📚┃suno-school

[charred merlin]

Trying to organize by song name, renaming them to add edit/extension points, can only go so far. I'd love a workspace function

[slate idol]

Hello,
I have 2 technical questions.

I generated wonderful music with lyrics.
Now I would like to generate the exact same music but I would like to change the lyrics.
Is it possible ?

My 2nd question is:
Is it possible to separate all audio tracks (all instruments and all vocals)?

[charred merlin]

Use the search feature to look back through previous conversations about 'stem separation' or 'track isolation'

[slate idol]

In other words, it's not possible to do this with the suno software ?

[vestal rapids]

no, it's not possible from within Suno

[slate idol]

So sad !
They have to improve this.
I think it's easy for them

[vestal rapids]

it's not as easy as it might seem

[slate idol]

When they generate all the tracks together.
I guess they can also generate the tracks separately

[low remnant]

It's trained on mixed together tracks. The model doesn't (in this version) know how to generate vocals separate from instruments.

[vestal rapids]

the way the model works it isn't easy to recreate precise tracks across different generations

[slate idol]

It is a crazy software !
I love it

[winter condor]

your best option is probably a vocal/instrumental separator tool

[slate idol]

I am a newbie.
Which one is the best ?

[winter condor]

i think i know one wait

i think its called ultimate vocal remover

[winter condor]

its not very user friendly but its still usable with some learning

its not hard but for a newbie its not the easiest one

[weary wharf]

If you have a daw, and know how to use, do a stem seperation and re-write with your own instruments.
If you can sing record yourself.

Stem seperation comes in Logic Pro - needs Mx chip -- arm processor.
There are online seperations tools.
There is also RipX.
Search and you'll find a lot of tools for this.

Check youtube for comparisons.

[graceful condor]

gonna throw my 2 cents and recommend BandLab, it's free and has a splitter tool that is pretty neat

[fluid bridge]

i am currently experimenting with trying to do long form music generation with bark, but am having some trouble where each snippet of audio sounds completely different from the rest. i understand that bark is nondeterministic, but is there some way to ensure that each snippet is somwhat similar to each other so that the long form output is cohesive?

[vestal rapids]

that's really tricky to do, not feasible

[fluid bridge]

ok, how does someone in general go about doing long form generation? is there some other model i can try using (i have heard of chirp, but not sure how to leverage this)

[vestal rapids]

music generation has moved onto the Suno website, you can begin creating by going here https://suno.com/create

[vestal rapids]

I am doing long form music with Suno v3.5 which is available as a early access model to subscribers, it has generation length of up to 4 minutes which can be extended in 2 minute blocks

[fluid bridge]

ok cool, are you doing this programmatically via some sort of model they've given you? i am definitely interested in learning more about what this entails and how to access this

[weary wharf]

You can use the website, no models or apis.

[vestal rapids]

the way I am doing it is to give the model a text prompt and selecting the v3.5 model from the model chooser

[vestal rapids]

let me know if you need any further help, happy to lend a hand

[fluid bridge]

thank you! just to clarify, there is basically no way to use bark to generate good audio songs?

i am trying to programmatically generate audio given some lyrics

[vestal rapids]

everything has moved on to the website, I believe Bark is now deprecated

[fluid bridge]

ok cool, thank you for letting me know

[vestal rapids]

what genre of music are you making?

[fluid bridge]

mostly pop

[vestal rapids]

ok then the new v3.5 model is definitely for you

[weary wharf]

Does anyone know if the WAV is generated from the original spectrogram that is generated? I hope it's not decoded from the mp3 😄

[low remnant]

lol that would be funny. "You wanted .wav files, wish granted"

[vestal rapids]

the devs could answer that but in my testing, the WAV file is of significantly higher quality than the mp3

[unique stone]

Spectral analysis of wav shows presence of high frequencies missing from the MP3 so there is some difference

[weary wharf]

Ok, did not check if there were differences with the mp3 vs wav. I think this is generated directly from the model output -- which I assume is a spectrogram 😄
Thanks 👍

[undone star]

depends on the subgenre of pop. some pop things are not as they used to be. ^^

[vestal rapids]

I have been getting some nice euro-pop tracks, some pushing 4 minutes

I only do instrumentals generally

[opaque canopy]

lets say i have a website (i want to trade / sell my magic cards collection) i was thinking to use SUNO to make song that describe what i want to trade or sell would it be possible or i would be ban or what so ever lol ?

[final hazel]

@brittle minnow The lyrics contain suggestive content that is not suitable for a PG-13 rating. I have to remove it

[exotic forum]

The new 3.5 update won't let you extend songs

[brittle minnow]

what, are you serious?

[final hazel]

Yes I am. Sorry. Again, this discord is PG13

[brittle minnow]

Ok, i didn't think it was that bad, well remove it then, can i keep the song on Suno, but not share it here, is that what you mean?

[final hazel]

It's only Discord here.. for the rest it's up to you 🙂

[brittle minnow]

But to make sure that i dont repeat this mistake, its not okay to have jokes about sex?

[final hazel]

It's not always easy to be sure what is PG13. In doubt you can always paste your lyric in ChatGPT and ask if it is PG13

[brittle minnow]

i see, thank you 🙂

[final hazel]

And it's absolutely not personal, what so ever. Just following the rules Suno wants us to keep here. 🌼

[vestal rapids]

yes it will

[brittle minnow]

wow, 3,5 got my song right on first try

[hollow hare]

As for Suno site itself, there is a portion of the Terms of Service (under "Conditions of Access and Use") you may be interested in.

Violating these conditions could have a consequence that ranges from neglible (e.g. not listed on trending/Frontpage, usually for small things) to severe (suspension/banned, possibly reported to local authorities, if the content warrents such a thing).

All in all, you should be able to tell what the limit is even if you do not think it is fair

[brittle minnow]

O_O

[exotic forum]

I've tried several times and it keeps giving me credits back before the update it worked fine

[brittle minnow]

Now i am scared lol

[vestal rapids]

I have just finished a 13:30 song

[hollow hare]

Don't need to be scared, standard ToS terms.

For the severe reaction, as an example, it would be something like explicitly naming a school and implying an attack against it

[final hazel]

It's fine.. your song was only suggestive. I wouldn't be scared. 😉

[exotic forum]

I'm not sure what's going on I'll try with a different song

[brittle minnow]

I thought Suno autoamaticly stopped you from generating if it considered the content u trying to make have prompts that dosn't follow guidlines.

like i wanted to make a song about my friends suicide, but it wouldn't have it.

[hollow hare]

There are ways around the censorship system, it is moderated by ChatGPT.

And the best content are made by people trying to skirt censorship (while, y'know, not being obscene about it)

[brittle minnow]

I see 🙂 its mostly okay xD

but man i am impressed with 3.5

the accuracy on the lyrics is so much better now

[hollow hare]

Your song was all about the implication and nothing bad about it 😉

Though not PG-13 appropriate, it is nothing obscene like the one person who made a song about... er, let's say forceful action which is illegal everywhere

[brittle minnow]

i made like 50 clips of my song and none was this good as the last 20 i've made with 3.5

[brittle minnow]

Geeze, not writhing that shit xD

[glacial scarab]

3.5 is amazing where it's good. It has it's strengths and weaknesses.

[brittle minnow]

Hmm, must try making another emo, metal with 3.5

Dystopian, emo, metalcore seems to be my top 10

[brittle minnow]

imma find out

[glacial scarab]

biggest "weakness" for me was that the prompting seems quite different. And for some reason it sometimes just really wants to end the song. The noise seems to occur a bit sooner as well but not enough data to verify (had only 500 credits when it launched) but it can do stuff v3 couldn't and the bigger context is amazing

[brittle minnow]

sounds good 😄 writing my second song now on 3.5

my new song is gonna be FIRE!! So pumped

[hollow hare]

Just remember that any song you let out in the public (let it be by making the song public, posting the link/video here, etc.), there are likely bots that will try to steal and distribute the song (Well, moreso if they rock, which is why Trending pages are primarily targeted).

If you are planning on having it monetized, ensure you give it to the distributor and upload an unlisted/private video to YouTube.
If you do that, do not make the song public until it is published.

If you are planning to get the song fingerprinted but do not want to distribute it or do not have a distributor, you can use DistroLock (just be sure to upload to YT unlisted/private first before using DL so you do not waste time unblocking the upload)

[brittle minnow]

Ok, thank you, i will keep this in mind 🙂

[final hazel]

@hollow hare please check your DM 🌼

[brittle minnow]

So far i've just made songs for personal use, i prefer making songs i wanna listen to instead of listening to existing music now since i found Suno. Its amazing how you can just enter a prompt and get somewhat what you wanted to hear.

[low remnant]

Uploading them to Spotify can be addictive as well. It's really amazing when it makes a mix playlist with your songs dropped in with mainstream songs.

[brittle minnow]

when do you think we will be able to save a style of a song that came out with super good melody and song structure but terrible quality and have the AI refine the one song we choose.

[brittle minnow]

That would be cool.

[lone oak]

3.5V seems to have solved the intro, and the accuracy of the lyrics has also been significantly improved. However, the tags and V3 seem to be recognized differently, and the probability of generating a suitable melody has become lower. I wonder if you have felt it?

[hollow hare]

Jonathan Fly's made a note about that. If you did not see it yet, check out #📚┃suno-school message

[lone oak]

thx

[warped furnace]

Anyone knows how to prompt a part that should be spelled letter by letter? Does this work? L-E-T-T-E-R? 🤓 Or do I put something in brackets?

[low remnant]

I tend to spell phoenetically, NPC = Enn Pee See. But there's many ways to skin that cat.

[warped furnace]

Thanks for the tip. 🙂

[alpine venture]

amateur drunken, bad, low-quality neo-anti-creepy gross, anti-noise fx, jumpscare piano, accidentally catchy

most reliable simlish

I've ever seen

https://suno.com/song/c774d6ef-e39e-40db-856a-a16b8d0229f3

[robust skiff]

Does anyone know how to shorten a song that was generated. It is great, but it goes on too long and doesn't end correctly

I could just download it, but I wasn't sure if there was an in app solution

[final hazel]

Look at the time where you want it to stop, hit extend, fill that in on the time stamp, fill in the lyrics box {end}, and then click the button to extend it.

[alpine venture]

https://suno.com/song/7cfb41e1-b07a-4f70-b5b5-7b14b8eebf7c

[robust skiff]

Awesome! I appreciate the help

[short prism]

How do you make a song last 3 minutes and the sound gradually quietens down at the end? Now I only have 2 minutes and the song cuts out.

