub will have segments that belong to Suno. If you choose to make a song public, from what I remember of reading the ToS, you're basically allowing anyone to do anything with that song - I mean, it shouldn't offer the Extend option on other peoples songs if this wasn't allowed.

[shrewd lantern]

its in another language, but if it doesnt matter, I can show you in a pm

[red moth]

Hello, do you know how to make Suno use my voice to the best advantage when I upload a sound?

[unique stone]

It's not something I have explored in detail though because I keep a majority of my generated stuff private, and under a paid sub, so it's my own little (ha!) private library of samples to export and manipulate.

[charred merlin]

Hard to get it done in a single creation, but piece together a bunch of extensions & create the song exactly the way you you want.

Before the verse gets sung again, create an extension point & keep playing with different lyrics & meta tags, making the extensions instrumental to try to reintroduce the instruments as the focus over the vocals, and being willing to spend a couple credits to get a song that's "curated" 🏆

[shrewd lantern]

I understand that this cant be influenced directly, then I will follow your advice

[red moth]

I’m talking about this new feature, and I would like to use it so that Suno uses my voice by simply giving it a sample of my voice.

[solar rose]

what exactly does that do? the link doesn't work for me. GPT not accessible or not found

[clever bloom]

Can anyone here tell me how to make the AI not skip over things in the prompts I'm using?

[charred merlin]

Sounds like you have the gist of it, record your voice and extend from the clip! The usual extension tips apply, like making sure that you're not attempting to extend from absolute silence. It's such a new feature that it's not really well documented on https://www.suno.wiki/, so any experimentation and observations you make can help everyone else as well! ❤️

[red moth]

To make our voice, what I’ve noticed when we do an extension is: -1 if we do the extension before our voice has passed, it doesn’t work, -2 when we launch the extension after our recording, it reproduces the voice quite well but it boycotts the already passed text even if we add it to the extension.

Sorry if this is hard to read, I don’t speak French natively; I’m using an AI to talk to you. ^^

I speak french*

[fringe topaz]

Trying to access my library, but it just keeps spinning. Any help would be appreciated!

[unique stone]

Can you access other pages like the Subscription one for example? (Seen some reports of this in #🪲┃bugs-web )

[clever bloom]

I take it no one can answer my question?

[charred merlin]

Prompts are suggestions, not explicit instructions. I find it helpful to create songs in multiple parts using extension points if I have a specific direction I want to take the song.

Without greater context, I’m guessing that you’re providing a moderate amount of meta tags, and some have “overpowered” others while the AI is attempting to cohesive tie everything together.

[clever bloom]

Pretty much

@charred merlin

This is the I typed out:

[Shakuhachi Melody]
[Boom Bap Drums]
[Repeating Bassline]

(...)

[Shamisen Instrumental Hook]
[Record Scratches]

(...)

[Drum Breakdown]
[Shakuhachi improvisation]
[Shamisen Harmony]

[hollow hare]

I wrote up a workaround here: #💬┃general-chat message

[clever bloom]

For some reason it just outright skips the first part and goes straight to the hook when I tell to extend from another song

[charred merlin]

Then only add the first part. Don’t focus on parts after that until you find an extension that can comfortably handle the very next part that you want.

No point focusing on parts 1, 2, and 3, only to throw away however many generations because it didn’t handle it in that exact order.

Focus on the first part and while it might feel like “wasting credits” to generate a 2 minute extension while you might only use the first 10 seconds of a clip, you’re already taking a more hands-on approach that gets you more experience actually using the product and learning all of the quirks and best practices.

So when people say “I’ve wasted 2000 credits and haven’t been able to make ANYTHING good”, you can roll your eyes and laugh because you’ve amassed a collection of unique clips to use in your future songs.

Just because you only use the first 30 seconds of a clip doesn’t mean the rest of it has to be wasted either! Download the rest of the song, clip out any cool sounding parts that weren’t used in the song the clip was extended from, and use those new clips as uploaded audio to start the process all over again.

[unique stone]

Good way to approach this 👍

[weary gulch]

today i get rarely clips that using my uploaded tracks. What is wrong?

[random bobcat]

Hey @unique stone, I have a question if you don't mind. Got a subscription yesterday, was looking for a way to impact my prompt with uploaded audio. Looks like when I click upload it just adds the audio to my list of audio. Is there a way to use it to impact new tracks? Am I missing something?

[unique stone]

The uploaded audio works as if it were the start of a new song.
On your uploaded audio, press the "..." and choose "Extend". Fill in the style and lyrics (if you want words) and then press the "Extend" button.
Your uploaded audio will then be used to influence how the new part sounds.

[random bobcat]

Oh, got you. I thought extend literally just extends the length of audio. I just tried that, it didn't simply extend the track though, it kind of mixed it with the given lyrics, cutting a part of it

yeah, extend should be basically called something like "remix" 🙂

ah, I can "extend from" as well, I see. Ok, a lot to experiment with now

[unique stone]

It really varies as to what it does tbh. I've had it continue a piece of music, layer stuff on top of what I've uploaded, style the vocals based on what I uploaded, follow chord sequences etc.

[near bear]

How to generate music in suno???

[charred merlin]

https://youtu.be/lj8utBd_IUM

[bronze shore]

Guys i was supposed to have my credits renewed today but it seems it is the saem

[clever bloom]

@charred merlin Welp...even breaking out down into parts doesn't help.

I finally get the flute to work, but now it won't give me record scratching sfx which it has on multiple occasions

[glacial scarab]

it will refill at almost the exact time you purchased your sub the first time. If you are a paid user that is. Free subs refresh at a fixed time i do not know exactly

[bronze shore]

oh so literally the exact same time in terms of clock not a specific day ?

[glacial scarab]

Yea my sub refreshes in the evening hours for example

[bronze shore]

oh okay then i will wait and see

thanks !!

[charred merlin]

Three options to choose from;

1. Break each segment into smaller parts and repeat the earlier process, now refined to specific sounds or styles you want to introduce to the existing track (eg. using only dj scratch as the style prompt, using the tips on the suno.wiki, even if you only use the first couple seconds of a generation before extending from that and going back to style you used before)
2. Make a post in #🪲┃bugs-web letting them know that your lyrics / metatags / style prompts aren't being respected/recognized during song generation.
3. Conclude that it just doesn't work.

[junior sun]

shouldn't a ending last like tops 30-45 secs... lol

[blissful nexus]

@craggy grove @regal vale @crystal prairie hey there I purchased the pro membership because inside the create panel it was implied one could upload an mp3 and have suno generate audio based off said upload, but now that ive paid for it I can see that is not the case. Id like a refund please

and apologies if you guys arent the ones I should be reaching out to, if so, can you please refer me to the right direction

[hollow hare]

Contact [e-mail] for this

[hollow hare]

Out of curiosity, what exactly is not the case? You should be able to upload unless you are talking about how the extension of said file does not work the way you wanted/expected it to be?

[regal vale]

Thank you @hollow hare

[blissful nexus]

thanks!

[blissful nexus]

yeah the create tab doesnt allow me to use the uploaded file as a source for generation, like its implied in the UI

Upload Audio at the top of the create panel suggests the audio can be used for generation

[hollow hare]

From what I read by others, you click Upload Audio and use a suitable Audio clip (must be between 6–60 seconds and format as mp3 or wav [unsure if other formats are supported])

This should upload the file (so long as it does not get flagged by the content ID check) and you would treat it as a Part 1, using Extend on it, and so forth.

I know some people had some issues uploading their audio in the past but I do not know if that still persists.

Note that if the Audio clip contain vocal-like sounds, the song cannot be set as public in Suno as a way to prevent non-first party voices from being published without permission (does not stop you from sharing the link or the downloaded media file).

(Again, this is based what I have read here so some information may be incorrect).

Though I do wonder if the upload screen simply failed to load properly. This type of issue happens to me when I try Reuse Prompt before but often refreshing fixed it

[forest cliff]

Am I supposed to get 50 credits daily? Because I didn't get credits today.

[fringe tundra]

the songs wont stop when you tell it too stop

or outro

it just keeps on playing costing more credits

[unique stone]

Ah - I think I see what you're expecting to happen here. The upload button just puts an audio clip into your library, instead of creating a new song (so it doesn't generate anything immediately.)

To make use of that audio in a song, go to the Library page and press the "Extend" button next to that song. You can then type a style prompt and lyrics, and if you want to you can also adjust what time to extend from.

This will generate a "part 2" of the song which should be influenced by your audio.

Basically the first step when creating a new song is either to fill in the "Create" details, or upload an audio file. After this, extending the song works mostly the same way.

(This is a slightly expanded version of part of Paimon's message)

[thin girder]

is there a way to minimize vocals leaking into "Intrumental" songs? Since 3.5 I don't think I've had a single instrumental that doesn't have faux vocal artifacts

[slow widget]

What is the problem? Please help!

[undone quarry]

I too am having sudden connection issues

[peak whale]

Hello, I have been showing the library in my membership for 3 days. Please solve this problem. say twitter dm mail

[floral sonnet]

sometimes i put 'oooh' and 'awwww' into the lyrics so at least it doesn't make up stuff also if you write [instrumental verse] [instrumental chorus] etc.... it seems to get around it... i found that if i set it to 'instrument' i also got those artifacts... sometimes the are cool tho'...

[drowsy tulip]

Wow.. what happened to Suno's bitrate?? It sounds really bad this morning 😐 I generated like 5 tracks and it sounds like I'm listening to something on AM radio!

[drowsy tulip]

It also seems to be repeating lyrics on every song right now - I didn't have this issue a couple of days ago... This is causing a lot of wasted credits 😐

[thin girder]

I feel like I've seen this as well with the lyrics

[alpine mist]

I have a question: is it possible to add your own Voice so that Ai can sing with it?

[drowsy tulip]

closest you can get right now is to record your voice and upload. It isn't going to sing anything in your voice though. That's probably not going to be a thing for some time.

[alpine mist]

so it won't work

[astral flame]

1. i have a song that just ends. it ends at a spot where it is starting to do some really cool things...but it never gets to complete. is there anything i can do about this?

2. with suno, if i upload myself singing will it create accompanying music?

[drowsy tulip]

1. 3.5, especially anything orchestral or without lyrics at the end has a habit of just carrying on. It's hit or miss, but I get the best results by Extending the song about 10 seconds before the end, and in the lyrics just add [End]. Eventually it will generate an actual ending. You can also do this and change the version to v3 - and then get the full song as 3.5 - but sometimes this disrtorts the quality of the whole song.

2. I've never tried it, but the idea is it will create audio using the upload as a sample

[inner current]

is suno an end-to-end large model or a pipeline with multi-models arranged?

[naive quail]

Have you tried extending it?

[cold zenith]

Hello, when do you think we could expect the release of V4 ? 🙂

[astral flame]

trying now...

[shy hound]

Hi. Billing Question: Can I make someone pay for my subscription, by sharing my Stripe payment link?

[astral flame]

thank you

[naive quail]

That feature is also great for building up a song in smaller stages, giving you much more control over the results.

...but also potentially burning through all of your credits in a flash

[shy hound]

I wish I could put a note in the prompt

[astral flame]

im new to this site, have been using soundful, and i am very excited about learning tricks like that. thank you!

[astral flame]

#💡┃suggestions-web

[floral sonnet]

eh up guys... how do you delete stuff out your library?

[shrewd lantern]

I write the name of musical instruments, but instead of their sound, AI sings these words 🤣

[unique stone]

Press the "..." menu and choose "Move to Trash"

[floral sonnet]

ooooo ta !! (thanks)

[unique stone]

You can also look in the trash by pressing the trashcan icon at the top right (in case you accidentally trash something.)

[snow zodiac]

🤣 I wonder if there is a command to distinguish the actual lyrics and the note

[charred merlin]

Are you asking someone to pay for your Suno subscription?

[shy hound]

Was my plan, but I sorted this out 😉 thx

[shrewd lantern]

these should be square brackets, but they are not perceived by AI sometimes

[snow zodiac]

I see

[astral flame]

i love that you took your shot. i might start asking for random things to be paid bc you never know! hell i want a sub too

[charred merlin]

Please don’t. (to clarify, please don't normalize soliciting donations or asking for people to pay for your subscription, because other people will do it, like what happened just here)

[shy hound]

@charred merlin That person sent me money, but now because of law, I need to make him an invoice for it, and include it in my tax declaration... 😆
Just wanted to avoid that, but f. it.

[astral flame]

oh, i was not asking a question.

[unique stone]

I don't think they were asking for someone here to pay for it. Just whether it's possible to have a different person pay, to the one who will be using it.

[astral flame]

what is the thumbs down doing on a song? it removes it and thats it?

[charred merlin]

#📚┃suno-school message

[undone sleet]

is there a way to force the singer to belt a line? like instead of
my life~
its
my liiiiiiifffffeeeee~~~

[shrewd lantern]

what command should be written to make several necessary musical instruments sound together, for example djembe and maracas?

[unique stone]

It stays in your Library but disappears from the Create page. As far as I know, that's all it does.

When new models are about to be released, there's usually a period of time where paying subscribers have early-access to help refine the model. The devs use the thumbs up/down on songs made with the new model as guidance as to what sounds good/bad. This is the only time it makes a difference - otherwise it's just "I do/don't like this."

This way of gathering feedback could change in future though.

[unique stone]

Ah. I should really read to the bottom before replying 😄

[charred merlin]

I just woke up, and after a second read, I think you were right about the context of the original question, but it seems like both thepresence and I interpreted it as solicitation, though I gather you were right all along, my dearest hamster.

[astral flame]

Ah! Thank you.

[shy hound]

I've read my own question and can see how it could lead to this interpretation ^^ mb ^^

[lavish plover]

suno, are you ok? 😦

nevermind, it's loading now 😄

[unique stone]

It has a little burp now and then

[charred merlin]

Cloudflare 520 is just a minor hiccup; like calling a friend and hearing static or a fax machine-- just need to wait a second and call again, maybe they have issues with their phone line, but probably just a temporary glitch.

[alpine venture]

Alright, so it should be launched in all countries now

https://open.spotify.com/album/03ej1SP7RJ5A5WqJGtY7g9?si=1-bUIeIcRWaqHlhQFTelzQ

https://suno.com/playlist/60489b14-8275-448a-9b37-a7a5f45af7be

the suno link has all the prompter

prompts

go crazy, feel free to use them to make cool music

not the lyrics, but the style prompts

this is my least fav album, which in my experience means it might be my most successful

😄

in any case, if anyone anyone technical questions on any songs, I can always look through my history and see if I can recall

[lavish plover]

Hi, guys... How do private playlists work? 🤔 If I open a private playlist URL on incognito mode, I can see the playlist info and play the songs 😶

[naive quail]

It's a bug - report it. The more people we get reporting it, the better the chance is that we'll see a fix.

If anyone has the url, they'll be able to see it. Same goes for non-public files in public playlists.