[hollow hare]

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

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

[short prism]

thx

[craggy delta]

DAMN, V 3.5 DESTROYED THE 80S STYLE OF VOICES! NOw it keep singing on that stupid Green day stinky modern rock style of sing. WTF? I hope this get fix on the official V 3.5

https://suno.com/song/c0fc9203-fe8c-4f29-ae0a-bef00cc83819
this TRUELY SUCKS!!!!!!!! WTF Suno???? WHAT HAD YOU DONE TO MY FAVORITE GENEREE???

THIS SUCKS! BIG TIME!!!!! 😡

[hollow hare]

Using dashes instead of spaces for a compounded terms will help (e.g. Glam-Metal)

See Jonathan Fly's post about 3.5: #📚┃suno-school message

[broken lotus]

v3 is still there. Why are you throwing a tantrum?

[sharp kraken]

this was soooooo fun

https://suno.com/song/84e412d6-f3b3-4766-8f69-eb746c053bd7

and Octotonic Fugue that goes from Baroque to DnB

with a [Vocal Morphs All Instrument Continuously]

[ruby condor]

I bought credits. A few months ago. I was temporarily homeless. I come back and the credits I bought are all gone?

[craggy delta]

NO WAY!

IT WORKEEEED!

https://suno.com/song/0aaca5b8-d8f6-41df-bc69-7eecb4ee2201

FREAKING EPIC SONG!

[finite quartz]

so, I can only get songs on the create page to extend. Tried it on several songs from my library going back a few pages and all I get is "An Error Occurred" . Specifically I wanted to extend an instrumental song and I tried it both with and without "Instrumental" toggled, still get "An Error Occurred" . I tried it on others as well, same error. Are only the latest songs that show on the create page extendable?

[errant meadow]

I have noticed that during loud portions of a song on 3.5 it gets a weird feedback or reverb.

this is on one song, but wondering if anyone seen that too?

[sharp kraken]

it can happen

[ocean grotto]

how would you try to prompt to get something like this: https://open.spotify.com/track/4oMgue1JcsV7qirqfRKawD?si=a90ee02584ae4149

[final trail]

Vocal flow was made super basic is what it means

[wet minnow]

if you get tracks that sound weird or unlistanable please give it a thumbs down. that marks the track as bad for the devs to look at.

[south kindle]

To be honest, I’d be more willing to experiment with an early access model and provide testing feedback if the generation were at a slight discount. Otherwise I might rather stick to what gives me more expected results since it is at the same cost.

[fierce bone]

Im curios... i use "thumbs down" just for versions i dont like even tho there is nothing wrong with them. It an easy way to like keep "workflow" clean. is that bad bc it will add tracks that by no mean sound wierd or bad but just was "not to my liking".

[wet minnow]

the devs asked to use the thumbs down as a indicator for "hurts my ears" kind of bad generations.

[fierce bone]

yeah i mean i do that to! But is it bad that i do it on "good tracks" that i just personally disslike

[south kindle]

They should make it a 😖 instead of a 👎

[prisma wraith]

But what if I distinctively aim for something that will hurt your ears?

[wet minnow]

they will also end up in the potential pool of stuff to look at. but thats just a guess

[wet minnow]

make it as long as its possible 😄

[fierce bone]

i feel like you are intentionally miss interpreting me 😩 ... i get that they will! My question is should i therefore avoid using thumbs down for tracks like that if i wanna be helpful? Like i understand they will be looked at but does it make a difference?

[wet minnow]

i dont think it makes a difference. i also used the thumbs down just to clean up my create tab from general stuff. now for the time of testing 3.5 i have shortly changed my thumbdown use to just mark really bad stuff. maybe its account dependend if they see a user does that they look at more of their thumbed down things. i can just guess because i dont really know more then you.

[south kindle]

I just instead put the stuff I like into a playlist and only thumbs down the stuff I really think is wonky.

[glacial scarab]

it should really become seperate things the upvote/downvote and quality feedback. But maybe then it's "too much work" to give quality feedback for the devs... but also then it would only be used by people that only want to give feedback about quality and not "hey that one was funny I like it"

[naive veldt]

I thumbs down a few of my songs to let me know that somewhere in the song it goes off the rails and dies, with screeching metal sounds or something else wrong

[unique stone]

My strategy for v3.5:
Bad audio quality - downvote
Good audio quality - upvote
"OK" audio quality - no vote
I find the music interesting and might want to work on it - rename it and put the word "INTERESTING" at the end
I don't like the music at all - trash it

[wet minnow]

my startegy is similar. all my internal tracks have gibberish names like 4ggr4t or so... and if i like it it gets a proper working name while listening to it.

[prisma wraith]

^same, but I also downvote things with O.K audio quality, if it goes against the style tags

[naive veldt]

yeah i go a step further that after a thumbs up, and giving it a custom artwork ill put it in my "best of" playlist to finalize it was a decent song

I kinda wish suno would auto label a song with (1) and (2) or something if its the same name

[wet minnow]

a quick picture change is a brilliant idea... have to steal that... 😄

[naive veldt]

yeah a custom pic is a quick look down the list to see which ones of thumbs up in the created songs list that you really liked

[unique stone]

A big "APPROVED" stamp 😄

[naive veldt]

i also wish you can see all the plays and upvotes on every song and can auto order the playlist to which is highest played or rated, thatd be nice

[cloud cairn]

Hello guys I have a question.
I am trying to make a song which sounds simmilar with Melatonin Dream About Followers.
Here is a link about this video. https://youtu.be/ZQ1olyUUYSU?feature=shared
But I can't found it's genre. Can someone advice me about that?

[foggy jasper]

I think ive annoyed suno lol

getting this error

using the same prompt that's been working

weird, the error is fixed when I switch to 3.5

[lavish hinge]

Me too on all my tracks honestly the start is really good but after maybe 1min or 2min the degenerate and fucked up all the song 😩 💩

[broken lotus]

Paste the yt link into sonoteller.ai (some free use per month). They actually listen to the audio and give an analysis of the genre as a weighted list which is surprisingly accurate for Suno v3….

ChatGPT an also tell you ‘opinions’ on a song’s genre/style but that seems less accurate as a style prompt. Try free gpt:
“Describe the music of <band> <song title> to an AI music generator in 80 characters or less, without mentioning the band or the song”

[valid musk]

Anyone figured out how to force an instrumental intro on v3.5 yet? I’ve tried so many things and still trying to brute force it. But it refuses to do so and just does a breath and starts lyrics instantly.

[broken lotus]

Use Extend for time, and cut before the audio becomes compromised

[broken lotus]

I use v3 for instrumental, and instrumental openers. V3.5 has been trained to be very vocal… 3.5 can force a v3 clip to sing when v3 refuses

(Use each model to their strengths)

[cloud cairn]

Thanks a lot

[valid musk]

Would be nice if it could do both vocal or instrumental. I don’t like having to stitch. Much more streamline getting all lyrics in one generation.

[lavish hinge]

Good ideas , but it's a little annoying for two reasons the first one is, this eird things do'nt appear so much before the 3.5v and the second is, suno use the 4min length and direct download method like an argument in the update.

[broken lotus]

I am sharing what gets ME good results🤷‍♀️ Do what you want.

If it’s actually TOO DIFFICULT to use a tiny dropdown to switch the model on the fly… idk if any advice could ever be helpful.

[cursive jetty]

quick work around is start the extend section pre noise/gate threshold before it is noticeable otherwise just compound 🙂

[floral pollen]

Hi guys, I generated a song and I really really like it but then I realized I misspelled something in the lyrics. Is there a way to edit just the lyrics and keep the same music?

[prisma wraith]

not yet, I believe there's going to be inpainting in the next version. so you might be able to fix it then

[floral pollen]

Cheers friend. Thank you 😊

[oak path]

Hi
there's a way to select the same voice for another song?

[vestal rapids]

no

[brittle minnow]

hmm, is it just me or is 3.5 generating better song structures and melodies, with less hallucinations and mistakes, but with an overall quality that is worse than 3.0?

[swift moss]

ye, 3.5 is bad

[charred merlin]

Gotta love the depth of technical discussion.

It's objectively good, but might not produce the results you subjectively want. Perhaps you're not exploring and experimenting with the tool in the right ways?

A couple days ago I was making jokes about how 3.5 wasn't capable of making what I wanted, and needed to use 3 to get non-poppy sounding results-- I don't know if the model has been updated as early-access continues, but it's absolutely blowing previous content out of the water. I've seen a lot of people combining the different models with extensions to use the structure from one and the style of another.

Some wildly creative folk out there 🙂

[vocal cosmos]

still getting 9-45 sec smh waste of credits

[vestal rapids]

instrumentals?

[unique stone]

v3.5 does generate some interesting things. Just gotta catch it before the point where it starts getting noisy. Which is unfortunate when it does something different at a later point and you want to keep that bit 😄

[vestal rapids]

I'm not getting the metalic ringing as much now, have you seen an improvement?

[unique stone]

I need to test some more really, I've just been busy trying to export and edit stuff I want to work on. It still does this but perhaps not as early on as it was.

[vestal rapids]

what genres do you like to do?

[unique stone]

It really varies. Mostly electronic stuff I guess, but then I swing towards real instruments from time to time or even choir stuff. I got a choral drum and bass the other day which was quite interesting.

[regal fiber]

Hi! :), I've been working with Suno for a long time and have a subscription, and also realese official , I think it is possible to improve the quality of the sound and the supply in "psy trance", the kick is very spread in most generations, and the psychedelic sounds are very few, we need more supply for this style, for example rock and metal is amazing, But in psy trance I would like to see more improvement if possible.
I find myself doing so many generations until there is some good and basic rhythm to the style, it can reach 5000 for one track, if you can put more emphasis on this style, it will be really helpful.
It seems that the AI ​​needs to learn more references from Psy trance tracks from the leaders in the field, and of course they have subgenres such as Progressive psy trance, Goa Trance and more, which it would also be amazing if it improved and learned them, so that we could create such tracks.
🙏 ❤️

[vestal rapids]

nice, I have been trying some sythwave the past couple of days, 3.5 seems pretty good at it

[prisma wraith]

I've had the ringing appear in the last 6 tracks I've generated, and it appears to always show up if it's a 4 minute long generation. However in all of these generations, the ringing started at various times. I think the earliest was at 20-some seconds, and the latest at 2:10

[vocal cosmos]

yes instrumental

[vestal rapids]

yeah I am having issues with instrumentals being too short

[hollow hare]

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

With the tags, it avoids hallucination and pads the time (as well as giving you control on how/what instruments/melody you want)

Examples:

• RayanAIR's Novo Lands which shows how flexible these tags are
• My generic catch-all tags I use in v3.0 to pad random instrumental pieces based on the wiki entry for the instrumental tags: #🪲┃bugs-web message

[hard topaz]

I gave it a shot and added 60 different tags [... ] mentioning the instrument in each. It still only created a 52 second song.

[obsidian tendon]

Question about "make a song from any sound" annoucement. Watching the clip, my take away was I can sound out the melody or beat and Suno will take that as a starting reference to add the musical style to the timing and generate the song. Is this accurate?

[vestal rapids]

do you have the link for the clip you are talking about?

[obsidian tendon]

Yes, it can be found in #📢┃announcements https://x.com/suno_ai_/status/1794367911408353349

Suno (@suno_ai_) on X

Make a song from any sound. Coming soon 🎧

VOL-1: A watering can, but make it psychedelic rock

Twitter · May 25, 2024, 2:00 PM

[vestal rapids]

ok yes I saw that, I thought you were referencing something new. The make music from a sound feature is more like act as an influence rather than being able to continue actual music

[obsidian tendon]

Oh okay. Okay so it will not pick up on tempo.

As intructions but influence

[graceful condor]

@unique dome @graceful cliff 3.5 noise distortion in all tracks and at the end and also does not say lyrics clearly and also does not respect the genre you want and on instrumentals you cannot specify the exact instrument you want it always adds drums will not do solos also need 6 minute tracks for v4 also hip hop and rap music is crap also need god like ability to send in any mp3 that is a minute long or more and have the ai extend and build upon that existing first mp3 song or sound

[graceful cliff]

thx I know

but if you ping me again because of this I will have to ban you

[graceful condor]

the build on mp3 is what we need too that would be amazing

😉

[graceful cliff]

yeah we have it, but if you keep on pinging me again I won't let you test it 😉

[graceful condor]

i am making my violent call of duty FPS with chat gpt sorry for disturbance need to buy Ferrari

[charred merlin]

Every time the devs get unnecessarily pinged, v4 gets pushed back another day 🥺

[graceful condor]

well v3.5 was a dumpster fire so v4 needs to take time

LOLOLOLOLOLOL

[unique stone]

Days since a developer performed a table flip in response to being pinged: 0

[unique stone]

v3.5 hasn't even been properly released yet. It's early access.

It's there if people want to test it. Testing is optional.

[graceful condor]

no its for pro customers and since its so bad i suspect v4 will be launched before final v3.5

[vestal rapids]

it's not terrible

[graceful condor]

v3 god like though compared to v3.5

[obsidian tendon]

Question, , I like to use Suno to help me generate from pieces of music I wrote that I've hit a road block on. When I find one I like, use another AI tool (or soon now will be AI Agents) to extract just the guitar track. Then feed that to another AI tool that will then listen or read the raw audio data and transcribe it into tabs. I can then learn and play in the real world. Thus bringing something from the digital world into the physical world. Anything like that possible today?

[broken lotus]

all of the mods regret asking Suno Devs to make 3.5 available to members. That was our mistake.

We certainly won't recommend they ever do that again.

[graceful condor]

no this is good i like when the community tells the devs about how bad this version was so v4 will be good

[charred merlin]

Now you've gone and poked the hornet's nest 😛

[broken lotus]

not directly through Suno.
Stems and midi have been requested. If you'd like to also add your voice, please post in https://discord.com/channels/1069381916492562582/1154587937611190354
Devs do read it

[unique stone]

I was thinking back to my suggestion of making v4 available as alpha asap and realising it'd probably end up with the same reception even if it had loads of warnings about being buggy etc.

Actually I think I have a follow-up suggestion, thinking about it some more

[charred merlin]

Hopefully the early access experiment with v3.5 is a net positive after you take all of the unhelpful comments out of consideration-- unfortunately there will always be some who use beta programs as a means of "getting the next version before everyone else", who won't contribute to any form of expected back-and-forth that alpha & beta programs are intended for, and just complain about bugs & quirks 🙄

[broken lotus]

if you aren't using the thumbsdown on specific clips, you are just complaining.

[graceful condor]

this is a global issue down voting on specific clips does nothing if all the generations are bad

[manic vapor]

i feel i have had more success with v3.5 than v3, in my usage

[charred merlin]

Dude, at least explain your reasoning instead of just saying "it's bad". Not only are you objectively wrong, you're subjectively silly.

[graceful condor]

i gave very specific details scroll up

[broken lotus]

I'm not the dev, bro

[charred merlin]

"I'm not smart enough to be a film critic, maybe I could be an art critic-- "this picture is bad", or maybe a food critic, "this muffin tastes bad"." - Andy Bernard

[graceful condor]

tony saw it he is a dev

[charred merlin]

500 credits are the extra credits that roll over between, sounds like you must be at the start of a new subscription period

Like every other credit-based subscription service? yeah.

[hot fog]

a question, can I repeat the same melody and the same voice with different lyrics of a song already made?

[covert fjord]

Beginner question: Does it make any difference (better or worse) if I type [Verse 1] [Verse 2] etc. or do I need to/should I only write [Verse]?

[manic vapor]

might just be my own bias, but using [Verse #] seems to prevent the AI from reusing the previous verses' lines if they are similar, but i havent done many without number the verses, so could just bias

it does on the payment page.

[charred merlin]

Correct me if I'm wrong, but it sounds like you signed up for a subscription, got the big pile of credits, then didn't continue the subscription? So you have access to the extra "roll over" credits, but expected the bulk of the monthly credirs to remain indefinitely,right?

[manic vapor]

i'm not particulary joyed about it either, but its not like they didn't warn you

also the pricing per credit on top-ups is wildly different than subs, but i guess since topups dont expire its a tradeoff

[charred merlin]

I've found that numbering the verses can be helpful if you want to differentiate styles between them, or recall a previous verse's motif

[unique stone]

It's on the "Subscription" page, underneath the subscription choice which state that the credits renew monthly.

[covert fjord]

In one song Suno generated verse 1 verse 2 verse 2 verse 3 even though I had written verse 1 verse 2 verse 3 so I thought maybe the numbering made it misunderstand

[unique stone]

Be aware that if you do this, it may affect the license of the songs you created during your subscription (i.e. you'll no longer "own" the songs.)

[obsidian tendon]

Thanks posted in #💡┃suggestions-web

[unique stone]

Additionally, Suno's biggest competitor handles subscription credits in the same way, as does at least one AI image generator that I know of.

[obsidian tendon]

Any thing generative uses a credit system, its become the standard adoption.

[charred merlin]

Exactly, “use it or lose it” for subscription services.

You’d feel silly calling up Netflix and demanding free service because you signed up for an account but then didn’t bother watching.

“You took money from me, and I didn’t make use of your services during the agreed upon time period. This is illegal. I paid for 30 days of service, and I expect to be able to decide when those days are counted.”

[hexed epoch]

In 3.0 it seemed much easier to blend genres mid song. I tend to do a lot of drum and bass/dubstep/metal, but I like them to transition into each other to make distinct sections of the song. In 3.5 the generations seem to be locked into a specific genre once the first one is done. Also 3.5 doesn't seem to be able to figure out what things like a breakdown, drop, build or half/double time are. 3.0 was able handle that with using it in the prompt. Is there a specific way to prompt these things?

[broken lotus]

maybe try using [metatags]...(?)

v3 wasn't very good at them (seems to apply them all at once, sometimes out of order), but v3.5 seems a more reactive to tags, and less so the style prompt.

[sacred cipher]

It still sings them sometimes, if the tag isn't "common"

like [dark solo]

it likes to literally sing that

[hexed epoch]

I'll give it a try

[broken lotus]

[husky cabaret voice]
🎵 Daaark SOOOOLOOO 🎶

[hexed epoch]

hahaha this is exactly what I got. I did [drum and bass build] and the music cut out and a dude randomly just said it then it went back into the music.

[sacred cipher]

exactly like that lmao

to avoid this try to use more common tags

which ones? idk, just try lol

[manic vapor]

you can also try stacking them

[Build]
[Drum and Bass]
...
...
..!

[outer imp]

can we find an official SUNO video tutorial somewhere?... with more details how to improve a song in SUNO?

[charred merlin]

not a video, but this has been helpful for me https://www.suno.wiki/ (pretty sure this is an unofficial Wiki)

[obsidian tendon]

I'm wondering how Suno interpets musical annotation?

I fed it this and can somewhat hear the structure but it's not clear

[verse]
[e]
[h]
[A A7 D H7]
[e]
[h]
[Cis7 Fis7 h Fis7 h]

https://suno.com/song/fa72862f-be9d-4410-a0a9-99239d976899

[agile veldt]

The genre makes it hard to hear if it worked out or not.

but it sounds to me like it's the same chord all the way

oh wait, at 1:37 there's a chord change!

[obsidian tendon]

It sound slike the same chord but there are changes. The bracketed letters like [e] and [h] likely indicate notes on the staff, possibly for a melody or bassline. The bracketed chords like [A A7 D H7] show chord progressions, likely for guitar or piano. <<<LLM's analyzation of it

[obsidian tendon]

Yes it seems to transition but without a tag like [chorus] possible just continuing or making a new verse?

[agile veldt]

i have a feeling it's gonna be 3% hit and 97% miss tbh.

but then again, noone knows what the meta data used in the training data contained

so if it works just 50% of the times, it means you're on to something

[obsidian tendon]

Agreed. Hoping to find ways to better fine tune the prompt or how to break down the generations due to the percentages.

[agile veldt]

most of your progmetalcore song is in B minor (which you didn't ask for), and the chord change at 1:37 is: Bmin, Gmaj, Emin

[obsidian tendon]

May I ask how you dechiphered that? A future state idea I have with Suno is one where I can take the instrumentals I generate and get them tabbed through another tool so I can learn how to play the tracks myself.

[agile veldt]

i just played the same chords on a piano

to double check

[obsidian tendon]

Ah got ya

[agile veldt]

So I tried it too: https://suno.com/song/b1bc017b-500c-4ac9-8eff-d1ddf686ee9f
Actual result: [Am Bm Am Em]

https://suno.com/song/c28477ef-e799-424c-bef1-cd9413960fbd
Actual result: [F#m Bm E A] ........ again quite wrong! 😄

seeing how weak the tendency to follow these instructions is – when almost nothing else is specified – tells me that Suno doesn't understand this.