[lavish plover]

Gotcha 🙂 will report now

[hollow hare]

Is it a bug though? I thought that was how they were designed. (More appropriately termed, these are unlisted, not private)

[naive quail]

Ahh. Yea, if that's the distinction. That part is probably not a bug then.

But unlisted files being visible in public lists seems buggy to me 😄

[lavish plover]

I'll post this in #🪲┃bugs-web 😄

Normal behavior 👍

[naive quail]

..and the toggle being titled "Public" sorta implies that, when it's not toggled, the song/playlist is set to Private. Probably falls into the suggestions category.

[charred merlin]

The "public/private" toggle is whether it's searchable within the "social media" side of Suno. @naive quail

[snow zodiac]

Hi~ How can I control the gender of the vocal? I want certain stanza be sung by male or female.

[charred merlin]

https://www.suno.wiki/faq/metatags/voice-tags/

[snow zodiac]

I see. I have already tagged it. Seems the AI didn't perceived it

[woven flume]

does anyone else have issues with the "style of music" when creating a custom song? whenever i define the style suno does not use my definition but rather chooses its own style

[hollow hare]

You will need to work at it to find out what the AI likes.

Especially for 3.5, for compounded/joined words, using dashes instead of spaces is very important as not to separate the terms.

E.g. "dark trap" would be tokenized/treated as "dark (emotion), trap (genre)" whereas "dark-trap" would be treated as "dark trap (genre)"

See Jonathan Fly's note about this #📚┃suno-school message

I mentioned about my preferred styles and weighting of the style terms in #💬┃general-chat message

[hollow hare]

Also in case you were unaware, there is https://www.suno.wiki/faq/style-and-lyrics/styles-and-genres/

[woven flume]

i tried that 10 times now and i am still getting the wrong style of music. i define the style as "HipHop, Rap, male-vocals, slow" and the AI creates a pop song. this is very odd

[hollow hare]

Hm, yeah, 3.5 is overtuned to the pop side though rap should have worked. It likely does not understand hiphop but would understand hip-hop

Some people resorted to making an intro piece on 3.0 then extended it into 3.5. Especially for genres that 3.5 forgot like phonk

Are you trying to make a fusion between hip-hop and rap? Wanted to know because I am going to test one generation on a similar set of prompts

[woven flume]

a fusion is not the main goal. first i want to figure out how to get suno to create something in the hiphop style. right now it seems to be always some sort of pop rock.

[mystic flume]

Anyone had any luck taking a separated stem (from a Suno wav) and uploading it as a sound to Suno and getting an "upscaled" version of, like, a bass track or something?

[hollow hare]

Here are my generation with similar styles:

• A more hip-hop ver./track
• The slower rap ver./track

Hopefully it helps you learn something new

[mystic flume]

Currently I'm using Lala.ai to separate the vocals, drums, guitar, and bass, but even though it sounds the best—by a small margin—of the options I've tried (including Replay, UVR, and RVC) I still get gross phasing issues when I try to remaster in Audition using only the resulting stems. Haven't found a way to get the end result sounding better than just mastering the original WAV.

[charred merlin]

It might be tricky to explicitly get an upscaled version in isolation without other instruments included. Who knows though, maybe someday? 😉

I've been meaning to try the workflow you're suggesting here, but I've been caught up with other stuff.

[mystic flume]

Guess I'll try finding the best 60 seconds of a bass stem and uploading it.

[rough girder]

Does anyone know why uploading audio failed and saying "Upload stalled for 30 seconds, aborting"

[somber fern]

if im doing a hard rock song, how would i make the rap section rap faster? this

[Rap]

[Distinct rapper voice, fierce and rapid]

seems to kinda work but not. anyway to speed up the rapping a bit faster?

[mystic flume]

So it still adds other stuff even if you like upload a sound and then specify "funky bass riff" or whatever?

I guess I can just try it myself and see.

[mystic flume]

The internet is a series of tubes

[daring drum]

musicgen and music spleeter probably can do that 🤞

[mystic flume]

are those other websites?

[daring drum]

both huggingface models and spaces, one moment

[rough girder]

hmm Sorry I dont get it

[mystic flume]

You likely had an issue with your internet connection during the upload.

[charred merlin]

https://en.wikipedia.org/wiki/Series_of_tubes

If I had to guess, likely a funny oversimplified way of saying "the internet is silly and doesn't work as we expect always."

[mystic flume]

https://rollcall.com/2018/02/16/flashback-friday-a-series-of-tubes/

[rough girder]

LOLOL I see, thanks

[forest cliff]

try something like [96 bpm] (beats per minute)

[daring drum]

split the music in vocals and instruments with that,
https://huggingface.co/spaces/derina/MusicSpleeter
then hopefully try remixing it with that
https://huggingface.co/spaces/artificialguybr/musicgen-songstarter-demo

both are free spaces for demonstration so it's not ... professional software quality but it's nice

[somber fern]

so like this?

[Rap]
[96 bpm]
[Distinct rapper voice, fierce and rapid]

[forest cliff]

yeah, or you might need a higher number depending on what you need

[unique stone]

I'm fairly certain that Suno does not understand tempo expressed in this way

[mystic flume]

try [Suck it in, suck it in, suck it in if you're Rin Tin Tin or Anne Boleyn, make a desperate move or else you'll win and then begin to see what you're doing to me, this MTV is not for free]

And then the hook brings you back.

[forest cliff]

I experimented with bpm tags the other day while making some klezmer and it works somewhat nicely

just make sure the successive tempo markings aren't multiples of each other

[charred merlin]

I had a similar idea, but hadn't taken the initiative to try it out yet-- I'm basing my guess on how tricky it can be to specify a specific instrument unless that instrument by itself is a popular enough within the training data to have a chance of "being diffused".

Even isolated drum solos often have some kind of background, though admittedly, I haven't played around with Upload as much as I've been meaning to, so any observations would be appreciated.

[forest cliff]

(e.g. going from 30 to 60 doesn't work)

[charred merlin]

@everybody, I like the cut of this one's jib.

[daring drum]

did turn a piano into a guitar the other day 😄 so the concept works, now to see if it can actually produce entire songs

[mystic flume]

Yeah, it really seems seed-dependent, whether you'll get a drum solo when you ask for one. I wish we could re-use the seeds.

[charred merlin]

⏳

[daring drum]

death to chronos !!

😄

[mystic flume]

lol this is my greatest hope

[charred merlin]

And this is how we became friends.

[mystic flume]

Do you think seed exposure is actually coming, or just hoping?

[daring drum]

😄

[mystic flume]

ugh and I just noticed I typed "to" instead of "by" there

[grizzled slate]

I've been working with the voices generated. I wanted to mix and master the song. However, one of the main problems I've found is that of words being mispronounced, or artefacts interference. Is there any way to improve the quality and clarity of the words produced?

[forest cliff]

sometimes you can trick it into saying what you want by typing other words

like I got it to say Balin by typing Ballin

[hollow hare]

^ Pronunciation can be a difficult problem for AI, especially when you have situations like with homonyms (Live vs Live).

In these cases, it is best to spell the word phonetically, e.g. "Liiv or Liv" vs. "lyv, lyv, lie-v, lyev", etc.
(... well, you get the idea, you will likely need to spend one generation tweaking it)

For clearer vocals, usually "clear [male/female/blank] vocals" works out well

[charred merlin]

"I eight a person" works, while "I ate a person" doesn't.

It more dislikes specific words than concepts.

[forest cliff]

For some reason, Mjolnir and Mjölnir come out differently. The first one says "mih-dzhohl-nir", whereas the second is "myohl-nir"

[charred merlin]

I think anyone unfamiliar with Norse mythology or the Marvel franchise would make a similar mistake-- how would you spell it out if you wanted your extremely Anglicized grandmother to pronounce it?

[mystic flume]

Currently the best way I've found is to do a soft "reset" of the song by expanding from a nice spot after the verse in question, and using the same verse again to see if it produces something you can cannibalize in Audiition to paste into the original, or just swap it out entirely. It's worked well a bunch of times, but you can waste a lot of credits this way.

[forest cliff]

m'yollner

[mystic flume]

I had to try like 20 times to get "Greta Gerwig" to come out with the hard G on her last name. Suno really, really wanted to say Greta Jerwig. Eventually I found girrwig worked.

[forest cliff]

"gu" would be safer

as "ge" and "gi" are usually soft

[gloomy anchor]

Has anyone found a newbie friendly way of cleaning up the tracks produced with Suno? To reduce the robotic sound/artifacts and make it clearer?

[mystic flume]

I tried gerr, gurr, guure, and a bunch of others. I'm sure the seeds had something to do with it as well.

[mystic flume]

How newbie?

[gloomy anchor]

I don't mind some learning curve, I've played with Adobe Audition, but nothing crazy heh

[charred merlin]

Hot damn-- I know it wasn't the intended result, but the funky bass riff track is a really cool adaptation of the upload. Not sure if the original track sounded similar to this, or if it's a sufficient reimagining? If it's the latter, that would be a really neat way to take an existing track from an otherwise undesirable track and save it.

[mystic flume]

Ah, so... nothing is foolproof. If you have disposable income, I really like Landr's mastering process; it tends to sound better than what I try to accomplish myself in Audition. But you can use Audition to clean up the artifacts if you're willing to put in some time. Lemme see if I can show you a quick example.

[charred merlin]

There are some programs that auto-master your tracks for you. I've heard others mention Landr like Tinda just mentioned.

[gloomy anchor]

I'll take a look at Landr, thank you!

[mystic flume]

yeah, it's nothing at all like the original, which is a loud glam-rock anthem. But it does slap.

[devout meteor]

Is there a way to make the beat drop on exactly the lyric i want it to? Using [Break] , [Bass Drop], [Beat Drop] after and before doesn't work

[mystic flume]

It wasn't cheap, but it wasn't super expensive either. Plus they do this distribution thing where I can get my finished tracks uploaded to spotify etc as part of the subscription fee. I think it was like $200/yr maybe? I'll have to check. EDIT: Yeah, $200 USD approx.

[mystic flume]

Reroll, reroll, reroll 🙂

[charred merlin]

Makes me wonder if downloading a resulting track & isolating the bass might clean up some of those strange phased sections? I know the extension doesn't always adhere exactly to the uploaded track, but it could be neat to explore that as Suno improves!

[devout meteor]

basically gamble until it happens lol. Thanks regardless, hopefully v4 will improve upon my issue

[charred merlin]

I use Distrokid which is around $30cad/year, but a lot of the functions are extra charges, but I can upload as many submissions as I want and it's 100% royalties. I'm afraid if I try the automastering processes I'll be bumped up into a new audio bracket and not be able to enjoy my tracks in ignorant bliss.

[mystic flume]

So, like, I used lala.ai to separate the bass track and then in Audition I was able to isolate the artifacts and remove them like so:

[gloomy anchor]

I'm loving what Landr did with it... darn it!

[charred merlin]

Does that workflow address the warbling / phasing you get from stem separation?

[mystic flume]

Mine? No... it's solely to remove the high-pitch artifacts.

[charred merlin]

I got overzealously hopeful for a second 😛

[mystic flume]

I haven't found a reliable way to fix the phasing because so much information is removed from the "noise" when you separate out the vocals.

[charred merlin]

A couple of us have been chatting about different models in UVR, I'll make sure to send any "breakthroughs" your way should any arise. 🙂

[mystic flume]

Yeah, I've spent hours in Audition and other tools making stems, trying to produce something better, but just tossing the Suno WAV file in Audition applying the "Kill the Harshness" filter in the multiband compressor, then exporting a 24-bit WAV file takes like 2 minutes, and sending that to Landr gets me something pretty great (all things considered).

[mystic flume]

Please do! I've tried UVR but it's so resource-intensive and takes so long that I get bored of testing pretty quickly because I can't really do anything else while it's running.

[charred merlin]

Are you running it from CPU or GPU, out of curiosity?

[mystic flume]

I have a 4060Ti

I'm sure it's the settings, because I keep increasing them to see if I get a better result, but it takes too long to find out, lol.

[charred merlin]

Fair enough, I have a 4070ti (shouldn't be WILDLY faster), but I've really only scratched the surface in regards to all of the segment overlap options.

[mystic flume]

But RVC gets me nearly identical results and it's super fast. So once my paid time at lala runs out I'll probably just use that.

Yeah, the overlap... uh... stuff... it's like "only increase this if you have a good computer" and I'm like sure why not! It's a trial and error process involving a lot of crashes too, and I just get bored of it and want to make new songs instead. 🙂

[charred merlin]

Increase this for improved accuracy; " I have no idea what any of these knobs mean, but I want the best sounding result, so I'm just gonna turn a safe number of them up 50% and start there."
-Me, with every new piece of software I play with.

[drowsy tulip]

Sorry to hijack... what is RVC? My spidey senses tingled and I was drawn to this room 😄

[mystic flume]

Oh uh... let me remind myself.

[unique stone]

I go 50% then 0 and 100 😆

[agile veldt]

UVR is fairly fast on my 3070 card. 🤔

[mystic flume]

Ah, "Realtime Voice Changer" -- it has a vocal stem separation tab.

[agile veldt]

"fast" considering what it does at least

[drowsy tulip]

Ahhhh this is relevent to my interests 😄

[mystic flume]

Yeah. I'm sure it can be better with more precise tools like UVR but who has the time?

This is the one I downloaded: https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI

[agile veldt]

yeah it's cool

you can use UVC to separate vocals, and then put them into RVC to change the vocal into a different person

the future is here. 🙂

(i haven't tried RVC's own vocal separation feature yet)

[mystic flume]

It's as good as the results I was getting with UVR, and faster, but it doesn't create as many stems.

The voice changer works well, but it doesn't improve the quality of the vocal track very much. Lala's voice changer seems to do some stronger work in that regard, but they have only a limited number of voices. Lady gaga sounds great singing my songs though.

[agile veldt]

there's a big difference between the different voice models for RVC

most of the ones i've been able to find seem to have been trained on mp3 compressed material, youtube videos and what not. and trained with too few iterations. with a proper set of studio quality recordings you can make better models than most of the online ones.

[mystic flume]

define "you" 😅

[agile veldt]

i did one with 500 epochs and 72500 steps which turned out really well. appx. 10 minutes of input data. so that's the amount of training it takes for a proper result IMHO.

hehe fair

that sort of recording is hard to find for sure

[mystic flume]

proper result meaning? voice changer? or vocal stem?

[agile veldt]

voice model... and the resulting output resulting from using it

[mystic flume]

Does it have to be 10 minutes of singing? Because I could generate 10 minutes of audio using ElevenLabs. When I clean it up they sound pretty fantastic.

[agile veldt]

i used 10 minutes of talking, no singing

[mystic flume]

Just feed it a chapter of my book and get a two-fer.

[agile veldt]

and yet the resulting voice profile is capable of singing, which is pretty impressive

[mystic flume]

reeeally... have you used it with Suno wavs yet? To swap out the vocals?