[hexed epoch]

I've tried that, but stacking tends to cause the metatags to become lyrics. it would be nice to have a song structure tab separate from the lyrics tab.

[agile veldt]

that happened to me too. 😄
" 🎶 Instrumeeeeennnnntalllll brreeeeeaaaaak doooowwwwnn!!!! 🎵 🎶 "

[hexed epoch]

it gets me every time. I know to expect it, but I still laugh.

[manic vapor]

we just need a little in-paint

i hope thats something before the long wait for v4

[agile veldt]

why even inpaint if the quality is this fuzzy anyway?

[undone star]

What long wait? V4 might be here summer.

And number one priority needs to be audio fidelity imho.

[agile veldt]

yeah, i don't see the point of inpainting if it's just mud inpainted with more mud

[wintry saddle]

It's a complex black box (or series of boxes) we may well need V4 to improve the fidelity

[wheat rose]

suno can't even get "male" right 80% of the time so expecting specific notes seems far fetched 🙂

[obsidian tendon]

^ True ha

[nova ledge]

BUT, you can change the image and make the song longer!

[wheat rose]

I had high hopes that i'd get a single generation of a full song with 3.5 but I haven't completed a song since they made it available

[charred merlin]

Is Chirp the reason that Suno has an obsession with generating birds with arms playing guitar for instrumental pictures?

[glacial scarab]

it was the original name yes. And the birds are placeholders playing on that yes

[vale isle]

That's what I'm seeing as well.
Lots of distortion with 3.5

But I really am glad to be able to have longer songs

[sharp kraken]

I flip back into 3.0 to refresh

like do a :15 instrumental

then extend it back into 3.6

[wheat rose]

I'm for sure seeing fewer generations ruined by losing the song structure, deviating from quatrain lyrics... but throwing them away for other reasons

[sharp kraken]

I get so much of this because im always pushing what you can do with prompt shifts and eventually the entropy just kills it

Im still working on a formula for what you can do to refresh it

[wheat rose]

I end up combining unnecessary punctuation and low syllable lines to keep things on track but it really limits what you can get out of it

[sharp kraken]

[Reset All] [Reset FX] ??? Something about EQ

[wheat rose]

The drift towards whitenoise I'm lucky if I get that far lol

[sharp kraken]

Man I can Go from electronic to classical into rnb with a gangsta-rap doom metal finale

[wheat rose]

Thats great if you are doing it on purpose

[sharp kraken]

Yeah - I use Cross-Genre / "" Influence and the Tag [Genre Changes to ""]

[wheat rose]

I need to try more verse prompting, maybe specifying male vocalist there will work better than in the style

[sharp kraken]

For instance, the Genre I used here is "Freak Folk" - but this is not Freak Folk

https://suno.com/song/c15d8dba-5be3-4906-9bf6-1519591bf8b9

[sharp kraken]

OOOhhh i know a lot

[wheat rose]

Annoying that sometimes the prompt is turned into lyrics BRIDGE!

[sharp kraken]

So you want to use BOTH for sure. There may be factors you want on ALLMvocals right? like male, female - whatever, put those in style

then for each verse you set the tone

[Uplifting Fast-Paced Bridge]

and you can specify a vocalist - but it a little wonky, it works better if you use a vocal break before

like [Vocal Break]
[Female Street Rap Vocal]

At least for me - Suno only understands it as a Vocal or Vocals - no singers or rappers

always use Vocal to talk about Vocal tracks

[flat robin]

Me too. Wasted a couple dozen generations trying to get a rapped bridge.

[sharp kraken]

You can do it tho

When you goto extend it, add "style" Rap Influence

into the style prompt

then tag the verse with the style of rap you added

[wheat rose]

I will give that a try. I'm not trying to switch between male and female, just it ignores even if I put several male callouts in the style

[sharp kraken]

i.e. Jazz, Freaky Rap Influence, Female Jazz Rap Vocal, IN STYLE
[Freaky Rap Verse One] IN LYRICS

[sharp kraken]

https://suno.com/song/eff399de-fc43-4a34-aa67-f55212964118

[tawdry bramble]

Anyone seeing issues downloading? This just started with .WAV file, I download and it generates 5 or 6 files, all exactly the same

[sharp kraken]

YES

[tawdry bramble]

Oh good, so it is a bug and not just me

[young orchid]

3.5 gives high-pitched sounds around the 2200hz band. Especially when I do drumstep.

[graceful cliff]

again plz help us downvote these clips!

thank you

[young orchid]

I am already doing it. I just wrote it as a warning to other friends here.

[unique stone]

Just curious: Are the upvotes/downvotes automatically having some effect or direct influence or do they just get manually reviewed?

[steep talon]

prompt cheat sheet?

[hollow hare]

https://suno.wiki

[sharp kraken]

use the .pdf

PROMPTS FOR AI DANCE MUSIC _ 🙏mista.lewys.pdf

[eager topaz]

This effect started for me around 25th April and reached a peak around 6th May, continuing to make my Instrumentals unusable and adding the effect to vocals. Now no improvement with v3.5.

[solar briar]

hi guys was just wondering if its somehow possible to make vocals without the beat with suno?

[sharp kraken]

i think you can use Acapella as a style right?

Ive never tried it

[untold briar]

You can, but the voices will be in the musical style of a cappella too, not quite the same as just a voice without music behind it. Try checking Suno search to find examples.

[hollow hare]

You can look into Bark, the text to speech system Suno AI made which is incorporate into Suno:
https://huggingface.co/spaces/suno/bark
https://github.com/suno-ai/bark

If you want more of a speech thing in Suno, there are [Announcer], [Narrator], and [Spoken Word] and their variants (e.g. female narrator)

https://www.suno.wiki/faq/metatags/voice-tags/
Old metatag list: https://suno-ai.notion.site/Using-Metatags-34944efe09ec41d693e314f13af44695

[solar briar]

ah thank you thats actually very useful

[zenith depot]

I want Ok Google to play music from Suno or Udio just by talking.

[violet anvil]

Hi guys, has it been discussed about editing a portion of a song, for example the whole song is good, but from 1:35 to 1:38 it mispronounce a word?? and I want to edit just that?

[graceful cliff]

you will be able to, just not right now

[thin flint]

Interesting. How'd they get the tempo change midway through this song? https://suno.com/song/296b9f3b-3e36-46d6-94c9-1d64337e3c7c

[plain shoal]

quick feedback of 3.5 after one day of usage with a pro subscription
Sound quality is super impressive,
I haven't seen yet a song generation of 4 minutes ( I mostly do instrumentals)
Doesn't really understand/apply the style prompts . When I ask for drill I get military drill time music. Does not recognize amapiano either.
i downvoted the clips
I think I'll keep using V3 for now because I waisted too many credits trying to figure our 3.5
Still love SUNO. though !! Hope this feedback will help making it even better.

[serene viper]

So not sure how to report this. I extended like two times already with v3 and they are supposedly generating when I go back to library, they dissapear.

[serene viper]

AI Gods on my favor.

[cold zenith]

3.5 ?

[final hazel]

Using an "unofficial API" for Suno is against the TOS. So it's not only the wrong place to ask, I also ask you to stop using it altogether.

[cold zenith]

damn, i just didn't come one week and i got this ??? Wow !!!

[wild pine]

is there a way to get actual specific sound effects in a song? I tried [sfx] and (sfx) but they dont work

[trail badger]

I get a lot of static noise in songs. This is very disturbing.

[final hazel]

It's not perfect yet. It belongs with the current state. V4 is around he corner though (4 to 5 weeks from now). And it gets better with every new version.

[trail badger]

I'm sure of that. Thanks

[crimson terrace]

do the wav files in 3.5 have better quality or is it still the same low fidelity we always hear on the web page ?

[unique stone]

It should be slightly better (high frequencies aren't cut off) but it won't "fix" any distortion or scratchiness that come from the model. Try it?

(also even v2 and v3 can be saved as WAV)

[crimson terrace]

Thank you, I don't have a paid plan 😁

[unique stone]

Ahh sorry, I should've guessed

[tidal sequoia]

Will I able to upload my song to create similar one with v3.5? If I go with the pro or premiere plan, will i have access to v3.5 model? I lost my old song which was important for me. I want to remake that song or is there a way to find old one (I haven't deleted it. All of my old songs have disappered)?

[broken lotus]

did you log in to the right account?

[jovial widget]

whats the best way to put a pause between sentance?

[broken lotus]

I break up lines and use "..." to make a small pause:

there once...
was a girl...

a longer pause might need
[break] or [pause]
or just some un-singable lyrics:
•–•–•–•–•–•–•
–•–•–•–•–•–•–

[jovial widget]

Thank you

[tidal sequoia]

Yes

[broken lotus]

if you have access to your old songs, v3.5 can extend on them.

there is no option for us to upload audio to copy.
but Suno has teased this feature 'coming soon' without any additional info... #📢┃announcements message

[tidal sequoia]

We are ,as an agency, ready to get to professional plan to reaccess that song. It is important for us because we urgently need to use it. Is there any way to get that song back? We have a copy of it but we want to extent that.

I kindly seeking for your help. Thank you

[jovial widget]

What would be cool is we could upload our own vocals and it would make the music to go with it

[broken lotus]

do you have a suno link to the original song?

[tidal sequoia]

Unfortunately

[broken lotus]

I don't know how to help you then..., just pay for a subscription and recreate the song. It won't be exactly the same, but you will have the rights to use it commercially.

[tidal sequoia]

Thanks

[charred merlin]

@moderators, if I'm making general discussions about technical aspects (observations of differences in vocal capabilities between 3 > 3.5, for instance), is that appropriate here, or better suited for the general channel?

[final hazel]

You are in the right spot 🤓

[charred merlin]

As it's tangentially related to technical aspects, I'd like to share my absolute new favourite game in the world-- Suno Roulette!

The rules are simple;

• enter random lyrics (<50 words for best 'game results')
• add [gibberish] at the end of your lyrics, or wherever you'd like to insert points of repetition
• set the custom style to "experimental"
• Generate!

I've spent more time than I'm comfortable sharing making generation after generation, marveling at how wildly different they are in every aspect. (and almost making my wife pee her pants laughing in the process)

I have a big chunk of credits that expire in a couple days, so I'm going to be playing Suno Roulette for a while now, wish me luck.

[rocky oak]

ugh question : i've publish an instrument sound in my profile and it visible, but i cant find in "create" menu or library. any suggestion ?

nope cant find there too. i want to extend it

[charred merlin]

I misunderstood, you can't find the song -- I thought you lost the "create" and "library" options.