[agile veldt]

Here's Suno v3alpha --> UVC --> RVC (with a voice profile i created using a girl's voice)

[neat field]

theory:

1. one randomly stumbles into a "Golden Seed" suno song with a superior vocal track
2. extend the vocals to 10 minutes with Suno
3. stem separation 3rd party tool
4. teach RVC the golden seed voice
5. use RVC to produce new audio inputs for Suno audio upload feature to always have a consistent golden voice in the same style

[agile veldt]

here's the original separated out using UVC:

i wouldn't train a model on suno's currently horribly fuzzy sounding vocals. you need a proper studio quality recording for the best result.

🍌 "Bana". I can't spell.

[neat field]

audacity has great noise reduction tools that are pretty quick these days for almost studio quality, or close enough

[drowsy tulip]

That lalal.ai is pretty good!

[charred merlin]

Similarly to being able to recognize text written by ChatGPT, I love being able to recognize lyrics written by Suno.

"Saxophones cry under the moon's soft glow, trumpets blare as we dance slow, in this jazz carousel...", something tells me "Slow trumpet, saxophone, jazz" were lyrical metatags or style prompts 😛

[agile veldt]

AI texts are even more recognizable than AI voices 😄

[neat field]

https://arxiv.org/abs/2406.05587 give it a few years or a few months of creativity fine tuning - the problem is quantified

[charred merlin]

#1252267444773326848 You absolutely rock, @final hazel

[mystic flume]

Yeah! And it's not as expensive as it seems at first. That 500 minutes is a LOT for only $20.

[drowsy tulip]

I might have some fun with this... if only to remove that distortion... (or I could just sing myself.. use Melodyne to autotune me.. noone would know... 😄 😄

[mystic flume]

please keep me up to date on your progress.

[neat field]

this will be experiments for after June credits are used up (best guess may even be August, and i will make a note to remember you regardless)

[mystic flume]

Bear in mind though that with the hours I've spent trying to remaster Suno songs by extracting stems, I never get a better sound result than just sending the original wav to Landr.

lemme see if I can find an example.

[drowsy tulip]

mm never used Landr

[agile veldt]

mastering suno's output is tricky

it's already very compressed and over-effected, and you cannot easily remove things from audio

[mystic flume]

It's not free but it's pretty great. You can accomplish something similar if you are good with Audition, though.

[agile veldt]

Tip: Try sending something into the RVC voice replacer that is not a human voice, and you get the most hilarious results.

[mystic flume]

I recall seeing a video where someone put a Minecraft villager voice in and used it to make a song and it was absolutely hilarious.

[charred merlin]

This kind of stuff is why Roko's Basilisk comes back to kill us all. 😛

[molten geyser]

Whenever I try extending my song, no matter the part of the song- it always starts with the 1st verse. Does anyone know how can I avoid that?

P.S: I extend the song from a hook that transitions into a chorus. Instead, the singer sings Verse 1.

[mystic flume]

Without seeing your structure it's hard to know. Got a public example?

[glacial scarab]

It might be more promising to just upload the splitted voice sample into Suno itself an extend from that.. as getting 10+ minutes of cleanly splittable vocals will be quite hard..
We could get away with less data but then it will not sound as great
For example I split this track
https://suno.com/song/2e7a300f-fd6c-48bb-b861-7a0ec0f78e55
And trained an RVC Voice
and relpaced the voice in this track:
https://suno.com/song/e21944ce-589b-4d2a-9097-4e4ea274a262
"rvc_input" sample of quality of the splitted vocal stem
"rvc_output" the result

[unique stone]

I consider UVR to be quite fast even on my trust ol' 1080

[mystic flume]

Generally, I find numbering every verse useful. Chorus too if you wanna change wording at some point. 3.5 is a lot better at respecting this. So, like [Verse 1] [Verse 2] etc. Or letters.

[mystic flume]

Even when you increase the settings?

[unique stone]

Ah, tbh I did not try changing anything except choosing different models yet.

[mystic flume]

I did a lot of testing but once you start fiddling with the (let's be honest, too many) options it bogged my 4060Ti down pretty fast.

[neat field]

neat, got them both playing mostly sync'd now in browser tabs your method works - should put this in new #1252267444773326848 🙂

it's got potential with some workflows

[grizzled slate]

Good idea! I'll give it a try. 👍

[molten geyser]

I think I managed to deal with the problem. Thank you anyway

[grizzled slate]

Now that's a clever idea...

Where would you add this; clear [male/female/blank] vocals?

[hollow hare]

In the style prompt typically, like "clear female vocals, dark-trap, ..."

I do not know how much of an effect that would have when used as a metatag

[grizzled slate]

Many Thank! I'll give it a go 👍

[mystic flume]

I get good results with making "female soprano vocalist" the first part of the style prompt, and then adding "crisp vocals" at the end if I have room.

[agile veldt]

@glacial scarab great result there! 👍

[astral flame]

if i get a paid plan can i upload vocals and suno create accompaniment?

Anyone used and liked the results of Logic Pro stem splitter with suno? or anything they recommend?

i am trying to find a way to be able to mix and master after i add my vocals to get a high quality song.

[maiden turtle]

Hello everyone, I’m a new member

With my respect can someone help me about prompt to generate song?

I really want to know how to generate song like this
https://youtu.be/EByD5F9Hkvs?si=mho6rWA_uOQyVnxZ

I try a many prompts but the results it’s not be like that.

Can someone to help me, I really want that.

Thank you very much.

I really want to know how to generate song like this
https://youtu.be/EByD5F9Hkvs?si=mho6rWA_uOQyVnxZ

I try a many prompts but the results it’s not be like that.

Can someone to help me, I really want that.

[forest cliff]

Is it possible to generate a song using the extend button and omit the first part?

[unique stone]

No - I think this would be good to be able to do though (may have been suggested in #💡┃suggestions-web previously.) Generally being able to trim the start/end of the song would be nice IMO.

[charred merlin]

What variations in prompts have you tried?

There's no magic keywords, it's really a matter of experimentation. Personally, I might generate an instrumental custom track with "meditative, ambient drone".

[unborn birch]

Hi new here. Does Suno have a definitive way to make a duet or at least generate male/female specific audio?

[hollow hare]

Old copy-paste made for v3.0, I haven't tried for 3.5 but the metatag reading is much better:

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.

[forest cliff]

can't seem to get a bassoon solo

even tried calling it fagotto like the italians do

out of credits, i'll have to try again when I have more

[unborn birch]

Thank you! I didn't want to edit in other programs. I've been alternating between v3 & v3.5 for different reasons. Using different qualifiers before and after verses and such. I'll try doing some of the alternatives.

[maiden turtle]

Thank you very much I’ll try.

[shell cairn]

Is something wrong with the bot?

[final hazel]

There is no discord bot anymore. You can now make music only on https://suno.com

[shell cairn]

Thank you, sorry I’m new so my bad

[final hazel]

This page will help you getting started: #1151542418991239168 message

[sweet wigeon]

Drummers... I've tried to create two song with drums instrumental, giving the instruction like in lyrics I am used to. But becase there is just 200 chars I had to make the instructions very short. Tell me about the song, what do you think did Suno followed the instructions? https://suno.com/playlist/fe7b8fbd-a2c7-44e0-9cc2-01c79d337ddf

Hi, I'm first time on discord.

[alpine venture]

OK, so I got some insight into "soulless music syndrome" today

I realized I haven't encountered it much bc I use a lot of freeverse

but this one I drafted up today was meant to be very very trad

https://suno.com/song/7aa03dd4-afa7-4257-9c30-2713ec1fc736

and I had issues

Basically, what got it out of its funk was removing all line breaks

I have no idea why

but it increased the pace and brought life back into the composition. It's still traditional structure (what half the people i know would unironically call x-pop), but there was something very wrong/missing in the gens before I removed those breaks

it was just completely lacking in dynamism

or emotionality

oh, and I brought ska into the mix

[neat field]

i listened to it, but it didn't resonate with me.

but then your song from yesterday came on https://suno.com/song/a96f9a08-b0f7-40e4-bde6-6270c369205a/ and the accent of this singer is the closest i've heard the model to output colonial american (weird ancient southern accent) i've heard so far (not quite actually the accent, but close enough)

[alpine venture]

aw man, it won't let me make the joke about ska being an incomplete word I ran out of room for haha, automod blocks that shitty extremely inappropriate dad joke

[alpine venture]

I love that one

the voice is so good if you have quality headphones

if you don't, it's too muddy to understand

[neat field]

i think it's the Irish Folk (with some sort of weird americanized gaelic accent) coming through somehow

added to my inspiration playlist !

[alpine venture]

but hot dog it's rich

[alpine venture]

oh no, it's not that actually

that's the v2 opener

the voice comes almost verbatim from the styles

on this one

[neat field]

v2 hidden sauce 📝

[alpine venture]

https://suno.com/song/38e2032a-5009-422d-8525-d692a1936b8d

but those African musical prompts

brought huge depth in

[neat field]

refeed that voice into v3.5 audio upload sometime, i'm here for it

[neat field]

this is a great voice, but the content is not pg-13 and i must defer it out of my lists unfortunately... very unique voice

[alpine venture]

translation is rough haha (it's about death)

and loss

although it does do this through biblical metaphor

involving onan

but yea, the main point there was the vocals

and the prompt

[lyric sonnet]

"traditional Australian Aboriginal Didgeridoo music" - Suno, why does your didgeridoo sound like a clarinet..? https://suno.com/song/5197e8b1-cc97-4f1d-bc53-97e557d12671

[neat field]

some interesting dialects here i've not sure what prompts hit these https://www.youtube.com/watch?v=03iwAY4KlIU

[sacred swift]

https://suno.com/song/44f3bacc-a15b-40ad-b26e-4d82eb4a8e3e

I think you can get away with it if you put the lyrics in [Instruments] and write the instruments firmly in the style.

[lyric sonnet]

I see, did you use 'custom mode' for this? Yes, it has a didgeridoo, but otherwise sounds Irish

[sacred swift]

yes, I guess you have to specify the tune as well as the instrument properly.

[lyric sonnet]

let me try some traditional Japanese Shinto music

i get piano and rock guitar ...

i guess they need to augment their training data

[charred merlin]

Starting with v3 or v3.5?

3 excels at creating a broad range of sounds, and is very versatile. 3.5 excels at song structure and flow

[wild shuttle]

idk if i should ask here or in general- but what kind of credits do we get for referals? are they like top-up extra credits that won't expire?

[lyric sonnet]

This was 3.5 but I'll try 3. I wonder if the word 'Japanese' strongly pushes it to anime music

[wild shuttle]

and most importantly, if i have a pro account and get credits from a friend signing up, would i still have the rights to the music, or do "free credits" mean it'd be like using a free suno account?

[lyric sonnet]

I took out 'Japanese', so here's 'traditional Shinto music' from V3, enjoy. 😆 https://suno.com/song/4f588db3-a8b5-4549-87c0-a769723eaac1

[naive quail]

Custom mode is definitely the way to go for didges. I did an extension of your Dreaming Land (I kept it unlisted) and tossed in a didge... and a tuba for grins https://suno.com/song/d66314e8-da24-4368-bb51-0814691c119c

[lyric sonnet]

hm, what does "////////" do?

[naive quail]

nothing consistent. It's just prompt noise. I sometimes toss that in there when I'm not getting what I want

[lyric sonnet]

i see, gotta try that

[charred merlin]

If you have an active paid account, you own songs generated by it.

But whether they’re standard credits or extra credits is a serious question that I’d also be curious about.

[naive quail]

It's hard not to keep playing with that track. I want it to go back to the didge 😄

[wild shuttle]

it'd suck if i get my dad to sign up, get credits, then get the month refreshed before i can use them lmfao

[outer shoal]

So I saw somebody with a song but lyrics were Morse code.... dots and dashes. Is that a real thing? I regenerated but could not reproduce. So maybe it was some recompose trick then? https://suno.com/song/f193f729-cd82-48e9-8454-39e48235e3cf

And the words did not seem to match to Morse code translated version....

[opaque orbit]

That’s a bummer

Sounds fancy though 🙂

[outer shoal]

And enough weird edgy things do work- I was all hopeful.

[naive quail]

It's really just prompt noise from what I can tell. It does have SOME effect, just nothing that seems repeatable. I add stuff like that when I'm trying to add something with an extension but Suno's just ignoring me. Sometimes that bit of noise will get Suno to fall in line.

[wind locust]

can't download as wav, it always breaks

[vital plaza]

Can someone tell me how to read this insane syntax? I prefer writing a regular notes:-) all this MORSE Code For god sake!:-) Do you have ... ---... to Notes converter? ^-))) WTF do you rerally use a MORSE code for writing notes and rhythm???

[naive quail]

I wouldn't expect it to produce anything repeatable.

[brittle apex]

Yea, it isn't an exact science. If you want control over a riff or beat, upload a sound to extend from

[outer shoal]

Just words....

[vital plaza]

Why write Morse If I Can write c# or javascript? Oh it is Just a sarcasm :-)))) I will use XML next time :-)))😋

[radiant plover]

So basically we upload the sound, extend from there to build off of the sound we want?

The goal isn’t to get the original upload to be part of the track? Like I wouldn’t “get full song” for these right?

[outer shoal]

Because it is Funny/Weird. But hey, maybe XML would work. For emphasis or something. Something to try....

[brittle apex]

Correct, the original sound might be there in the beginning but then the AI will take over and change it according to your prompt

[naive quail]

Ideally... But you won't be able to do any other extensions on the piece and combine them w/o Suno adding your sample clip to the beginning. 😦

[radiant plover]

So is there another way upload the sound as the prompt itself?

[naive quail]

Right now, we only have an option to extend the uploads

[sharp kraken]

What are the Genre Groups in the Showcase Section for? Should I submit appropriate tracks forconsideration in those channels?

[charred merlin]

Same as the previous showcase, no consideration beyond sharing your creations, but just a bit more structured instead of having everything thrown into a single chat.

[sacred arch]

I've made over 100 songs using v3.5, but I still feel the quality isn't as good as v3. Specifically, the songs created with v3.5 sound more like radio quality and aren't as clear to me. I've gone back to using v3. Has anyone else experienced this? Any tips on how to prompt v3.5 to create better songs? Thanks!

[mystic flume]

Easiest way to do that is with editing afterward in Audition or Audacity, etc.

[mystic flume]

Hard to know without hearing what you mean, but I suggest creating an intro using v3 and then extending that from a nice spot in the first 20 seconds using v3.5.

[carmine osprey]

is it possible to have uploaded songs appear as the reference from an extension, as is the case for normal suno generated songs?

[crystal owl]

Is there anyway to prevent the songs from ending abruptly?

[peak kelp]

Is there a way to edit just the lyrics of a song but keep the music and voice the same?

[mystic flume]

try [outro] and [end] or, if you like what you have and it just cuts off at the end of the track, extend it with those tags.

[mystic flume]

Extend from just before the lyric you want to change, regenerate and hope for the best. It might take a lot of tries. Then you can edit it into the previous version in an external app like Audition or Audacity.

[unreal trench]

check this out: https://www.suno.wiki/faq/metatags/song-structure/

[peak kelp]

Thanks. From a technical standpoint, what is the difficulty in just adding an edit lyrics feature? It is probably not that easy, as they would have done it already. But I am interested in the techical reason for it.

[peak kelp]

thanks, that's helpfull

[unreal trench]

most welcome - rest is learning by doing 😉

[maiden turtle]

With my respect, I really want to know how to generate song like this
https://youtu.be/EByD5F9Hkvs?si=mho6rWA_uOQyVnxZ

I try a many prompts but the results it’s not be like that.

Can someone to help me, I really want that.

Thank you very much.

[pseudo patio]

Suno is doing some really screwy stuff with the mix on intros I upload, like it sounds really bad sometimes... is there some sort of mastering pass it tries or is the AI actually messing with what you've uploaded as if it were its own generation?

If the latter I understand. The former I don't want or need and would like to be able to disable lol

[jagged cosmos]

Is Suno died for anyone else?

[pseudo patio]

ex: https://suno.com/song/3ea51786-bdf9-4995-8685-7b2c1ae19297 vs https://suno.com/song/b9064710-30a8-42a3-b8b0-6a20f286f1b8 crickets are just gone, phasing effects are screwed with and everything is muddy and muffled, etc.

[hard mango]

suno runs it through some filters for sure, not sure what exactly. there's no way to disable it, but i'd recommend doing a 20/18,000 cut on your track before upload and make sure that you don't have any harsh resonant frequencies that might throw off however suno analyzes your audio

[wise blade]

hey everyone i have a little question not about suno but about AI generated videos i was wondering what is the name of the App/Site/AI that can make those kind of AI generetated video https://www.youtube.com/watch?v=wCRftfx62uY

[mystic flume]

I'm not sure. We're colloquially referring to the concept as "inpainting" like you'd do with an image, and I've heard Suno is working on it. That's about the extent of my knowledge on the subject.

[grave wraith]

With the new upload feature...anyone tricked Suno to use the "original" melody for the chorus?

[astral ocean]

Haven't tried. Maybe start with [chorus]?

[hard mango]

https://suno.com/song/d079ed40-262e-42b7-956e-289291e4b28a

i made the melody that you hear in the chorus, starts around 30 sec

https://suno.com/song/3d3d6f85-cfae-4a0b-aa48-de1e636cc30c
here is the audio i used

[hard mango]

the trick IMO is to upload a loud, repetitive and preferably isolated melody to help ensure that suno uses it appropriately

[unique stone]

It seems like it needs to convert the audio into something the AI model can process. Not sure what it does exactly but it's most likely a necessary step in order to be able to use the music. Like turning a sample into granules for use in granular synthesis for example.

[sweet wigeon]

What Suno definitely needs is import of MIDI files. Just imagine to have your own drums. U know they are so important when you want to create the song because the drummer defines the dynamic of the song, like the tempo and time signature. So if we were able to import either the MIDI file as sample of drums, or maybe as a sample of song melody... it would be great enhancement. There is program like tuxguitar, where it is easy to generate notes for some instruments. It could serve as a pattern for AI. Second idea is that it would be nice when I have created such drums instrumental, let AI tries to mark the lyrics which I will try to add to it. If it does not fit because some words are too long, or like in Czech where we have some long vowal so the lengh if vowel is extremly important in some languages. And you cannot add some text when it does not fit to the rhythm, so AI could suggest make some edits, where it could be possible to shorten it, to change to short word, or to add an instruction to change the pattern like drums and quitar. Like to create transition between two parts of song which does not fit well togheter when your drums or the words don't fit each other. Also in Czech, when the tempo is too fast sometimes it is impossible to pronounce long verse/line of text or long word, because when you shorten the vowels then no one understands what you are talking about. That's why new language learners are so funny when they don't manage the vowels, and they sound like so short vowel speakers.

[sweet wigeon]

I cannot really answer but I think AI needs to decode the audio first which could take reasonable resources. In fact to do it well it would need to separate the instruments and analyzy every one of them. If it does not do that because of the low quality model and low performance, you cannot obtain the results as you wanted. But this situation could be easy solved if developers would allow to use MIDI files. You would need first to convert the audio .wav file to MIDI files with the notes for each instrument, including voice if it is present in your song. Then you would use this MIDI to tell the AI how to modify, extend or create your song.

[modest pelican]

Hi, how to export video without suno logo?

[unique stone]

This is a good thing to put in #💡┃suggestions-web 😉

The audio upload feature can sort of do this, but not with MIDI. You can set a tempo by hitting an object, a melody by humming, or a chord sequence by uploading an audio containing one. I've had it use drum loops I've uploaded. It'll usually try to continue music that is uploaded so you may be able to get it to add vocals to an instrumental or the other way around, but the AI will still be "creative" so there's no way to tell it exactly what you want.

MIDI upload would be useful and could be used similarly to do the things I mentioned above. But the AI does not use MIDI itself as far as I know so it'd need to convert the MIDI to audio first. This is easy - just use e.g. a SoundFont to render a WAV of the MIDI and then process the WAV.

But, it doesn't do that yet... So for now it could be done manually, by recording/exporting a MIDI performance as audio from whichever program you use, and uploading that.

[silent viper]

I believe that more useful than midi file upload would be the midi file output.
Then we could open it in a DAW and use the VSTi of our choice. And it would solve the problem of low quality of audio files.

[unique stone]

That would be useful. But it's a lot more difficult since it would need something to go through the generated audio to work out what notes are being played (pitch detection)

[naive quail]

Deleted a song last night by accident - no chance in hell of actually finding it in the trash, which seems to be randomly sorted and probably has 200+ pages of rejects by now. I still had the link to it, tho... so I just did a v3[End] Extension on it to make a new one, which I'll try not to delete. 😄

[lofty garnet]

Hey! I've got bit technical question bout chirp. As I generate song using v3.5 ( 3 verses and 2 choruses) AI makes song where it sings second verse two times and last verse too. Is there any prompt that I should use or some settings that I am not aware that I should change to sing it only once? (at least second verse one time)

[lofty garnet]

Sorry, my mistake, I am not sure if it is chirp. As I do use web version on Suno

[abstract hare]

is there any way to add ambient sounds like rain to an instrumental?

[naive quail]

the best way would be to upload a sound sample and extend from that

here's an example of thunder that starts with a sound sample https://suno.com/song/3e572c78-0c5e-4845-97cf-dde077a86465

[charred merlin]

No worries-- Chirp has been retired (they haven't gotten around to getting rid of the chirp bot, but it doesn't work anymore).

If you look at #😜┃suno-memes you might pick up on the fact that a lot of others run into a similar issue in regards to verse repetition. While I find that 3.5 is better at following the flow of a song (eg. wrapping a song up if the extension it's being asked to create is positioned right as the "response" or "denouement" begins, as opposed to at the beginning of a new "call" or somewhere that it could understandably try to insert a rising action), it's a matter of experimentation.

The best advice I have is to break the song up into smaller & smaller segments using extension points until you're able to get exactly what you're looking for-- the less specific of a song you're looking for (exact type of sound, specific phrasing and usage of lyrics) the less work you'll need to put into it, but the song will ultimately sound somewhat similar to others who've hoped for a one-shot generation.

[north jungle]

Salve
come faccio ad inserire un grido in una Strofa o fare la Strofa urlata?

[charred merlin]

https://discord.com/channels/1069381916492562582/1155316661927297115
Mi scuso, questo é un forum inglese

[north jungle]

Sorry

[charred merlin]

https://www.suno.wiki/faq/metatags/voice-tags/

[north jungle]

Thank you

[sweet wigeon]

You can have more tracks/instruments in the midi. So in the case, you would need to upload more files... Which would cost the performance. But why could not be the audio file generated from the MIDI on the side of server and then passed to AI? I am not aware what licence is used by textguitar or it's component, I suppose open source code).

[sweet wigeon]

So the problem is not licence of the software but of the audio files. They would need to buy some soundfont to convert .midi to .wav. Then why not.

[shadow valve]

Does anyone know what Suno Considers a "Listen"? Is that some one hitting the play button? Getting 5 seconds in? Is me listening to the same song on loop 4 times adding 4 to the count? If I refresh the page and hit play and refresh it again, does that add to the count? Just curious how the metrics work!

[charred merlin]

I have nothing to base this on Basing this solely on my own experience, but I feel like it's if >50% of a song is played, it counts as a play.

I don't share my songs through Suno, so all of the plays are specifically from me, and often I'll see a handful of plays on tracks that I skipped through while searching for suitable extension points.

[shadow valve]

Thanks for the input!

[charred merlin]

As I'm getting more hands-on experience with mixing and mastering, I've played around with Audacity, and while it gives you a great deal of control, it requires a certain level of technical savvy and understanding of how a lot of elements come together to create the best results. I've looked into some of the automastering processes like Landr, and Mixea from Distrokid, but I've found at a glance that they more or less just apply normalization and compression which sometimes can have undesirable effects on specific tones you were trying to create.

I've just come across BandLabs-- instead of creating a track on the site, I've uploaded my Suno track and I'm having a much easier and more intuitive experience "playing around" with the process on the site, but I've yet to figure out what "the catch" is, because there's no way that I'm getting better service for free than the companies trying to charge $99~150/year for an inferior service.

[lofty osprey]

Hi Everyone. I just starting using v3.5 and now all the songs I write seeem to automatically extend to 4 minutes even if the song should only be 3 minutes or 315, etc. To make it into 4 minutes Suno just starts repeating verses or the chorus and the song gets extended and screwed up. So my question is: Is there anyway to set the song length or make a song that

s lesss than 4 minutes in v 3.5?? thank you!!!

[naive quail]

We're all stuck with this for now. The trick that many have been using is to Extend the 04:00 track using v3 from the point in time where you want it to end (or a few seconds prior) and have only [End] in the lyrics box. You may need to do a couple of rolls and play with the exact time a little bit, but you'll probably find something that works. I've had extensions as short as 3 seconds doing that.

[sacred trench]

So, I have a really interesting question. How does Suno handle microtonal compositions?

[lofty osprey]

Thanks very much, SixString. Really helpful work around. I'll give it try 🙂

[weary heath]

Hey today i was on fire because the new 3.5. But i have recordet some vocals. If i upload and then extend the vocals dont be the same. no beat under vocals what i do wrong?

[lone mirage]

peace and greetings, I have looked thru the web app and the docs and I am trying to find out if Suno can generate isolated vocals ie acapellas. If so does it require specific instructions? I understand there is vocal seperation. I was wondering if there are api tools or scripted functions that would make it easier if it is not already available. Thanks in advance.

[charred merlin]

Oh my god, 3Daizy and I were talking about this the other day. I had almost no success using the anglicized "acapella", but she was having regular success using the traditional "a cappella".

I don't have anything conclusive as a takeaway from it, but I found it interesting that there was such a wild difference in results. If anything, I would have thought that presenting it as a single word would have given a more conclusive result, but I suppose that particular style is more closely associated with the classic spelling.

[naive quail]

I haven't played with it too much, but it always ends up with a barbershop feel when I do. 😄

[rocky star]

I've used bandlab mastering tool as well, and what i found is that the noise gets even more emphasized, i didn't try the Distrokid mixea, its a bit pricey and one track is not enough to test the quality. I am curious, what tool did you use in Audacity for mastering?

[charred merlin]

I used the single free generation from Mixea that I got with my subscription, and while it's easier than manually mixing & mastering, I still found that it "painted everything with the same brush", a similar feeling that I got from Landr.

In Audacity, I use OpenVino to separate the stems (I've found better results in successfully isolating vocal tracks using it compared to UVR so far), then use the effects suites from MeldaProductions and MuseFX.

Did you use a one-shot process in BandLabs like you'd find from Mixea & Landr, or did you open the song up in the timeline editor and play around with it?

[lone mirage]

cool I wanted to make reverb slathered vocal stabs and pads and I thought it might fit the bill. Might have to combine tools to make it work. It is weird because I can't see why the platform wouldn't allow us to isolate vocals if it has the instrumentation toggle on it. Are there other discord channels that let us approach the matter via text commands instead of using the web app?

[charred merlin]

No, just the web app-- but there's a bunch of tools you can use in tandem with the stuff created by Suno if you have a specific sound that you want to create. Now we can even feed the new creations back into Suno and get even wilder creations 🙂

[rocky star]

I used the Studio for mixing, or joining tracks and for mastering, the one shot options that they offer, from universal, fire, etc. I am not happy with the results, because the noise just gets heightened and I'm almost at the verge of uploading the original Suno tracks to Distrokid without mastering. I am new to all this, and technically challenged lol.

[charred merlin]

I LOVE how AI is lowering the educational bar of entry for so many fields! 🫂 ❤️

You're clearly not technically challenged if you're taking it upon yourself to play with all of these different tools that are available to you-- it's just that you don't have the experience under your belt, but now it's easy to get that practice from your living room, instead of having to move to a city with a good music program at school in order to get your foot in the door.

[lone mirage]

Yeah I guess I will take that route

[rocky star]

indeed so! but when you have a problem, then the human skill is needed.

[charred merlin]

@mystic flume got some cool results from using a Stem Separator to isolate a bass track from a really intense song, then feed it back into Suno in an attempt to tidy up some of the phasing & warbling that ended up on the track, and she was able to make some pretty cool variations from that bass riff. Mind you, those results sounded way different than the original song the bass was taken from, but just playing around like that will help you spot really neat tracks instead of leaving them behind because the rest of the instrumentals weren't up to your par.

Maybe you could do something similar with the vocals?

[lone mirage]

Yeah I’m gonna look around and plan things out. I’m sure it can be done…it’s just I don’t need suno for much else except vocals because I have synths and gear, etc…but I did like the idea of uploading and uprezzing our own audio

It’s just it would be great to have an ai prompt text to singing synth versus text to speech.

[charred merlin]

I don't know if it necessarily succeeding in upscaling the track's clarity, but it definitely breathed new life into it, reimagined it in certain parts.

Agreed, something like a Bark that Chirped would be neat. 🙏

[lone mirage]

Exactly! When I saw bark and chirp I was like ARE YOU KIDDING ME SIGN ME UP….and then I saw the web app and it felt restrictive, lol.

[charred merlin]

I wasn't here for Chirp, but from what I gather the web app is an evolution from it, and Bark is open-sourced so anyone can run it locally & improve on it

[lone mirage]

Oh ok ok! I’m not downing what I see so far it’s just organizing which subscription goes where and why, lol. I am thinking I can buy Synthesizer V make the “clips” and upload those with some prompts…but good looking out on Bark because I did not know it was open source, fam

[lyric sonnet]

i've tried stem separation in fl studio, which works fine, but the stems then really lack quality. some kind of uprez function would be really useful..

ideally a tool which does stem separation, then separates the reverb etc. from the clean original and fits sounds from your sound library to each of them (except voice i guess)

i'm sure someone is working on this somewhere

[simple osprey]

is it possible to let the voice whisper? like in an intro?