Close your browser, reopen & see if it's there? Shouldn't need to go as far as clearing browser history & data, but perhaps check Suno from another device to see that your creations are syncing?

[rocky oak]

done, but still 😦

i cant find the correct icon in create nor inthe library. help ?

[broken lotus]

if you can get to the song's page, use the [..] menu and choose Extend

[rocky oak]

i cant find it in my phone too,

[valid flax]

does anybody know if the WAV is actually uncompressed original AI result or decompressed mp3 file back to WAV

also both mp3 and wav says that the sound is 48khz tho some songs has some weird spectrum limits

i noticed that in FL studio when i am often post processing the songs from suno

[rocky oak]

ok thx, ill use this temporary. so it was a bug the original song not shown in my library

[valid flax]

i can do full spectral analysis tho i am afraid the result will be off per song

[broken lotus]

the AI is the thing that is limited, as the Dev's have always said.

Don't expect significantly better sound quality with wav. MP3 compression is NOT the source of the audio limitations.

that said there may be some extra info in the wav at very high frequencies (potentially meaningless without equipment that can reproduce them)

[valid flax]

oh i see now that's a quite interesting information

[charred merlin]

Sometimes it doesn't show the "extend" button if the song is still in the process of generating. Off the top of my head 1) could be a heavy load on the server at the moment, 2) you're using 3.5 and it's attempting to generate one of the longer songs (~4 minutes), 3) your internet might be slow

[rocky oak]

nvrmd thx, it was in page 2 LOL

yes my bad, twas in page 2 haha

[broken lotus]

I lost a whole playlist that way... hahaha

[rocky oak]

🤡

[unique stone]

The WAVs seem to go up to ~20KHz, last time I looked the MP3s were more like ~18KHz I think? The classic effect of MP3 compression is that it makes stuff sound progressively more watery as the bitrate goes lower (so 128kbps and below is very noticeable.)

Higher bitrate MP3s still lose some of the definition but won't be resulting in the noise/scratchiness that come from the AI itself. Essentially you get imperfect quality from the AI, and then some further imperfection from the MP3 compression. So having the WAV means you get as good quality as you can get from the AI, without losing anything else.

[broken lotus]

If I want to remaster or edit the song I will download wav...
but for listening, i don't think it's worth 10x the hdd space per song.

[charred merlin]

🤖🦻🎧

Somewhere, a sad argumentative man is preparing to argue that every kbps above 320 is clearly discernible, and when presented the options, one should always choose lossless.

[unique stone]

It's reasonable to want lossless formats. But yeah I'm happy to just listen to the MP3s in most cases.

[spare comet]

Hi, have anyone else the Problem, that instrumental songs are not instrumental since the last update?

[valid flax]

Well i sometimes use another AI to isolate songs into stems

songs with bad quality often causes to AI badly differentiate between voclas and other parts of the song

which results in incorrect separation

[vestal rapids]

stem separation in the latest version of Logic Pro is amazing

[valid flax]

i am using FL studio tho i never used its stem separation ... i used some close sourced stuff with some fine tuned models for certain genres etc.

[vestal rapids]

I use FL Studio to make my music videos

[broken lotus]

v3.5 is trained to sing... like, it REALLY wants to sing.

For instrumentals I am still using v3... but that might change as I learn 3.5's prompts.

[charred merlin]

I’ve only had a few “lyrical hallucinations” I’d be curious what style prompts seem to be causing people to get them more frequently?

[valid flax]

It also really wants to play on electric guitar

its favorite i guess

[charred merlin]

I’m purely speculating at how the songs are generated, but I feel like “the vocal melody” gets “diffused” into a specific instrument , like piano, electric guitar as monty just said, etc

A few times the hallucinated lyrics are very simplistic and resembling Simlish from The Sims (2000) with very basic rhymes and repeating structures, and I can imagine that being the starting point for adding a slight bit of noise to then diffuse with the desired lyrics.

Or it was easy enough to assume it worked that “simplistically” until I saw how 3.5 was able to adjust lyrical flow, I literally have no idea how it pulls that off so well.

[low scarab]

something like this? Seems to cut it always right after the last [gibberish] tag.

https://suno.com/song/06c062f0-795c-417b-99d3-8bbc5bf13876

[charred merlin]

Ooh! I haven’t played around with inserting “experimental” into other style tags, I’m still wrapping my mind around how “experimental” works as the only style tag used.

I’d like to imagine that it’s some sort of magical tag that will insert a lovely degree of randomness into your song, but I’m more inclined to think it’d be outweighed by the other tags.

[valid flax]

Yep ... sometimes the saxophone in the songs starts saying things ...

[charred merlin]

I feel I should note I’m just an enthusiast; [gibberish] itself seems to insert a degree of randomness; sometimes it encourages the lyrics to repeat after they’ve finished, sometimes it causes stuttering, repetition of specific lines, or wild pronunciations.

Only using “experimental” as the style tag, a recent lyrical prompt I used was

“Oh my god, there’s a spider! (Oh my god!)

There’s a freaking spider!

I hate spiders! [gibberish]

To be honest, I’m not comfortable with spiders and I was a lot more comfortable just a couple seconds beforehand when I wasn’t aware that there was a spider in the room.

Maybe it’s true what they say about ignorance being bliss when it comes to spiders. [gibberish]”

[low scarab]

I see what you're getting at 🙂
https://suno.com/song/5c69959b-7ab6-413b-9f4d-c2ebf2836ebc

[charred merlin]

I wouldn’t know how to go about organizing the varying styles, lengths, degrees of “radio appropriateness”, etc., but none of it WASNT interesting.

[charred merlin]

Yeah, that time it seemed to ignore most of the uses of [gibberish] except for the last instance when it kept repeating “generate”. (From my experiments, generating that same prompt should give different results in terms of what [gibberish] does.

Pretty sure if you exported it as a video (where it highlights the current lyric), i wouldn’t be surprised if [gibberish] is highlighted when the repetitions of “generate” are playing.

[charred merlin]

My running theory is Suno is just an evolved version of Frampton’s Talk Box, and the devs haven’t learned how to shut the damned thing up.

[valid flax]

proly yes

[remote cedar]

site's not fully loading for me. spins instead of populating the song list. tried relogging, deleting cookies, disabling ad blockers, etc. is it down right now?

[valid flax]

can be cache issue

edge cache to be more specific

try it on your phone

[remote cedar]

I'm on brave

i'll try another browser tho, that might be it

[valid flax]

brave should be ok so check extensions

[remote cedar]

might be my account. chrome doesn't work either

explore works. create and library aren't loading any songs and home is blank

[valid flax]

check the dev console

sometime it will report the errors like if something is err 404 or 500 thats a server issue or issue of your connection to their edge server

[remote cedar]

oh weeeeird ERR_BLOCKED_BY_CLIENT

[valid flax]

doesnt mean anything

[remote cedar]

oh hmmm

[valid flax]

for example adblock will block their telemetry stuff and you will see this message but suno will work anyways

[remote cedar]

scrolled up a bit and there are others

[cursive patio]

Can we have more than 6 'structural' parts generated by the simple mode please?

[remote cedar]

it looks like it might be trying to bring up some faq that isn't loading

[valid flax]

check if your error messages doesnt contain any sensitive data and send screen

[remote cedar]

sorry if htat's too mcuh but i think that's all of them. it repeats the blocked by client errors after that, followed by some timeouts

lmk if you want me to expand one

[charred merlin]

To whoever on the team added the ability to add multiple files at once into a playlist, or pushed for that idea to be included, just know you're a good person and I love you.

[valid flax]

i dont see anything that bad tho

bad CORS can be an issue tho thats their issue or DNS cache issue

[remote cedar]

thanks man

i can TS a dns issue, running a pi-hole at home. if it's on their end i guess i can wait

[neat field]

https://suno.com/playlist/ca526d7f-8abb-482a-b398-68c7bcca7981 whispering lyrics seems rare to me, making a playlist to collect good examples

even lo-fi on the front page only has a few examples of "real" whispering throughout a song (vs a few lines) - which seems surprising to me

[sharp kraken]

do you know how to do it only one verse?

[cold zenith]

I've just done a test and I realize that the music in V3 was downgrading vocally, so if I extend the music and put it in V3.5, this corrects the problem to 85%. So the solution, if this problem persists in native 3.5, would be to make the music in V3 and then upgrade it to 3.5. Or upgrade to 3.5 and wait for V4.

[neat field]

sometimes [tags] like [whisper] or [quiet] will work
<lyrics and stuff [whisper]

but other times you will want to do

{extend the song}

[prompt 2 on extend section]
lo-fi anti lullaby
<lyrics and stuff [whisper]

{extend the song}

[prompt 3 - k-pop, metal, classical, etc]```

i find putting the whisper in the beginning of the song during [intro] or in the end at [outro] also seems to have more examples in the dataset and "works" more of the time

[sharp kraken]

Cool thats really helpful. Ive noticed it has a hard time with stuff that wasn't in the style of the prompt you extend from, even if you add it for the extention. Very interesting to finally see some syntax that works for <lyrics and stuff [bracket command] because if thats how it works it might help me with a bigger problem. Kudos

[neat field]

yeah the model will still have problems with some tags in some genres, my best guess is worse case ~~15% success rate (so 3x generations for 6x samples on average) for most clashing [whisper] - but my sample size is not large

[sharp kraken]

I made this goofy lil rap song about prompts hehehe
https://suno.com/song/846180ae-61cb-4b47-9914-8de906330c85

Might help ya i goofed off like 4000 sunos trying to figure that out

[paper lichen]

Песня про любовь на расстоянии двух людей

[broken lotus]

Пожалуйста, используйте веб-сайт сейчас
https://suno.com

[autumn skiff]

Pitching a few ideas with Suno with my agency, wondering where I can find the latest updates on when a public API will be available?

[broken lotus]

Suno has not announced an API. There is no eta.

[glacial urchin]

For some reason I upload an image as the cover image to suno and it just shows this

Does suno have an image file size limit?

[final hazel]

1024x1024 I believe (not 100%, but let's be honest, that's already overkill) and it has to be less than 10 mb.

[glacial urchin]

Oh mines 44mb

[charred merlin]

https://imageresizer.com/ should be able to trim that to a reasonable amount of display colors

[final hazel]

Great suggestion 😄

You know.. the simple paint in windows can do a lot too. Not that hard. Or any other software made to edit images. Most people have at least one on their PC I think? Heck, there are even simple apps for on phone.

[bronze helm]

It's nice to see (and hear) that .wav downloads have significantly better quality than what is heard in .mp3s and the website! Suno's known for the slightly robotic vocals but they sound better in .wav files.

[golden cove]

is it passible an existing v3 song to re creat in 3,5 ?

[final hazel]

no

[clever bloom]

Have you ever had an issue where you plot out an entire instrumental, yet the AI doesn't listen to anything you tell it to do?

[outer imp]

I noticed that in 90% of the drops, SUNO uses the same synth and that gets annoying... the same synth tone appears on the drop side!

[charred merlin]

The closest is taking a song created by v3 and trying to extend it using 3.5.

Especially if the extension point is far enough into the song, you can likely get good results attempting to recreate specific verses, etc.

[cerulean steeple]

Hello. Does Suno have official api?

[final hazel]

It's on the roadmap. Release/Availability unknown at this moment.

[cerulean steeple]

Thank you

[untold briar]

I haven't been able to try myself yet, but at least one user reported that if the V3 song is short-ish (not much over 2 minutes), if you Extend From the end of the V3 song with V3.5 with the same original song lyrics, it can be something like a remake of the original 3 song in 3.5. (Because 3.5 has two minutes of memory it remembers quite a few verses/melodies/etc from the original.) This is a bit of a long shot but let me know if this works.

[cerulean steeple]

Oh, another question: is there any way to tell Suno that "this" letter should be pronounced longer, like "a" should be "aaaaa" XD
(sorry if sounds weird, English is not my native language)

[untold briar]

People try lots of things that work, for example: aaa a, a... AAA _a_ *a* a-a ayyye but there isn't one correct way you are supposed to do it.

Wrong Discord 😂

[weary wharf]

Oh yeah 😄 Wanted to test this... I'll go to off topic.

[cerulean steeple]

so these are just the ways that sometimes seem working. Thanks I'll try

[golden cove]

thanks to everyone who try help 🥰

[final hazel]

No sorry. Devs told us that inpainting only works with V4. For starters at least.

[pseudo wind]

Is there a reason why the sound on the site are put at a low volume?
Songs come out at a really low volume.

[shadow ridge]

So soundcloud does not allow suno just tbc

according to what i received

from soundcloud

so question is why am i paying for a commercial licence for something that is not allowed to be commercial

even though i wrote the lyrics

its a fair question

[charred merlin]

I'd like to explore this in more detail, but in the scenario you provided, it sounds like this question would be better directed toward Soundcloud?

(tangentially related story to emphasize my point)
I remember when vaping first became a thing, someone came into my coffee shop and started blowing huge clouds. I told them they weren't allowed to do that inside, and they tried to argue with me that it's not the same as smoking, and legally they weren't breaking the same rules as if they were smoking.

I told them I didn't care whether they were breaking laws, it was my establishment and I didn't want that inside (not worth explaining to senior citizens that it's bubblegum scented water vapour).

Same sort of situation-- soundcloud is well within their rights to refuse to host AI generated music regardless whether you own the rights to the song, just the same as you have the right to take your songs elsewhere.

[vestal rapids]

Does Soundcloud prohibit AI generated music?

[charred merlin]

Haven't looked into it yet, just based on #📚┃suno-school message

[vestal rapids]

what did you get from Soundcloud?

[charred merlin]

I'll ping you if I find anything one way or the other-- without context it might have been a one-off?

[shadow ridge]

expicitity will not allow ai suno on soundcloud

[vestal rapids]

can you share with me the text of the communication?

[shadow ridge]

no conversation i just basically removed the album was about to decide on what to do its just one phrase soundcloud explicitly will not use fully ai generated content

and detected i did that

i shared everything i needed to approve my licence

[vestal rapids]

how did they detect it and how do you know they detected it?

[shadow ridge]

did not say

thay said it

not what they told me

they told me i could not distribute

ai content

[vestal rapids]

are you a free or paid Soundcloud user?

[shadow ridge]

paid

[charred merlin]

Reaching out to soundcloud support directly to get clarification

[shadow ridge]

i do premimum

figured since i wrote all lyrics it was ok

I still have all the source material for the 3 albums they blocked

albeit i released 25 other ones so idk what to say about this situation

[vestal rapids]

how many AI releases have you done with Soundcloud?

[broken lotus]

WOW, demand a refund from Soundcloud.

[vestal rapids]

this might be a new thing from Spotify as I know someone who has had the same issue with Distrokid in the last 72 hours and they now will not release any more music from her

[copper obsidian]

Speaker spanish?

[glad granite]

Hi yall, I am looking if anyone here uses Sonu to develop cool applications, like using it in Code am I at right place? 😅

[final hazel]

Like an api?

[shadow ridge]

like what a refund for 2 years of service lolol

my ai songs have been on distro for 3 - 4 years before suno idk what to say about this

one of my albums is ai cover of my own song with my own ai cover voice so its me i guess??

[glad granite]

Ye

I saw they have a text to speech api called Bark

So i am like playing with it rn lol

[final hazel]

At the moment Suno does not offer an official API for Suno yet. It's on the roadmap. And it's not allowed for third party to create their own Suno api. It's against their tos.

[glad granite]

Oh okay

Well i was talking about the Bark which yall made open source so 😷 😓

[final hazel]

Yeah that's more the depreciated model for text to speech. But not many people here know much about it. One of our mods does though, not sure if he is around, but his name is @untold briar . He might be able to help you with that.

[glad granite]

its deprecated! Damn thats crazy it is probably the best model I have ever used for TTS lmaoo

although quite slow

lmfao

[final hazel]

You can download it yourself local or play with it on huggingface.. and maybe there are some forks? I am sure you can find them on github 😅

[glad granite]

yeah its already up and running on my system

lol

[final hazel]

Well for more questions Jonathan is your man 😉

[glad granite]

https://tenor.com/view/copy-that-call-roger-that-copy-gif-17713322

[lone galleon]

How does the model handle vocal progression specifically?

I.e.
Burn.
-Ing.
Bright.

The goal for this phrase is to hang on each syllable for a beat, but I'm trying to figuire out how the model understands the lyrics for vocalization

Would the optimal be to have it as
Burn-Ing
Bright.
; Captilization after hyphon
Burn.
Ing.
Bright.
; Seperate with periods
Burn
Ing
Bright
; Seperate without periods

?

[untold briar]

Bark isn't dead, it's just not supported by Suno. But it's been recently ported to iPhone for example.

[final hazel]

Ah wrong wording for my side. My apologies.

[wispy zealot]

What's the difference between the WAV and MP3 download option?

[charred merlin]

@broken lotus @final hazel @untold briar (just tagging active mods in case this comes up in future discussions ❤️ )

[vestal rapids]

sheesh!

[broken lotus]

or computer generated...?

[charred merlin]

My wife pointed out that new uploads to Instagram now require you to indicate whether or not the content was generated with the use of AI.

I suppose we're at the "honor system" phase, but mark my words it'll only be a matter of months before any content being uploaded to any service will recommend that you indicate whether it was created by humans, as that will start to become less common. But that's just my 2-cent speculation

[broken lotus]

🤷‍♀️ that's ridiculous.

[vestal rapids]

thanks for following that up btw 🙂

[charred merlin]

Clearly has to be awkwardly worded, meaning zero human involvement, because there's a LOT of music that would be invalidated if that's their literal interpretation.

[shadow ridge]

I mean i cloned myself

i released with my cloned self and now

not

[broken lotus]

I mean I stopped at Soundcoud when it was overrun with clickspam...

But refusing to distribute algorithmic music?
Who are they kidding?

[shadow ridge]

i know

[vestal rapids]

I use Soundcloud to "store" music, I think I might rethink that

[broken lotus]

yeah...

I have algorithmic music on soundcloud... from a decade ago...

[shadow ridge]

this is a weird place for production

in general sence

sense

Your distribution release, Mako....Rehaps...., has been rejected for the following reason(s):

There is a problem with your release. Please review the notes below for additional information.

Our terms of service explicitly prohibit generative AI (or content that is purely computer generated). Unfortunately, under these terms, content created via services such as suno.ai are not valid for distribution via SoundCloud for Artists.

[broken lotus]

WAV is usually a better file format for production.
MP3 is best for listening and sharing.
(Wav is 10x bigger file size)

There isn't much difference in quality currently. The AI itself is the source of the music's quality flaws. As each model gets better, and Suno ups the quality on the downloads to match what the AI can output.

[golden cove]

can someone send me the link to the most detailed description how suno copyright works ?

[shadow ridge]

I remember when deezer started used python models to stem music wonder how that applies

deezer stems are in tons of stuff

[sharp kraken]

Does anyone know how to use time-stretching ?

Or like better yet - what is the culture around prompt sharing? Like is it a faux-pas to ask such a thing? I've notice a lot of information is... decentralized. ? intentional?

[wispy zealot]

Thanks for the info!

[clever bloom]

Can someone explain what I'm doing wrong?

Am I not using the tags properly? The AI doesn't listen to them at all:

[Intro]
[Somber violin solo]

[Verse 1]
[violin, cello instrumental, minor key]

[Pre-Chorus]
[Build, forceful tremolo strings flourish]
[Distorted synth melody, fade in]

[Chorus]
[strings, distorted synth, swell]
[Taiko Drums, thunderous, pounding rhythms]

[Build]
[strings, swirling, ascending]
[glitchy, arpeggiated synth]
[Drums, taiko, industrial, driving groove]

[Drop]
[Pause]

[Verse 2]
[pounding, distorted kick drum]
[strings, pizzicato, dissonant harmonies]

[Bridge]
[string, furious, spiccato]
[Aggressive, atonal synthline]
[Taiko Drums, explosive, driving]

[Build]
[strings flourish]
[Drums, taiko, intensify]

[Intense Finish]

[Outro]
[Ambient Decay]
[soft, somber solo violin]

[sharp kraken]

[I would do it more like [Taiko Drum Intensifies]

I have alot of success going like [ "adjective" "noun" "verb" "target"]

[glad granite]

Being unbiased, do u know any alternatives to Bark like really good ones I can use with python?

[clever bloom]

I see. Anything else you can spot?

[untold briar]

There's a few good ones that good but not actually open source (can't be used commercially which arguably prevents even using it in a YouTube video with ads).
I've been watching the progress of this one a lot:
https://github.com/collabora/WhisperSpeech

[sharp kraken]

I've been picking through rave2

havent done anything with it yet - whole nother animal but kinda interesting looking

[glad granite]

Ok thanks both of yall. And i wont be using it for commercial purposes, only for my own knowledge 😅

[sharp kraken]

I think a lot of people started open source stuff using the FTT equations a few years back and then U-turned on that so theres a lot of Weird ai stuff thats practically abandoned on github

[glad granite]

yeahhh ikr

[sharp kraken]

Hehehee, - thanks for the FTT algorythm and realtime Music generator though

guess it helps when you know matlab too

[molten geyser]

Can I write a tag that stops the melody and leaves only the singer's voice?