[sweet wigeon]

My opinion: Nothing U do wrong; AI just does not have enough performance and knowledge to analyze the audio on very high quality.

[neat field]

https://discord.com/channels/1069381916492562582/1252286293992407140

see also https://suno.com/playlist/ca526d7f-8abb-482a-b398-68c7bcca7981

[silent viper]

It would be more practical if the output was in a midi file.
You put the VSTi of your choice and that's it.

[lyric sonnet]

once all the separation and mapping to instruments is done outputting a midi file should be easy..

[weary heath]

ok mh i hope it will work in future

[foggy jasper]

Not sure what to do. I've noticed people have uploaded AI music with Landr, but I got this email. I think I am okay, they just want to make sure I have a subscription and if I am officially cleared to use these samples. I can prove the subscription, but I am not sure how to prove if I am cleared to use these samples. Am I? (I have a pro sub obviously)

[hollow hare]

By the wording, it appears they are aware you own the song so long as you have a subscription so taking a picture of your subscription time frame and the creation date of the song (by providing the URL would be a way for them to confirm) should be sufficient.

If you want an additional proof of ownership, you can look in the Terms of Service for the following

https://media.discordapp.net/attachments/1102676217687195679/1249441403956760586/Screenshot_20240609_125633.jpg?ex=66732dfc&is=6671dc7c&hm=8f2c59763097083c0bbc5b609c9e39aaa7d990b1c48e3cc910cf01a7cb163181&

[foggy jasper]

thank you so much @hollow hare. I went ahead and sent them my sub proof, but I didnt think about sending them the creation date of the song. wouldnt hurt to send that too. thank you again

[hollow hare]

You're welcome 🙂

(This is assuming you did not use samples from non-licensed sources through audio upload of course.)

[foggy jasper]

oh definetly not

this was before upload was a thing lol

yeah im hoping it passes

this is like my most favorite suno track lol. I havent had another one like it

[daring lava]

Thank you @hollow hare for contributing to members learning here! You are greatly appreciated for sure ❤️

[charred merlin]

You're a real swell fella, David.

[daring lava]

Feel the same way about you and @hollow hare! You both help so many here! Pretty big deal if you ask me. Definitely noticed

[undone basin]

It would be very nice to generate "intros" for the uploaded files ^^

[unique stone]

This would be a good one for #💡┃suggestions-web

[silent viper]

The problem is that the suno's output has low audio quality and hiss, noise and artifacts that hinder the perfect separation of voice and instruments. Making the process of converting audio to midi extremely difficult.
That's why I say it's more practical for Suno to output in midi.
MidiGuitar2, Melodyne, SuperiorDrummer3, EzBass, Fadr already do this.

[dusty stag]

Did something change with how lyrics are generated? Suddenly everything sounds robotic with v3.5 after the upload audio feature came out

[mystic flume]

Pretty much everything you can imagine is possible, but it might take a lot of generations to get Suno to interpret your commands the way you want it to.

[mystic flume]

I've noticed no change. Did you try another style prompt?

[cold zenith]

new 3.5 ?

[smoky tree]

Is it possible to generate the same song but without the vocals? If not is this something that is coming in future updates?

[charred merlin]

Not currently available, but search this channel for "stem separation", "UVR" or "OpenVino"

[errant arrow]

One message removed from a suspended account.

One message removed from a suspended account.

One message removed from a suspended account.

[smoky tree]

Add more lyrics to fill out the song more or download and edit it in audacity to end at an appropriate time. Thats what I've done.

[errant arrow]

One message removed from a suspended account.

One message removed from a suspended account.

One message removed from a suspended account.

One message removed from a suspended account.

[crystal ibex]

Is there a server pause? I cannot enter the Suno website.

[charred merlin]

There isn't a major trick to making Suno do exactly what you want, it's a matter of creating extension points after the parts that you want to keep & continuing to build on your existing song that way. Use the tips found on the suno.wiki to help better direct each section.

[crystal ibex]

Problem solved, my web browser had a fault

[errant arrow]

One message removed from a suspended account.

[charred merlin]

We've found "a cappella" works well 🙂

[errant arrow]

One message removed from a suspended account.

[shell loom]

problem here #🐣┃suno-showcase message

[shell loom]

well obliviously not going to say anything, salty users

time to consider, cancellation of this, and goto the other site

[foggy jasper]

damn never thought i’d see so many points to agree with in one post! it would be really cool if there was a midi option. I absolutely love Melodyne and SuperiorDrummer3

[rocky belfry]

Does the style mean anything when extending a track? I'm trying to find a way to add more variation, musically, as a track gets longer.

[charred merlin]

Oh yeah, you can do a complete 180 on the style of a track. I find it funny to end bluegrass songs with a chiptune solo

[shell loom]

@unique stone never mind bud, i cancelled suno. to many post here from Salty users that can't be bothered with me having a go at someone here a few days ago.

[rocky belfry]

Do I change the style box, though, or do I need to add annotations into the lyrics?

[unique stone]

Oh. I must've missed that. Sorry to hear that it's made you abandon Suno though :/

[sharp kraken]

You can do alot with changing the style of the track, i use it to change instruments, key, even genre - although it doesnt always do what you want, eventually it usually will

[heady egret]

hi, does anyone know what the max context window is?

[sharp kraken]

3,000 characters on 3.5 1500 on 3.0

[rocky belfry]

I guess I'll keep poking it with changes every time I extend and see if it's just changing very gradually or something.

Very gradual indeed because I keep running out of credit.

[sharp kraken]

yeah if you want to make a direct change I would input it into the lyrics using [] commands, such as [Breakdown] or [Beat Switch]

there is a .pdf on the suno wiki that lists a bunch of those commands

[heady egret]

Thankyou!

I should have mentioned audio context window though. Curious how far back suno can see on the song before parts are lost

[sharp kraken]

oh gee i think its 6-60 secconds

ive not actually used it yet

[charred merlin]

The audio upload portion allows you to use clips that are between 6-60 seconds like Joshuasca mentioned, but if I'm correct in my presumptions, you seem to be asking about the context/memory in a similar fashion to LLMs like ChatGPT?

[kind dock]

Huh? some bot deleted my post bcos language? I didn't curse?

[pearl valley]

Try to to scroll in your playlist on IPad 10 vertical or horizontal does not work 😡 no way to organize or even play what you want . This is extremely bad display of Web skill design.

[unique stone]

Looks like the moderation bot doesn't like the word for one of those structures that go over a body of water, which is also used to describe a certain passage of music. It's a new thing. Seems a bit sensitive. I'll mention it to the mods to see if they can tweak it.

[final hazel]

lol.. I see what the problem word was.. going to edit that

[charred merlin]

...like the golden gate thing?

[naive quail]

rowboats are banned!!?

[final hazel]

😉

[charred merlin]

@kind dock 😮 that's worse than all of the swear words combined

[unique stone]

Chat 'bout to turn into a game of charades 😅

[final hazel]

Watch out.. 3 violations will end up in 10 minutes mute 😉

[charred merlin]

At first I thought you meant the commonwealth English term for a levee, but this was way more unexpected.

[final hazel]

I have to go through the looooooong list

[naive quail]

SunoNet awakens...

[unique stone]

Tonight's bedtime reading: The dictionary!

[charred merlin]

...is brXdgX entered as a bad word, or was there some kind of overlap that caused a profanity filter to catch it?

[unique stone]

It was literally just the word

[final hazel]

bridge

😛

[unique stone]

clutches pearls

[charred merlin]

Mind your manners, there are children about!

[unique stone]

stops car near some construction work
"This uh... This thing you're building.... Not very PG-13 is it?"

[charred merlin]

Oh my, I'm going to confuse so many people with this when they ask for help making a bridge for a song.

"Here's some helpful metatags, but I'd appreciate it if you didn't swear.", and just refuse to clarify.

[nimble sun]

I've done slight variations on the extensions with minimal change...if any.

[radiant mountain]

the audio input doesn't seem to care about style of music if you use audio upload, only if you give it lyrics it changes it, maybe im using it wrong?

[heady egret]

Sorry for late reply. The context window I'm thinking of is like if I have a song that is going on 5 minutes, at what point can it copy aspects that happened at the beginning before new extensions can no longer see them?

Say I'm trying to repeat an effect that only happened in the intro, at what time marker does that intro disappear to suno if I am extending from the entire song

[untold briar]

Currently 2 minute backwards/forwards, so the best you can do is 4 minutes between the intro and when you call back to it. See: #📚┃suno-school message for an example, the last note about how the last 2 minutes of the song was done with 1 clip for that reason

[untold briar]

If you want to maximize the time be sure to Get Whole Song before extending it thogh

[rocky belfry]

Why is everybody saying the b-word so freely without getting shot down by the bot?

I think I'll have to start my own hate group and appropriate lots of musical terms as slurs. Just to ruin it for everyone.

[charred merlin]

🌉 🫵

[soft smelt]

hey hows it going guys, im kinda new to this.
I made some songs and was trying to reuse some of the tags i made such as female vocals and EDM, and for what ever reason they wont add those to the new track? did i mess something up or am i doing anything wrong?

for instance, i did slow, female vocals, edm, dreamy.
and it made a song with slow and dreamy

[hollow hare]

Could have just been RNG sadly.

Try selecting a more specific subgenre (see https://rateyourmusic.com/genres/ for a list) for the style you want rather than leaving it open/broad. Just remember to use dashes for compounded words in 3.5 (e.g. dream-trance. See #📚┃suno-school message )

From what it sounds like, instead of just EDM, you can pick something like:
Vocal-Trance, Progressive-Trance, Uplifting-Trance, Euro-Trance, etc.

You may want to prefix Dreamy to the subgenre or the vocals, so something like dreamy ASMR female vocals, dream-trance, slow down-tempo adagio

There may be some genre that 3.5 does not understand as well, like phonk, which 3.0 does remember. In these instances, start the part 1 as 3.0 then Extend into 3.5

[rancid hornet]

Why did I lose my tokens I cancelled my subscription cus I didn't wanna get charged another month but I lost the ones I had paid for previously???

[bleak haven]

Credits included in subscriptions do not carry over from day to day or month to month. Purchased top up credits do not expire, but require an active subscription to use.
See the terms of service for limitations on commercial use. Email us at [e-mail] with any questions.

[rancid hornet]

Ok then send me my money back

[bleak haven]

I don't work for them lol, just a new chatter that saw your question. sorry wasn't trying to sound like I did

[hollow hare]

If it was the end of your month's subscription when you cancelled, then what likely happened is that it applied the daily credit renewal/refresh got applied when the renewal time was up (Refreshing for subscriptions happens at the specific time you subscribed iirc whereas free daily plan occurs at 0300 UTC, in about 40 min)

Regardless, nobody on Discord can help you with this. You must contact [e-mail]

https://media.discordapp.net/attachments/1069381916492562585/1238249445192237106/image0.jpg

[sharp kraken]

I think i might have gotten music box notation to work here, what do you think?https://suno.com/song/7a55986c-0c4a-4226-9eab-2584f549915d

[twilit scroll]

why did my credits not renew today ?

[hollow osprey]

Hi, i’m sure this has been asked before but sometimes when I use audio upload it flags my songs for copyright violations and won’t allow me to make them public. I’m creating my samples using paid iPad apps. Can I submit something to someone to get those restrictions lifted?

This has happened to me with straight up harsh noise music from hardware analog synths I recorded through my iphone mic as well.

I understand that you need a heavier hand in moderating audio uploads.

[untold briar]

If it's being flagged for existing copyright there isn't currently any way to handle special exceptions as it goes through a third party. If a non voice upload being flagged as a vocal sample (the upload is allowed, but can't be set public) then that may be more of a bug, and you could mention it, and my guess that filter has some chance of being tweaked.

[twilit scroll]

where are my daily 50 credits today?

[hollow hare]

Daily credit refreshes at 0300 UTC+0 (<t:1713236400:t> in your local timezone)

If you did not use any credits in the past 1.5 0.5 hours, try:

• Reloading your browser
• Clear your cache or Log in another browser (to ensure there is no visual glitch)
• If it still remains zero, contact [e-mail]

[hollow osprey]

Ok thanks. The one that was noise could have been it was hitting vocal ranges on the waveform or something. Will try reporting it

Mainly been testing the feature but annoying when I get a song I like.

[silk pumice]

Hey All, anyone got any theories on how Suno works? Me and some friends have been arguing for about an hour about how it could be done, but I don't think we're even close theoretically.

[hollow osprey]

It rebuilds compressed waveforms and is image based is my best guess

[silk pumice]

But it's not rebuilding existing waveforms, unless it's taking the closest reference and trying to swap out samples from other matching waveforms. But even then it wouldn't explain how it ends up with a different song structure.

[untold briar]

Check out older work by OpenAI, Jukebox from 2020 could do a version of something like this... very very slowly (like hours to generate one song at high quality) https://openai.com/index/jukebox/

[lilac wasp]

hi

[hollow osprey]

Stems will be interesting

[silk pumice]

Yeah that's sort of what got me so confused. This is way better than Jukebox and I thought the Jukebox project got scrapped. Is this some of the same people that worked on Jukebox?

[silk pumice]

I've been using Fadr for stem separation. Works well imo.

[untold briar]

No, but since Jukebox, there have been many advancements. But the overall idea isn't that different, compress music into tokens (or codebooks, a hierarchy of tokens) and predict them. MusicGen is a more recent example but I linked the OpenAI page because it explains things a bit, without just linking to the PDF of the paper like MusicGen

[silk pumice]

ahhh! Yeah that makes sense, would explain how splitting genres seems to work so well. How they are doing this so quickly is beyond me.

[hollow osprey]

Thanks for the rec gonna check it out

I assume suno will integrate in a future update

[silk pumice]

Yeah it seems to be in a lot of DAWs and there's a bunch of platforms offering it. Seems like a solved issue. RIP Izotope 🙏

[sage flower]

I’ve been building a cross platform web based platform targeting for audio editing that uses AI. Just so happened to randomly stumble across Suno about a week ago and as a musician I was really amazed, so I integrated stem separation in the platform. Figured out some ways to improve the overall output quality

[silk pumice]

Sounds cool! Probably shouldn't post links in here but if the platform is ready for testing I'd love to check it out, DM me.

[sage flower]

Awesome, yeah definitely would like to get others to test out the flows whoever’s interested. I initially developed it with podcasters etc in mind so stem separation and other features I’ve been building out has been newly implemented. Mixing, effects, multitracks etc already implemented. Just UI to get it looking pretty.