[untold briar]

It seems plausible there are tags that do that at least occasionally, but it isn't a feature or something I've personally seen. Well, one related thing I have seen: if you continue past the very end of the song (after the fade out) sometimes Suno will randomly hallucinate a version of the chorus that is voice only. Not a common result, can't find a sample clip, should have noted it somewhere.

[molten geyser]

I was initially trying to use that feature in like between the song before the chorus, so I might try the fade-out tag for it

[narrow cypress]

I haven't spent much time in this channel, has there been any serious talk of implementing negatives in the future? (even distant?)

[glacial urchin]

I accidentally made a duet

[glacial urchin]

And I just accidentally made a 2nd one lol

[hushed sequoia]

Greetings. New to the tool. Only been using it a week. Anyone notice a degradation in quality the longer the song is? I'm using v3.5. To the point where the 2nd half of a song a lot of times is unusable. I can hear a stark difference.

[glacial urchin]

I have not noticed anything, it works just fine for me

[hushed sequoia]

Like the first half of a song with be warm and clear. But by time it gets to the end of the song it sounds like tin with artifacts

[eager topaz]

It depends on the genres, but yes, it's fairly widespread. Hopefully something that will soon be fixed.

[hushed sequoia]

OK thanks!

[sharp kraken]

Hey I accidently flagged one of my songs

theres nothing wrong with it

can I report it as misclick or something?

[untold briar]

accidental flags must happen all the time, i wouldn't worry about it, they must be looking at trends. it's not going to nuke the song from your library or anything

[sharp kraken]

ok cool - i figured it would be that way

[wet vapor]

I just tried 3.5 yesterday and haven't been able to download anything in .wav

[broken lotus]

are you on mobile?

[vivid charm]

found this alternative that seems to be focusing only on AI music https://www.mastro.ai

[restive totem]

Hello ! Whats your tips to have a good music quality ? Some songs on suno website sounds great but sometimes mines sounds with noises (V3.5) thank you 🙂

[distant basin]

Is there a way to shape the prompt in such a way that it is more likely to output the correct instrument? I'm trying to generate music that features the 笙 (Japanese sho) instrument (similar to the sheng and khaen), but my prompts never produce the right sound

[sharp kraken]

Does Suno provide help for addiction? I feel like im playing a slot machine that kicks out emotions

😉

[cyan thunder]

how could we see the song cover in high quality?

[final hazel]

First, I keep in mind that it's never perfect. If you want the best quality possible, you'll need to spend some credits. I personally use the highest plan (premium) and I'm not afraid to use credits. If I'm not getting the right quality after a few tries, I tweak my music style and see if that helps (sometimes less is more, or the other way around, depending on the genre). Once I find a decent generation, I start with that. If the quality drops after 2 minutes, I extend from a point before the degradation and retry until I get a good result.

[final hazel]

Here you find a trick how: #💬┃general-chat message

[dreamy dagger]

Hello i have a question,
I did a music and i wanted to know if it's possible to use the same vocalist on my futur musics ?

[final hazel]

Sadly no.

[dreamy dagger]

Alright thank you

[cyan thunder]

thanks !

[lone oak]

Although V3.5 has solved the prelude problem and improved the accuracy of the Style of Music tag, its own melody consistency is very poor. Although it can sing any lyrics, its abrupt connection is like Google Translate in a different disguise. I really think that the weight of the Style of Music tag should be lower than the weight of the information marked in the lyrics.

[hazy osprey]

hmm cant stop my song ... just extend with [END FADE OUT] track showes 20sec ... and at 18sec it will automatically extend to 2 minutes ... 60 trys ... 😄 suno wont stop my song 😄 hrhr tryed ... [Outro], [End] [Stop] [Fade out] ... no way to stop xD any one has the same prob ?

[sharp kraken]

Thats just suno havin a good time

[hazy osprey]

hehe jea ... i think suno is a fan of my song and wont stop party 😄 hehe

[sharp kraken]

Its much easier for me to anthropomorphise on Suno because I feel like it's talking to me sometimes, it's wild - how perfectly it hallucinations fit sometimes, like Suno had it's own Idea and its a good one. The way it yells your broken tags back at you - is absolutely priceless - so weird and yet useful at the same time

I feel like im working with Suno

[hazy osprey]

hehe, in any case, suno has his own soul and is sometimes a stubborn child who doesn't want to share his lollipop 😄 "I'll do it now whether you like it or not!" 😄

[sharp kraken]

@untold briar The "Compressed Mp3" has me frikkin dying over here man! I just cant turn it off, - only bc that voice is just - perfect. I download as WAV hahaha! Gut 1

[sharp kraken]

Or maybe thats us

hahaha, Suno knows best

[wet vapor]

on my PC

[broken lotus]

Did you wait for the song to fully generate?
it stopped spinning?

[undone star]

none of the social links on the site work. This seems like a poor placeholder.

and considering its hosted on squarespace does offer me zero confidence 😄

[vivid charm]

Yeah noticed that as well @undone star - probably a very early landing page which isn’t necenecessarily a bad thing though

[undone star]

Even if its early - at least do some quality control. Being first out of the gate only does wonders for you if you do a solid job. So far this site tells me almost nothing and leaves me with tons of questions and empty promises.

[vivid charm]

Yup! Imma keep an eye on it 🙂

[unique stone]

I just looked at that site. That's quite an obnoxious web design. Bright red. Also all of the photos have an interference-like animated fuzz to them. I thought my monitor was having an issue. Pretty sure there was a different website being promoted a lot for this kind of service recently as well, so be cautious as there'll be some bad apples out there just looking to claim music as their own.

[broken lotus]

Amazing Opportunity to store your lunch in my mouth!

Act NOW! this offer is for a limited time (until I get hungry again)

also for a small fee I will HOLD YOUR WALLET FOR YOU....

[undone star]

https://tenor.com/view/lets-go-bananas-banana-woo-gif-23231193

[wet vapor]

yes even went back to songs I made yesterday and still getting the same error with the .wav download

i'll try on mobile and see what happens

[broken lotus]

is this EVERY clip...?
or specific clips...?

[wet vapor]

the full songs

[broken lotus]

Do you see this popup?

[wet vapor]

yes

on the mobile it only gives me the option of downloading the .mp3 file

[jolly robin]

is there a way to play suno songs through a discord bot?

[broken lotus]

hmmm, ok.
so it seems the wav is being generated, but not auto-downloading when it's ready... 🤔 (that's what I see, it just downloads on its own when it is ready about 10-20sec later)

I haven't seen this problem before... I'm not trying to 'blame' your browser or anything, but is it possible you have popups blocked or something...?

[wet vapor]

the .mp3 downloads fine lol

it just gives me an error when trying to download the .wav

[broken lotus]

what is the error?

[wet vapor]

and I've tried my other browsers too

Suno.com says: Download failed, please try again

[broken lotus]

anything else I can relay to the Devs?
are you using a VPN, or anything like that...?

I haven't heard of other reports like this about downloads (wav or otherwise), so any additional info I can give is helpful.

[unborn hound]

I'm trying to find where in the wiki it talks about using symbols in the lyrics box. Like dots and dashes, ascii code, etc..
Can anyone help me out?

[final hazel]

There is no really information about that. Some people tried it, and had some interesting results, but it doesn't seem to recognize the symbols in general.

[final hazel]

Try relogging.. for some people that helped.

[unborn hound]

Ohhh 9k, that makes sense. Thanks

[broken lotus]

there's 1 little part about adding 'unsingable text' to expand an instrumental...
https://www.suno.wiki/faq/style-and-lyrics/how-do-i-make-an-instrumental/

Mostly people are putting stuff in experimentally and the bot uses it semi-randomly. It's not a 'feature'.

[wet vapor]

I'll try but yeah no VPN or anything like that in use. Thanks!

[wet vapor]

try relogging in but no success

[final hazel]

Can you please report it to #🪲┃bugs-web then. I know you are not the only one, and it might press some urgency to the devs. It's a weird bug since most people can.. so it's a whack a mole game, find the mole.. in which hole, if that makes sense.

[wet vapor]

thanks

[glacial scarab]

Symbols can also help suno "remember" song parts. So if you put a bunch of symbols after the intro for example under an [Instrumental] tag you should pretty reliably get that melody back later when using the same symbols. Some symbols might lead to vocalisations though which will probably be dependent on genre. I have not found symbols that trigger specific stuff. I use it more so that suno has something to latch onto in place of lyrics for structure as meta tags alone where not successful for me.
If you use too much symbols next to actual lyrics suno might even confuse the symbols to mean it should repeat the lyrics though 🤣
Here is one of the better examples of the above
https://suno.com/song/35a6ab8d-0ef4-4b81-a212-a48e57e8e655

It is however pretty random what suno will associate with which set of symbols. Finding out what triggers what is best tested with an extension that only has that symbol set.

[serene viper]

Great content @glacial scarab I followed you

[glacial scarab]

And it will only be the same in the context of one song (and its extensions). The next song might associate different stuff with the same symbols

[lilac garden]

Whats the number Id in the music style prompt for?

[glacial scarab]

its testing out one of yolkheads theories. Here is a replychain for reference
#📚┃suno-school message

[weary wharf]

Has anyone found a way to tell v 3.5 to slow down?

[broken lotus]

like to start with..., or during the clip?

[untold briar]

[speed change] I think used to work in V2 or V3? The other thing to try is adding more line breaks, literally make the lyrics in the slow section use up more lines (less words on each line)

[broken lotus]

there are also Italian Composer terms for it...
https://www.cmuse.org/ritardando-vs-rallentando/

[runic dock]

Downloading wav was not working on mobile for me.
I have already reported the issue.
A workaround is to use the three dots at the top right and change the display to desktop even if you are on mobile. It works.

[undone star]

same behaviour for me - sometimes it also downloads the .wav twice.

[weary wharf]

What happens is, all of the generations read the lyrics really fast. No matter what the style is.

[weary wharf]

Not working. Empty lines are passed, lyrics are merged, even though they are on different lines, ending with , . or ...
Less words do not work as well. The outcome becomes something else.

[cold zenith]

i hope the downgrade of quality will be fixed in the V4. Because it's my main problem.