[sharp kraken]

im a DAW musician - psychedelic trance and weird idm/dnb stuff mostly - but i make odd ambient music as well and I use many different tools, ai and FFT tech

im interested

i use fl21 to seperate stems on all my suno tracks, then either make additional stems with fft synths like transmutator and morph3 or just record uHe synths

BTW Guys... im pretty sure suno gets the music box notation really good, look at this:
https://suno.com/song/794f2120-dc41-4711-8d2e-b4ea83add2df

[sage flower]

Yeah I think there can be a lot of ways to work on the tracks. I’ve been having a lot of fun messing with a few Songs. I can see the gaps that are left with out it generates the audio

[silk pumice]

https://discord.com/channels/1069381916492562582/1083129341870878780 showcase channel here dude if you wanna show some stuff 🙂

[untold briar]

There's also the new Workflows forum https://discord.com/channels/1069381916492562582/1252267444773326848 which is a good place for longer technical posts - technical prompting, or technical in audio post production

[sharp kraken]

ok thanks i didnt know that - yeah im not trying to showcase this, but i thought it could be helpful?

maybe ill just keep my secrets since everyones always blasting me when i try to share

[silk pumice]

Nah didn't mean it like that lol, sorry! Was just pointing it out. Not flaming 🙂

[sharp kraken]

at least i have not had a good time time trying to discuss prompting on this discord

[sage flower]

So Suno is basically generating the music from white noise? at first I thought it was generating the melodies/etc and rendering it server side / exporting it from a daw, but from what I understand it’s essentially “imagining” the instruments? Not using actual samples. I guess that would explain why when separating stems they sound blurry

[daring lava]

Sorry to hear that, very grateful for you sharing the awesome you make and the tricks you learn 🙂 keep up the good work. The music box example you posted was neat. Great suggestion from Jonathan Fly, posting it in #1252267444773326848 could inspire a lot of people floating through

[untold briar]

Suno is beatboxing all the instruments and the vocals

[sage flower]

Haha perfect way to put it

[pseudo patio]

Stem separation has and will probably always be hit or miss, it's amazing it works at all tbh. There's always going to be some overlap in frequency content so that you get loss, that loss is part of the music, it's unavoidable without absurd measures.

[hollow osprey]

I watched a video with ilya formerly of open ai explaining the big llm breakthroughs and how it all had to do with compression and predicting what the next bit of data would be

[sharp kraken]

Ill will add my insight to that forum i know you guys are still hashing out the organization on discord sorry if i disturbed the flow - thanks for listening - i just spent like 300$ my first month just trying to figure it out and i don't want other people to have to go through all that if it can be avoided

[hollow osprey]

Which makes sense for music since we’re all very used to compression with mp3s

i envision it building a song backwards

[sage flower]

Yeah it was one of the reason I developed a stem player / file format. I doubted AI based stem separation abilities and there needed to be a way to release music natively in that way

AI stem separation should definitely be a last resort as it’ll still be some time before it gets there, but it’s pretty decent for now if you can cover it up

[twilit scroll]

so when i used my creds in the past 1.5h before the refresh is, i dont get them ?

[hollow hare]

Oops, that was a typo on my part, meant 0.5 hr at that time of posting (since that was when the refresh happened)

This is also assuming you do not have a subscription plan either since they do not get daily refreshes

[pseudo patio]

sorry if I'm talking down to you ^^; I would guess there's more you could do with it that steps outside of pure reproduction, where instead of generating it wholecloth you use token cues to refine and 'deblur' using diffusion or whatever else from a separated stem

[twilit scroll]

i used mine like at 3:whatever hour my local time and it says it automaticly refreshes 50 creds at 5:00 my time

[sage flower]

I think there could be some really interesting things that could come from the stems actually. Just you bringing that up has given me some ideas 😀

[hollow hare]

So, to sum up:

• Did not use any credits since refresh (0300 UTC+0)
• Verified there was no visual glitch
• You are not on a subscription plan
• No other account issue (e.g. declined payment that needs to be resolved with [e-mail])

In that case, contact [e-mail] as soon as you can. Be sure to be as detailed as you can (e.g. account handle, what time you last generated a song with UTC offset [seen by looking at the latest generation time mark], etc.) And they should hopefully give you some gift credits for the missing time period after an investigation.

Sadly, this has happened in the past and this will hopefully help the staff find the cause since this issue has affected others (I remember one user stating they had missed several days worth due to the missing refreshes)

[uneven locust]

Hello All I been using the generated lyrics in the videos and want to ask you for more customizable features with them like font size also if you can remove the in between brackets info like {verse#] [Chorus] [outro] etc this feature change will make more attractive the video and use it as karaoke o just to understand the lyrics

[twilit scroll]

thanks for make it clean, i was confused little with the time stuff xd, i think i will wait for the next refresh shedule and if this not working ill contact the sup, thanks for answering

[lilac wasp]

Hi Guys

can someone be my new boyfriend?

[hollow hare]

Not the server for that type of thing.

[lilac wasp]

?

[hollow hare]

This is not a relationship server

[lilac wasp]

where can i find one

[hollow hare]

https://tenor.com/view/you-can-google-it-gill-engvid-you-can-search-for-it-online-you-can-browse-for-it-on-the-web-gif-10409511348401384306

[south kindle]

You could even write a song about googling it, but that's about it.

[untold briar]

https://suno.com/song/b6cad2f5-1a06-4ae8-8f53-cde392cc43e5 classic case of Suno lyrics using the song genre IN the lyrics, hah. (Otherwise pretty good job with the long instructions I gave it though.)

[lavish shard]

When are the credits from the referral program added? A friend signed up with my referral link and created 10 songs (actually 20, he used 10 prompts) but neither of us have received any credits yet 😟

[rocky belfry]

I feel like the more I extend the less different the two different versions are. To the point they I think is be better off with four more minutes of one version than two minutes each of two things I can't tell apart.

Is that just me?

[pseudo patio]

Think I answered my own question here. I don't think it likes noise, like anything atmospheric gets absolutely mangled, which is a bummer because I use an atmospheric layer a lot in my music but good to know. Makes sense, any time I've tried to get it to generate like rain or thunder or something it just 100% refuses.

[tacit shadow]

If a friend invited me before the referral thing was a thing, is there a way tonaward him points? @0utsidermusic

[tough kayak]

I'm sure I'm not the first person that's tried this, but has anybody noticed that it's impossible to upload dial-up modem sounds without getting the "recognized as an existing piece of artwork" error message? I've gone as far as pulling out an old laptop and spinning up my own dial up server. But even that gets flagged when I attempt to upload it. I know it's probably not a super creative idea, but darnit I got this song in my head and I really want to try making it 😦

[ancient glacier]

I have lyrics and video but I want to sync them any ai tool that can help em

Anyone?

[craggy kayak]

can i distribute the song created on suno with my own lyrics on spotify

[ancient glacier]

Only if you have paid for the premium

[craggy kayak]

okay if i buy 1 month plan and distribute the song , will i will get the rights of song even after my plans or subscription end

[bleak haven]

In the terms, it says with a subscription you have complete ownership and can use it for comercial uses and that lasts even after you cancel the subscription. Regardless of tier you always retain the right to the lyrics if you wrote them (if it's generated lyrics and your are a free tier then they own them). However that only go towards things generated while you are subscribed and does not apply retro actively

Though I would encourage reading the terms yourself, trust but verify

[low remnant]

I have not, so can't help you there. But I'm curious if it's so well known that Suno blocks it, have you tried just prompting for it? [Dial Up Modem Connect]. LOL sometimes it'll surprise you in stuff you're like "no way will this work"

[narrow jungle]

is there some essemble mode with specific modules for UVR5 to get a good outcome? even if its just sometimes? better than no good outcomes at all imo, i tried some guids the past few days but cant get it right

[tulip pasture]

How to "upsscale" audio and remove crisp and noise?

[south kindle]

I would download all the latest modules and try them all. I have found that different tracks give varying results between models.

[lavish shard]

I believe it used to be possible to tell Suno what motive to use for the artwork by putting it in brackets at the end of the song, like "(Cover art: mail singer)". Did I dream that? It doesn't work at the moment, but I was sure that I did it in the past with success.

[narrow jungle]

thats what i did but the only way to reduce noise is by using the only noise file and use that to remove in any edit app

[south kindle]

I found that I need to be careful doing that because sometimes what I though sounded like isolated noise can sometimes be a part of some other overall sound.

[charred merlin]

After experimenting, I genuinely prefer Audacity+Openvino, seems less likely to confuse the vocal & other tracks.

[narrow jungle]

yeah thats a big issue sadly

[narrow jungle]

nice i will try that out later, atleast adjusting levels on stems work somewhat if u dont overdo it

[charred merlin]

No direct means of upscaling yet-- a growing number of users are using automastering tools, but the best way to have direct control over the sound would be to use a stem separator like we're discussing at the moment so you can normalize & compress individual tracks.

[naive quail]

v3.5 Music Boxes are overpowered. They'll happily take over, even if they're the last item in the style list.

[tulip pasture]

Best de-noice de-crisp model in UVR?

[unique stone]

I've tried several times to derail a music box generation and make it do other things, it's a really strong style.

[naive quail]

I can usually force it out by excluding it completely from the style in the next extension. It'll still happily ride into that extension even tho it was left out, but by the NEXT NEXT extension it's usually gone, or at least concluding its assault on the other instruments

The plus side is that its real easy to bring back in

[fierce ether]

I have 2 easy questions if anyone would be willing to help me out.

1. How do I make the vocals more pronounced or clear through prompts?
2. Can anyone help me with uploading to Apple Music? I have one of the paid accounts and think I figured how to upload to Spotify, but need help with the former.

Thank you!!

[shy badge]

There is this kind of white noise in the background of basically every song. Sometimes subtle and sometimes almost unlistenable in some cases. Not sure if this is intended for a kind of Suno watermark? Or Suno is using noise Diffusion and the noise doesn't get cleaned up? Just bringing that up. Just in case.

Example: https://suno.com/song/5e0cd6e9-ee79-4379-a0b3-dd953397322f

[naive quail]

Noise has been an ongoing issue, but Ambient and Atmospheric can definitely increase it from what I've seen.

[shy badge]

Got it. And I used those tags I think for that haha

[naive quail]

Yea, I saw the tags, which is why I pointed it out. I've used Ambient the most... and it loves pulling in noise.

Handy for noisestep? lol

[shy badge]

Haha. I also noticed it does that with custom uploaded audio as well.

[naive quail]

I haven't noticed as much deterioration from start to end lately, at least. May just be the way I'm building up the songs in sections. I toss out the extensions that start getting too noisy unless I actually want to use the noise.

Yeah, I've noticed that with uploads - when they get merged in, they often sound noisy

[tender eagle]

I've extended a number of times without any deterioration. I think that what I do notice is less getting noisy and more. The vocals shifting over time. On part three. I had one track that was originally younger male voice. It sounded like and by the end it was definitely a more divaeque woman singing lol

[low remnant]

I haven't tried for a while, but at least in v2, "Clear Female Vocals" in style tended to tilt the volume towards the vocals over the instruments and make them stand out

[cold zenith]

when can we expect V4 ?

[fierce ether]

Ok, I tried "Clear" before or similar, but will try "clear female vocals" or "clear male vocals" and omit "ambient" or "atmospheric" thank you!

[tender eagle]

What is the context length of the extended tracks? One minute and 30 seconds, or 2 minutes? I'm thinking 2 minutes since the 4-minute generation time is cut to 2m maximum when extending, assuming =<2m of prior audio as context

[cold zenith]

twice 2min

[tender eagle]

So prior track =<2 minute and generated the same 2>=?

I meant the context of prior track's audio utilized for the generation

Eg. When uploading 1m upload audio then extend you do not get 2m you get 3m~

Cos I think it is simply 4m gen time. And prior track context length results in max=4m so 1m pre + 3m gen

[unique stone]

No information on this yet. Last update was basically that it was "cooking" and from what I've read they've been working on it since v3 was launched.

[cold zenith]

i wonder how they did V3.5 if they are working on it since V3....Did they divide their team ? One for the major update, and the other one for the " maintenance update " and minor update.

[tender eagle]

Yeah I know it makes 2x2m track I was meaning context length eg.

2m generated when 2m prior track as beginning, is total 4m context split between the 2m generation and the 2m prior generation.

It is going to work similar to any generative model with previous context used. For example with a text generative model if context length is 4096 tokens and you send a prompt of 1024 tokens you will only have 3072 tokens of the prior conversation as what it "remembers"

I guess simpler terms, how much of the prior audio is used when it generates the extension.

[cold zenith]

hum....it's true that it got to remember to maintain the rythmn

[tender eagle]

Yeah. So the reason I'm asking is if you consider some of the tracks how they end at the 4-minute. A lot of times you will get a outro break where it plays music and repeats things, but the beat and rhythm has changed at that point. However, if you know where the cutoff is for the context, since we know there's 2 minutes generated Max, we know that the context for a 4-minute maximum is going to be a maximum of plus 2 minutes on top of the generation 2 minutes.

So if it's shorter, maybe 30 seconds, and not the 2 minutes I'm assuming. Then maybe backing the song up to the point where certain things that you want to include for the entire song, such as drumline changes during a break prior to a chorus, will be repeated. You may need to go back further than expected if it was the two-minute length of the prior.

But making sure you do not include a long intro which does not need to be included could be also avoided

[low remnant]

I believe - with no specific knowledge - that it primes the generation with up to 2 minutes of tokens from the previous generations. So if it's not in the last 2 minutes of the previous song segment, it doesn't influence the current one.

[tender eagle]

Yeah that's what I assumed cos with upload if you upload 45s you get a generation of 3:15m.

But I know that it uses up to 2 minutes but I wondered if it actually did use that 2 minutes or if it was 1 minute. Because that changes things significantly is to how one would choose what to continue from

That gives me an idea. You know what they need to do. They need to allow us to extend from: 0:30 - 1:25. And so on

To segment the audio for example an 8 bar snippet

[cold zenith]

i totally understand what you said. But by experience with V3.5, it generally maintain the rythmn. So i think that we do have the 2min fully. But In the code it's something like 3min maybe, as they keep one minutes for the previous generation to keep the rythmn

[naive quail]

Unless I'm misunderstanding you, they already do this. That's how I build up songs in sections.

[tender eagle]

Absolutely not, it maintains the rhythm from about a 30 to 45 second upload. If you go with a 10-second you're probably not going to get a decent rhythm, but 30 to 45 seconds seems to do just fine and all the upload is is an extension

[tender eagle]

No you can back it up to a starting point. But you can't segment it at a end point. For example, begin the sample at 30 seconds of the prior generation and end at 1 minute and 25 seconds

[naive quail]

Ahh, gotcha! That would be nice!

[tender eagle]

Actually 13 seconds seems to maintain a decent rhythm itself. As seen here. https://suno.com/song/479a6101-a560-4d7e-bbf9-da916abfb82b. Which uses 13 s of this

As I only use the first 13 seconds of that sample I made

You can hear after the amen break that I dropped into it when it switches back to generation at the 13-second mark

And frankly the 2step it maintains is only about 7s of that

[unique stone]

I'm guessing this didn't happen and they just made time for v3.5 in-between other things. v3.5 is - as far as I'm aware - just v3 with some extra training data and small tweaks to reduce the quality degradation and thin out the vocal layers.

I expect this could have delayed v4 slightly due to the extra work needed but IMO it was worth doing rather than continue to have frustrated users trying to deal with v3's issues.

(Now instead we have fun trying to end songs in v3.5 😅 but at least that's easier to handle with post-processing.)

[low remnant]

v3.5 is v3.0 and v3a's love child.

[tender eagle]

Why end the song, extend forever , 😭😭😭

[cold zenith]

True because the quality degradation was too severe

[cold zenith]

it's easy to finish in V 3.5. Just finish it with V3. hahaha

[tender eagle]

They clearly adjusted the context length which I think plays a big role in the decoherence problem of v3+

[cold zenith]

finish it with V3. put [end]

[unique stone]

I think they mean: just let the song play forever

[tender eagle]

Yeah haha I was joking

I think I'm going to try extending until ded. See how long it can go.

I wonder if I can make what is somewhat like a continuous mix of electronic music, switching from DnB to brakes to trance over gens lol

[cold zenith]

you'll get the quaity degradation at certain point.

[tender eagle]

Are you sure?

[naive quail]

lol. The longest I've seen was like 20+ minutes

[tender eagle]

I really wonder about that yoktan. I think there may be quality degradation if it is a vocal track that expects consistency

[cold zenith]

i did with V2....29min...

[tender eagle]

But if I change the style and it's an instrumental or sample based like that track above, I don't think it will degrade

Since it changes enough, if you change the style heavily, that it might not gather noise from the prior generation

[cold zenith]

yeah most of the time, it's the vocal that decrease

[tender eagle]

That's because it can't maintain the consistency because each generation builds on the last and it picks up noise from the prior generation

If you've ever used local LLM text gen models. It does similarly with it's coherence.

It don't respond normally for the first few thousand tokens and then it'll start getting noisy responses from following its own gen style. Repeating things

"The chemical structure of the carbon ring is maintained by the chemical structure of the carbon ring is maintained by the chemical....."

Because it's trying to mimic a style that is becoming noisier and more repetitive over time due to some flaws and how the language model is handling the context

That's been ironed with some of the newer model types and layer structures. But with audio if we consider it the intended repeated states, such as individual Audio inclusions. For example, it's going to identify a trumpet sound, the drum style, the rhythm, the vocal cadence, the vocal pitch and style, the vocal sound. But but it doesn't do this in a 1:1 manner. It picks up noise cos the voices (instrument included as they're all treated the same) are not isolated. And those mimicry errors from the identifying the sounds with other sounds interfering, likely leads to noise entering the sounds for the next generation, after a while a clear voice may become a clear voice + white noise from a high hat that hit in about 3% of the vocalizations it used to mimic the prior. But over time. It will get worse and worse. Since it's only context is the prior audio... afaik. Which is akin to the text gen decoherence from context focus

That's my assumption.

Does version 3.5 have a paper by chance with the model basic architecture? May give more insight. I may be wrong how it works. But if It does work as I think from what testing I have done, I bet an easy way to get it to avoid the decoherence would be to have an option for generating the element style, for example always using a rising 3 octave spread for tonal sounds and a single hit of each sound of the drum kit. Via extraction from the prior track. And then begin the next track and all following tracks with the first extended generations solo element generations as part of its precontext. This could have it use these as a generated context which is only Gen the first time an extension is made so each subsequent gens use these to build its instrumentation and use the prior audio only for rhythm >1 extension.

Little bit more complex but I'm pretty sure it could be done fairly easily™

[low remnant]

I still think it'd be amazing if there was an eyedropper function for Suno, like Photoshop has for color selector. Mark a point in the song and this is what is uses to pull vocals and instrument samples, but is uses the rolling context window for keeping on beat, etc. And yes, I have dropped it in suggestions. 🙂

[tender eagle]

Eg. Song with trumpets / one vocal / drum kit / Acoustic guitar

Identify the sequences with the least noise and use a context 'training' layer to train each of those 4 voices.. then play pregenerate the trumpet playing 3full and 3half notes across 2-3 octaves then same for guitar, / have a short full phenime IPA output of the vocal + same octave rake / drum rhythm as from the song.

Train the model to accept these generated extension[1] samples as a voices [voices being and sound that is separate from another within an audio context, eg. A trumpet is a voice] context, and then generate as. Normal + the voices context which would need to be trained in to recognize as these being the voices needed to be used with the full audio context insofar as generating the next extension.

Now why this would work is that you would only use the initial generation from the first extension so that five extensions in it's still using the same voice samples in the generation of the subsequent tracks.

I think this would be a good method for both alleviating the coherence.

[tender eagle]

As well as allowing the user to do something like this since the first generation of the voices from the first extension would allow you to pick the instruments out

It would require some new layers in the model and some training, but I don't think it would be Just need to be fine-tuning rather than a whole new architecture.

[tender eagle]

Well not exactly. Something like. But more so maybe allowing the use of specific versions of instruments and vocals within other songs that you could transfer to pre-contextual voice set

[low remnant]

Eyedropper would also make duets like really feasible. "This voice at 30 seconds is the voice for this extension." and "This voice at 50 seconds is the voice for THIS extension." So it doesn't generate brand new voices each voice switch. But that's just me dreaming.

[tender eagle]

Exactly, so if the extension from the original song generates all the different voices that are variants , whether it's instruments or vocals, you could explode from the extension, it would work like an eyedropper, but would also serve primarily to maintain coherence between the prior track voices, which again I mean any audio segment that is uniquely identified within the entire audio context, not just vocals

[low remnant]

I'm happy in any way they find to implement function like that. Because, of course, I don't want the guitar to become over the course of the song a synthesizer any more than I want the voice to change or become more noisy. Hopefully they're going to make us all happy in v4. 🙂

Or v5. I'm a long-hauler.

[tender eagle]

Well, I'm wondering if something like that already exists. For the regular music that is. Because there has to be some sort of pre-processing going on, and maybe there's a model already for data processing which is extracting individual voices, again instruments, drum hits, etc. From the audio samples that are used to train Suno. Unless they're just using flat audio which they may be doing, but I would not be surprised if they had something built from an app like bark which is used to create pre-generated samples of individual audio from a sample.

I really need to sit down and look to see if they have any papers. I should just go read those If Suno has released any on their later models

[low remnant]

No idea on v4... but I'm reasonably sure up to v3.x that it's trained on all flat files which is why there's no stems available currently.

[tender eagle]

I mean consider this if it used a manner like this for a voices sample set. You could upload a drum beat. Label it as such, then you could upload audio of you singing a specific set of phonetics to meet as many phenoms as possible, and then you playing an octave roll on a guitar and then several octaves of chords and uploading each of those. And it use those in conjunction with its own audio set to create a completely fluid track, instead of the way we were doing with the audio. Right now I wag er that would not be too hard to build upon.

[tender eagle]

Yeah I'm going to look when I get a moment.

I'd really like to know what data set they're using that would interest me a lot. Is it one of the publicly available ones,

But then if I knew that I could probably get an idea of the dictionary. That's something I really wish they would release. Is the dictionary. Their data set tags and methods. If they had that information in a paper that would give us so much more control over our generations

I also wonder what encoder they use for the text input

[low remnant]

I'm pretty sure whatever v1 and v2 were trained on, v4 is going to be increasingly trained on our own generations. That doesn't mean they're not feeding raw samples of instruments in as well. But I'd bet they'd love to move away from using any external data for training as it just makes a wider and wider moat.

[tender eagle]

I sure hope they don't train it too much on our generations. And if they do only the first generation LOL. Otherwise he's going to pick up the noise just like context, the coherence in the extensions LOL

Either that or everything's going to sound like an anime song

[low remnant]

Ah, but the good/bad data they collect... plus some real samples to keep it grounded. There could be some magic there.

[tender eagle]

Well now I know, I have no idea how they actually process data sets. So I don't know what method they would use. And hopefully if they do use our generations, they have some glassy-eyed guy listening to 2 minutes of music for 16 hours a day. Picking and choosing the best. And not anything automated or based on our thumbs

Because I've used some stylistic labels in the generation that were not at all followed, but the output was still good even if it was that ultra generic Midwest radio rock That it always generates if it doesn't understand you.

[low remnant]

I know absolutely nothing other than that they've got some smart AI people working here and presumably trying to figure out how to maximize their own dataset of creations as that's something they have that's currently probably much bigger than their competitors.

[tender eagle]

Obviously and they've done quite well so far. I can't find a paper on v3. Neither could co-pilot.... Meh

[low remnant]

I'm not aware of any published papers so far. Must all be secret sauce. 🙂

[charred merlin]

Proprietary

Bark is open-sourced, though

[tender eagle]

Oh snap. Ok. So I see.

Like a text gen token vocabulary. Audio codebooks. So I never messed with Audio much it looks like the audio codebook is akin to q generative text model's token vocabulary. It's going to have snippets of audio in a higher dimensional vector as with token vocabulary of a textgen.

So it probably is going to have something like audio vector ≈ [a₁ a₂ ... aᵥ]. Where each vector represents stuff like frequency [261.6hz] Note [perfect middle C] instrument [piano] attack [52ms] attack tonal dynamics [represented mathematically somehow] plateau length ... Tonal change during x y z each being a different dimension.

I don't know the actual manner but maybe i should play with bark some more and actually check out its internals. It is going to have a token vocab also which is what I really want to see for Suno, how the training data is tagged and how it relates to relates the other audioᵥs would be so useful in controlling Suno.

Aside from understanding that knowing what encoder they actually use could be useful too, since it would give an idea of what relationship it can provide internally. If they're using meta's... I wonder...

I am going to see if I can shift bark and retrain using t5xxl

It'll give me something to do with these bored 30XXs lol

[naive quail]

It is going to have a token vocab also which is what I really want to see for Suno, how the training data is tagged and how it relates to relates the other audioᵥs would be so useful in controlling Suno.

That should be interesting...

[tender eagle]

Well it would be if they actually had it available

[low remnant]

You cannot control the Suno. It laughs at your primitive attempts to chain it! Mwa ha ha!

[tender eagle]

Hahaha well

Doesn't actually laugh, it either gives you Midwest radio rock or generic Midwestern accent electronic music lady If it has any question about what you prompted for style

[naive quail]

heh. I can get suno to give me laughter in a comedy sketch easily enough, but in the middle of a song... not so much.

[low remnant]

I got a laugh in a song, but it had nothing to do with the singer's voice. And to be fair, it was between a section marked comedy routine and the main music. But it sounded like a nice fake laugh you would expect in a song.

I won't link it directly, server rules, it's not PG. But if you go to https://suno.com/@portabob and search We were never friends. You can see the tags.

[naive quail]

I did get an unprompted laugh at the end of my first Lost and Gone track (crying about not being able to find my bong). At the very end, I added:

Oh shit... it's in my hand!```

Suno added a laugh. ![K_grin](https://cdn.discordapp.com/emojis/1057505487903531098.webp?size=128 "K_grin")

[icy depot]

Hi folks, I'm wondering if someone could help me with something please 🙂 Suno has created a song I really like, but the song ends abruptly. I've extended it but can't get it to blend into one track properly with Audacity. Is there a way I can keep the original song, but have it properly fade out?

[dusty moss]

can someone explan how suno creates music, someones telling me that Suno just copy/paste audio from around and cant create it . are they right ? lol

[low remnant]

What have you tried? Have you done a "Continue From" like 5 seconds before the end?

[icy depot]

So far besides trying to merge the two tracks in Audacity myself, I've only clicked 'extend' and generated another minute or so of it 🙂

[naive quail]

Try Extending with v3 from a few seconds before you want the track to end and have nothing except [End] in the lyrics box.

It may take a few rolls, but you should get some shorter extensions in there. I've had some that were just 3 seconds long.

Suno will refund the really short ones, but you can still use them

[icy depot]

Hmm, it's an instrumental synthwave kinda song so I clicked the 'Instrumental' toggle on 🙂 I think I need to turn that off then?

[dusty moss]

can someone explan how suno creates music, someones telling me that Suno just copy/paste audio from around and cant create it . are they right ? lol

[quick marsh]

hi everyone, do you know if subscription can be interrupted without any fee ?

[dusty moss]

im sure youll be fine

[naive quail]

Ya. Switch to custom, non instrumental for the extension

[low remnant]

Try leave instrumental, remove song style tag and put just "end" in song style.

One or both of the above lol

[naive quail]

Or that... 😄

lol

Between the two of us, I think there's half a solution there

[icy depot]

Alrighty! And that'll keep the same song as it is, but it'll just shorten it a bit?

[quick marsh]

i mean i would to test it for a month, you think i can unsubscribe at the end if i do not like it ?

[naive quail]

Yup!

[low remnant]

50% of the time every time

But they're only credits!

[unique stone]

You can pay monthly and cancel the subscription before it renews. (I have done this before.)

[quick marsh]

thx

[unique stone]

Nobody knows except the team that develop it, but it's definitely more sophisticated than copy/paste. Otherwise all of our songs would sound the same.

[icy depot]

Hmm, I must be tripping here lol. So from the generated original song, what do I do? I've clicked on it and on the drop-down menu, there's 'reuse prompt' or 'edit details', which just brings up a space to change the generated artwork.

[dusty moss]

That’s not true, I knew but I’m not technical and forgot

[low remnant]

You want extend. Then Extend from bottom left of image.

[icy depot]

Ahh right! I see, thank you 🙂

And in style of music, I'd clear that out and just put 'end'?

Or is there a special type of formatting syntax for it?

[low remnant]

Yep. Or uncheck instrumental and put [End] in lyrics. One of them should work after a try or two.

[naive quail]

Hell, try [End] in the lyrics box and End in the style. Two for one!

and name the extension End?

v3.5 will happily ignore it all, but v3 pays better attention

[low remnant]

I mean you can get REALLY fancy any put [Music Fades] in lyrics and end in style. There are so many ways to skin a cat in Suno it's hard to give the best advice.

And sometimes it ignores them ALL>

My best song ending - and this was v3a mind you:

"""[Music Really Fades]

[Song Ends]

[Kittens Die If There Is Still Music At This Point]

[Save The Kittens]"""

[naive quail]

Did the kittens die, tho?

[low remnant]

Nope. It saved them. Hooray!

[naive quail]

hahah

I was getting fed up:

https://suno.com/song/a64911dc-7f7f-45a3-82bf-21202171eac0

Just End Already

hah. I didn't realize that anybody can "get whole song" from shared parts

[low remnant]

Yeah. I wish there was some more control there. Like "ok for people to listen to, but it's mine. m'kay?"

[icy depot]

I've got the song ending 🙂 Cheers! Now I'm trying to merge the two tracks in Audacity without it having a weird break at the join point lol

[low remnant]

Can't you just "Get Whole Song"?

Then you don't have to do the merge

[icy depot]

God damnit, I can't believe I missed that lol

[naive quail]

hahah

[icy depot]

Literally the solution to the entire problem sorted in 1 click!

[low remnant]

/bow

I did my one useful thing today. Headed home!

[naive quail]

wish my workday went like that

[icy depot]

That's sorted it! Thanks very much @low remnant and @naive quail 🙂

[low remnant]

No problem and Happy Creating!

[naive quail]

Just remember us when you start touring stadiums...

[misty hearth]

hello a noob and Pro subscriber here. Is it me or a bug. I keep getting 8 or 9 seconds song generation and ofc credit was deducted 😅

[low remnant]

New song or extend?

[misty hearth]

new song hahaha

[low remnant]

Can you paste your prompt? Or share a link to one? How many times did you try? 1 short gen is different from 10x it did same thing, of course. 🙂

[misty hearth]

https://suno.com/song/b82d8e2b-5106-488a-99cc-b2a482414abd

[low remnant]

LOL I suspect it's the only lyric being "farewall", the AI is confusing with end and trying to end the song. Maybe put some more in there.

[misty hearth]

it generate 2 pieces and 1 is 8 or 9 seconds and a few 4 seconds that credit was refunded. It happen two days in a row to a few generation

https://suno.com/song/3c81b780-90df-4cc3-b579-ed5ba1e11bf1

[low remnant]

https://suno.com/song/4685ef35-431c-420f-bf93-1d17f241a57d I just took the farewell out of song style. And it generated 1x 47 second, one 1:30+.

[misty hearth]

https://suno.com/song/6ba9cf56-4cd9-4234-b95d-f9aa7aa7a8e7

[low remnant]

Check the one I linked and compare the prompt I used to yours. Keep the title out of song style, and no square brackets either. Mark as instrumental and I think you'll be fine.

[sturdy flax]

Hey everyone! 👋

Has anyone here figured out a way to equalize out or otherwise mix/master the infamous Suno static?

I'm looking for some tips or techniques to clean up the audio. Any advice or experiences would be greatly appreciated!

Thanks in advance! 🙏 #SunoAI #AudioMixing #MusicProduction

[naive quail]

A new way to force endings?

[misty hearth]

thanks

[misty hearth]

Thanks

[charred merlin]

Can you share a particularly notable example of the static you're referring to? It's important to gain agreement on what we're referring to as static before any helpful advice can be given.

I've heard some instances where the sound being referred to was just a noisy Chinese crash cymbal, an issue that's resolved in-studio by audio engineers (either by carefully positioning the microphone, not using "trash cymbals", or doing literal black magic voodoo in while mixing or in post).

However, there have been instances where it's just the bass continuously boosting the gain & amplification to the point where it peaks and you get a literal, audible static.

In both instances, I'd give the blanket advice of reporting them through ... > Report > Bug > Poor Audio Quality and giving a thumbs down to help improve the model, but otherwise it's subjective & case-by-case.

[sturdy flax]

I am referring to the 'noise' people hear that automatically allows you to tell it was made with Suno

For lack of an exact description

[charred merlin]

Without a specific example, I can't give any more specific advice than to report the audio quality, or perhaps apply some compression filters if you feel there's noise on a track.

Not trying to be difficult, it's just nigh-impossible to give any kind of specific advice without gaining agreement. 🤷‍♂️

[wheat quail]

Just want to know what's considered "ok" when uploading a song!
Let's say i take an old song like, cola bottle baby, remove the singer, and just ad me being silly.... (so whhat daft punk did basically) is it ok? (i mean if law didn't troubles them, technically it's ok no?)

[fluid robin]

is there a way to getl like [background vocals]

that is singing diffrent lyrics then the main one

[naive quail]

There's usually licensing involved to get clearance for using samples. The artists that don't get that clearance can run into problems...

[wheat quail]

Else you modify it of course, buut isn't singing on top already a modifier?

else how covers channel aren't all banned?

There is hundred of thousands of people doing that ^^

[naive quail]

Probably more a case of it not being worth the cost to go after little fish.

[charred merlin]

When you upload a video to youtube, Youtube doesn't say that you have ownership over the song. In fact, if you try to monetize something you don't own, you get DMCA'ed.

The expectation is that you use Suno to create original content-- that's why you're not allowed to use copyrighted material, because they don't want to give the impression that you can upload "Here Comes the Sun", ask for a metal solo at the end, and then own the song.

It's a pretty big deal when companies / copyright holders go after people who are using their respective content in fair use, like non-monetized youtube videos where you just want to sing and share your good vibes, or community modding projects that look to breathe new life into old games.

But Suno is granting ownership and permitting general commercial terms over the content you create using their product. Just gets legally mucky if they don't at least try to prevent you from uploading content you don't own.

[wheat quail]

Maybe they are not monetizing it you're right...
(but some really make no effort, in my exemple,)
((No worry i properly edit and modify my samples ^^ That's not forbidden as long as you don't straight reuse the same thing anyway!))
As i don't have material nor knowledge my best chance to have the same "type of instrument" and "basic feel" it to use sample (Especially for old games vibes)

I was just wondering about if singing is a thing

(...Still as i'm a bad singer that won't solve anything llol)

[charred merlin]

I'm sure there's a lot of instances where people manage to eek out a bit of monetization on content they don't own, but the process of detecting, DMCA striking & then hoping to collect might not be worth the effort for the proper copyright holders, and would only reflect poorly.

In any instance, there's a lot of grey-area in regards to all of this. I'm just a layman, so I really don't have an insight to any of it, but it does make me wonder where the distinction between being a legally distinct alteration or not comes into play. But that's a chat for another day in another server 😛

[somber fern]

I'm trying to do a female county singer, is there a way to make her sound not so awful?

[hollow hare]

Try to be more descriptive on how you want it to sound like "clear Texan Texus female vocals" in the style prompt

https://www.suno.wiki/faq/metatags/voice-tags/

[somber fern]

so wouldnt that be for the vibe: clear-southern-female-vocals

[hollow hare]

Dashes are typically used for compounded words, so more like:
Clear southern female vocals
Clear southern-Texas female vocals
Etc.

[charred merlin]

I would love this so much-- Suno's just trying so hard to be helpful that it doesn't want to keep singing and bother you if you've said goodbye to it.

[hollow hare]

Though I am unsure if Clear and Southern would conflict with each other...

[charred merlin]

I love this little music robot so much that it hurts.

[naive quail]

Suno will try to interpret whatever you put in the style field. Even "Not Dave" 😄

[somber fern]

ya the only reason I ask is cause the highs are horrendous

[charred merlin]

I'm sure SixStringToker knows a thing or two about horrendous highs.

[naive quail]

coughs

[somber fern]

so for the highs not to sound horrendous I would put what?

[naive quail]

haha. I saw that. So bad. 😄

[charred merlin]

I couldn't resist continuing the marijuana puns, but then I realized that I wasn't helping the situation at all, regardless how funny I thought it was.

[charred merlin]

Do the vocals sound horrible right from the start, or do they get worse after a certain point?

[somber fern]

they start out not too horendous and get better but the highs are the worst

[charred merlin]

Drats... if it was the other way around I would have suggested creating an extension point before they get bad, but if it starts bad then it's harder to hold onto the later sections without taking the song apart with stem separation.

[naive quail]

It could also be something in the lyrics steering the voice in that direction, unfortunately.

[somber fern]

i mean its not too terrible starting out but its the highs, and ive experienced through several songs that the highs are the problem.

guess ill try Clear southern female vocals and report back

[charred merlin]

In a future version we may be able to fix these types of issues with "inpainting", but as far as I know for now its just a matter of experimenting with different metatags & style prompts, as well as rolling multiple generations to pick the best option from as there'll always be a degree of randomness.

[naive quail]

I'd probably also experiment with the position of that callout in the style field (relative to anything else in there)

[somber fern]

didnt really put anything in there for the singer, just the music. oh sorry i thought you were referring to the []'s

I see what your saying, so like Clear southern female vocals should be first instead of last or both or wherever

[naive quail]

Suno seems (slightly) less inclined to ignore things like that in the style field vs within the lyrics field.

slightly.

[somber fern]

yep I've noticed lol

[naive quail]

Or 2nd, or 3rd, etc... It can sometimes make a difference, at least with instruments

[somber fern]

now the next question, how to get rid of that awful autotune sound

like it sounds like its overkill.

[naive quail]

It's too late. You're gonna have to reinvent the music industry.

[somber fern]

lol

i meant when i gen a song

[naive quail]

auto-tune/pitch correction is in damn near every commercial tune these days... probably all over the training data. It's just a crapshoot.

[somber fern]

wonder if I could type "not too much autotune" if it would listen

[charred merlin]

"natural voice"?

This is the same reason why it can be difficult to create specific sounds in isolation, like just a drum beat. It's not that Suno doesn't know what we mean by "drum beat", it's just that whenever it "thinks" about a drum beat, it normally associates it with a jazz standup bass or something.

Even if you solely use the exact metatags and style prompts that would describe a drum solo by itself, it still might take a handful of generations before it's able to "imagine" a drum solo without any other instruments.

Luckily, we can "hack" the process by making a handful of generations, taking the most promising candidate(s) with even a couple seconds of perfect audio at the beginning before it veers off into an undesirable direction, making notes as to where we can add extension points, and chipping away at that same process ad nauseum until we have something that matches what we originally heard "in our mind's ear".

That way, you've had a hand in every step of the generative process, and any odd static left in the result is because you specifically preferred the relevant extension track over the other generations that you chose it over.

[naive quail]

that may very well get you MORE autotune since negatives don't currently work 😄

[frigid pond]

Hi!
Did anybody figure out how to force v3.5 to finish short bits of songs?
It keeps wanting to use lyrics from before.
It seems it does not want to finish the song

[end] no matter how many times used, does not seem to work.

[somber fern]

actually I believe that helped. the highs dont sound atrocious, but that autotune is still there

[acoustic hornet]

I constantly have issue with that as well as it playing random music that does not seem totally in line with the rest of the song before it. I've tried [end], [outro], [stop], [fade out]. Nothing. Still does it.

[frigid pond]

As you wrote it, it managed to make one short version. But I should not mention how many credits I have sunk to find one short version.
I know I can use v3.0 but it just struggles with accents and punctuation and it does not fit the rest of the song

[acoustic hornet]

I only ever have luck with "short" songs if I stick primarily with only one music style. Honestly kinda boring, wanting to try a good mish-mash.

country, punk, and swing seem to do best at the so-called "short" song, not grabbing additional lyrics from before and not playing well beyond when the song clearly ends.

[frigid pond]

Well I don't want to be mad about 3.5 coming out.
It really helped with distortions appearing after multiple extensions appearing in V3 but making it not pushing for 2 minut mark in extended songs marked with [end] would help immensely.
Maybe in v4

[acoustic hornet]

If anything, I'm happy with 3.5 - but I would really like to see a tool to "trim" and "fade out" added into the setup, in case we get those long drag on moments. I don't like sharing a song that has wonky extra bits at the end.

[velvet bolt]

Same. I guess the best option now is downloading it and adding a fade out in Audacity or some other DAW

[acoustic hornet]

I have done that when sharing outside of Suno, but I don't even want to make the songs I make "public" on Suno because of the extra wonkiness.

[velvet bolt]

Yeah, same. It's weird, even if I do an "extend" with the lyrics just set to [end] and the song ended on a good note to stop at, 3.5 insists on adding at least 25 seconds

With v3 it wasn't too unusual for it to generate 2-4 seconds of fade out and refund me even when I wanted it to keep going, v3.5 has the opposite problem

[acoustic hornet]

Honestly, I am fairly new to this whole thing, and I only got to see v3 for a couple of days before 3.5 went live. I've had decent luck with some songs that I've been happy to DL and fix up a bit, so I subbed. Ever since subbing, my luck with it has, for lack of better words, all but sucked.

[charred merlin]

You can still choose v3.

[acoustic hornet]

I'm aware, and I've done a couple with it. Not too happy with how the extends have worked out though.

[charred merlin]

I only got to see v3 for a couple of days before 3.5 went live

Sorry, it just came across like you were implying that v3 changed

[acoustic hornet]

Oh, no. Apologies.

Random aside, different issue. Can anyone explain why using the "melancholic" style is akin to "pop" with a... uhm... phat beat?

[charred merlin]

v3 or v3.5?

[acoustic hornet]

Both

[charred merlin]

https://suno.com/song/1c0e9563-f83c-410d-9a7f-35f020d0fd77
https://suno.com/song/4083f6e1-9daf-4d6d-a975-9e385d34eaf7

I don't have the same findings.

Sounds like one of your other style prompts, metatags, or the lyrics themselves are overpowering what it might interpret as a subgenre as opposed to something stronger.

I'd suggest breaking your desired song into smaller chunks and try to create the specific intro before worrying about the rest of the song's composition or lyrics.

[acoustic hornet]

I've been trying to make a somewhat sad sounding medieval bard tune, and no matter what styles I try mixing, even copying a couple I've seen from the featured music of medieval sounding tunes, mine almost always end up sounding like synthwave-pop.

[sweet wigeon]

I believe the list of word they use to communicate in various languages, not just English or Italian is huge. Not an easy thing to do.

[acoustic hornet]

The last one I tried was styles: medieval, melancholic, female vocals - the only tags I have in my lyrics are [instrumental intro], [verse #], [chorus], [bridge], [instrumental outro], [end].

[charred merlin]

Then take out the tags and focus on just the beginning. Don't try to one-shot the song

@unique stone @final hazel where we did land on 🌉 , is that still a swear word?

[final hazel]

try it

[charred merlin]

You're trying to get me banned, aren't you 😉

[rare stirrup]

Try 'troubadour' and 'acoustic', that will probably give you what you are looking for

[naive quail]

BRIDGE!

[acoustic hornet]

I'll give it a shot.

Curious... why would "bridge" be a swear word?

[naive quail]

by mistake 😄

[charred merlin]

OH MY GOD you two, mind your manners!

[naive quail]

lol

such rudeness!

[charred merlin]

Woodhouse Bot went crazy last night, but it was funny enough to me that it's now part of 'Suno Lore'

[acoustic hornet]

Was it a mistake or a censorship thing to not talk about business initiative. I know what the initiative is, and I know other areas of Discord where even typing it out gets you a warning.

[charred merlin]

Complete mistake, afaik

[graceful condor]

@final hazel v4 coming soon is there any timelines?

[charred merlin]

https://tenor.com/view/nervous-shy-kermit-terrified-scared-gif-17697469

[naive quail]

https://tenor.com/view/john-cena-im-watching-you-i-am-watching-you-wwe-watch-out-gif-25856329

[ancient glacier]

V3 extend function doesn't work

[sweet wigeon]

Who made Bark model?

[wheat quail]

do someone know if there is a good AI to turn my unplaisant voice to something cool that i own ? (and that less us mess with making it lower, or else?)

[charred merlin]

Suno.ai, but they've open-sourced the text-to-speech model and closed-sourced the text-to-music model.

[sweet wigeon]

I've noticed that there is also HF Facebook model genmusic, it can generate music too. I wonder if Bark was trained from this genmusic

[ancient glacier]

It just hallucination