[sick helm]

Is it technically possible to extend song that where generated using the Suno plugin in Copilot ? Because I generated a music I really love but it cuts in the best part at 1:00

[flat robin]

This is gold, thanks a lot!

[lilac gulch]

Hi everyone i couldnt find in wiki. I want to add nature sounds (bird call, cricket sound, ocean and wind sounds) to my instrumental song. Do you know how can i do that ?

[hazy osprey]

🙂 pm

Try to define those sounds in your stylebox and then try to use something Like [Intro] (birds are singing). Sometimes suno really knows what to do then 🙂

[west geyser]

what does "A clip got stuck" mean?

ive been encountering this error the last 10 minutes

[twilit veldt]

Same

[vestal rapids]

same error here

[ebon rampart]

I'm almost wondering if Suno's got some maintenance going on than

Sounds like everyone just all of a sudden ran into the issue

[west geyser]

starting to think so

yeah, general is sounding off about it too

[static hearth]

Yes we are lol

[hybrid linden]

yes i just came here to say that too. the credit got stuck issue

[static hearth]

Good luck friends, off to edit my downloaded clips then.

I see 4 trying to genereate now, and now 6, it may be coming back online.

[hybrid linden]

Shit. now it started creating all the songs ive tried, and its burning through credits like hell

[static hearth]

LMAOOOO

[west geyser]

HOLY CARP

same here!!!

[glacial scarab]

Happens every time to somoen when clips get stuck

[static hearth]

I sense a 500 credit care package coming tomorrow as a sorry.

[twilit veldt]

Uh oh, I'm about to have 20 clips or so. Lol

[glacial scarab]

I can Only recommend stopping using suno for a bit once clip got stuck occurs more than one time

[west geyser]

welp...this will take a while to sort through now lol

[final hazel]

Thank you for bringing that up.. The dev team is working on this issue.. Stay tuned!

[static hearth]

I see album art so it's trying. And good idea, off for now until it slows down again.

[weary wharf]

It's time to return to mastering then 😛

[static hearth]

Saaaame! Pivot to another project for a bit.

[final hazel]

It should already be solved.. still some jobs in queu so could be slow.

[weary wharf]

Oops... 😆

[final hazel]

yeah.. still the aftermath.. it's fixed.. just some queued jobs

[weary wharf]

I am probably going to create more, so no problem for me hahahaha.

[hazy osprey]

Same here 🙈 and also i cant get suno to stop the songs 😂 have to create 6-8 minutes songs if i want my lyrics done completely. First the extendend clip is Like 43 seconds and then jumps the time to 2 minutes. Do you have the Same prob atm? @weary wharf

[weary wharf]

They are coming in one by one...

[twilit veldt]

I have found that if I switch back to v3 for my ending and put [End] in the lyrics, it is more likely to end the song.

[west geyser]

welp, ever since the clips started getting stuck, every extended clip generated has either ignored lyrics and just made instrumental or has ignored the music types, the structure, pretty much made an entirely different song and form of music...poof goes 110 credits

[static hearth]

When it's busy like this the quality seems to drop noticably, all my generations are now pretty "corporate" feeling and less "creative"

Another issue is that they can start abruptly or forget lyrics, or the dreaded demon chorus. Time to hop off for an hour and come back.

[true gazelle]

Is there any way to make the chorus sung by a man and the verses by a woman?

[west geyser]

[Verse: female voice]
XX

[Chorus: male voice]
XX

• it has worked for me about 60% of the time

[true gazelle]

ok

[thick wraith]

I have the same problem every generation its degraded since the bug I think the servers are too overloaded

[hollow hare]

Getting duets... that is more luck than anything else (well, not as hard in 3.5 compared to the other versions from what I heard but still hard).

For 3.0, 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 voice's 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.

[static hearth]

After looking at 3.5 today, it feels "cleaner" and has less of that radio static sound, but since this hiccup I'm getting all the above with the ones it's generated. I noticed that I still prefer gen B most of the times still.

[hazy osprey]

Good idea 🫡

[true gazelle]

unfortunately it doesn't work

[hollow hare]

See my reply above if you missed it: #📚┃suno-school message

And sorry Twigs, I replied to the wrong comment earlier

[rare stirrup]

Is there a way to have suno make a specific sound at some point in the lyrics, like an airhorn or a beep sound like you are censoring a word

[nimble smelt]

How do I end a song that is instrumental?

[hollow hare]

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

I Prefer either Big Finish or Fade Out followed by End

[nimble smelt]

That only works in lyrics from what I have found

I’m doing instrumental only atm

Thank you tho

[hollow hare]

Ah, I see what you mean. I specifically avoid using the instrumental switch for my instrumentals.

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

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

Examples:

• RayanAIR's Novo Lands which shows how flexible these tags are
• The generic catch-all tags I use in 3.0 to pad random instrumental pieces based on the wiki entry for the instrumental tags (#🪲┃bugs-web message)

.
But since you likely have a piece you want, you can Extend at the cutting point and use [End] (or {end}) with the style "End, Short" iirc

[coral plover]

https://suno.com/song/4ec1b0e4-60e1-4a36-ba41-72f5aca4b4a0

every single time i try to make music box melody it does the most crazy things.

sorry for asking this here but does anyone know a good prompt for music box melody? i tried many times and managed to do it once but never again

[unique stone]

What's changed between early-access v3.5 and this one (aside from vocals)?
I assume v3.5 is now "frozen" and won't change any further?

[unkempt spade]

I can hear that it still does the super high pitched reverb sound sometimes - will that ever go away completely or is that something that can't be fixed? It's always a bummer when you get a really good generation and it's got a big SKREEEE over it

[azure crane]

IMPORTANT!!!
There is a bug after the last update when you create an Extend part. If you don’t refresh the browser, you cannot create a new track. Instead, you bring up Extend again and again…

[manic vapor]

and Clear doesnt work?

[lone galleon]

Is there a Build up type meta tag for song structure?

[glacial scarab]

wait wait there was an update to the V3.5 Version?

I might have to buy credits after all then

[azure crane]

Exact… Frustrating. They fixed the last Extend bug with the 3.5 update but with today’s update, the Extend bug came back in another bad way…

[unique stone]

v3.5 is no longer early-access and is available to all.
There was talk of vocal improvements.
I'm now regularly getting 4:00 long instrumentals without having to shove every known symbol in the universe into the lyrics box 😄

[glacial scarab]

aight imma buy credits

😄

[unique stone]

(slight exaggeration there)

Check out what people are making in #💬┃general-chat as well.

[eager topaz]

Can you share how you get 4 minute instrumentals in Lyrics? Mine are varying from 1:27 to 4:00 . Tks

[unique stone]

I was finding that I had to put loads of dashes or dots or whatever in the lyrics box. I no longer need to do this for instrumentals. Just turn the "instrumental" toggle to "on".

[eager topaz]

My Instrumentals in the toggle are varying from the same lengths as in Lyrics, just over 1:00 to 4:00. I'm just not getting many lengthy gens.

[unique stone]

Thrown a few random different genres and combos of instruments at v3.5 - audio quality no longer seems to degrade as much (and if it does, it starts much later in the song) so I'd like to say thanks for that!

[graceful cliff]

you are welcome

thx for labeling those noisy ones as thumb down

[unique stone]

Will the thumbs-up/thumbs-down still be used from v3.5 to help train v4?

[low remnant]

I'm a ruthless thumbs-downer. At least for noise. I'm more tolerant of lyrical hallucinations, because sometimes the singer knows better than me. I've had more than one song where I changed the "official" lyrics to match the ones sung. But noise = instant thumbs down.

[dreamy dagger]

There is a special command to stop immediately the music?

[weary wharf]

You can try [stop] or [end] but Suno does what Suno wants when comes to outros and ends.

[dreamy dagger]

Thank you 😄

[molten plume]

palio

[nocturne fable]

if i put at the begginging of the song. (Male Elvis Singer) or (In the Voice of Elvis Presley) Will Suno use a similar voice or style?

[rapid bluff]

[Instrumental Fade Out] works well for me too put directly after your outro

[weary wharf]

Question was about "immediately stopping".

[rapid bluff]

i know, sometimes suno thinks that means abruptly stop after the lyrics

[weary wharf]

Too random for my taste 😆

[rapid bluff]

finishing a song is always rolling the dice, really think it depends how your song is flowing up to that point

[weary wharf]

Yep. Sometimes it feels like a miracle. So good outros and endings.
Sometimes it feels someone just poured coffee over my keyboard, and something came up.

From time to time I get some clapping from audience and people screaming "bravo" or "bis" / "again"... I feel I'm special 😛

[rapid bluff]

hahaha honestly the creative add-ons suno throws in is what sets this apart from other services, its done that since day 1. More control and inpainting would be nice but I dont want suno to become too robotic of a model because of it

[graceful cliff]

well first we don't use it to train, second really depends on the label qualities

[graceful cliff]

at this stage I almost care more about the quality of the feedback rather than the quantiy of the feedback

so I may ask us to roll out some special groups, let's see

but keep your feedbacks with 👍 👎 , again, much much appreciated 🙏

[weary wharf]

Like, how can you say "no" to this intro and outro...

I disabled the vocals on outro, so first few seconds sound broken a little.

[rapid bluff]

goddamm sounds soo clean, didnt even notice it had a break in the outro

[weary wharf]

I am mastering most of the outputs. Or add my own instruments on top. Sometimes it goes a bit too "noisy" at the end, so need to correct that.
Takes time but the outcomes are clean.

[broken lotus]

we can't use artist names... but try
'rockabilly' as a style

you will definitely get Elvis-like voices...

[rapid bluff]

nice! what program do you use personally if you dont mind me asking?

[weary wharf]

Logic Pro. For instruments I have Native Instruments (came with the kontakt keyboard) + East West subscription.

[rapid bluff]

ohh gotcha, never used logic, dont have a mac. But everyone says its solid

[weary wharf]

It's really really good. With the right tools, I don't think you can pass it.
The midi part is a little hard, do not much like it, I go to Reaper for that, but it has it's own problems -- do not like setting busses outputs etc. Too complicated.

[visual palm]

Really wish it didn't turn most of my outputs to Rap halfway through.

[dark pike]

I'm tired of getting country renditions of a Pop style song. How do I get Suno to understand negative prompts?

[scenic geyser]

Hi im new here. how can i manage to make 4 verses matching with melody and voice??? after i had 2 awesome ones i extended the song but 3+4 are ONLY way off....voice different or pitches are higher or so on..

really need help here...im blasting out my credits and thats not what im subscribing for...

[hollow hare]

The issue is that the AI does not remember the original style used. What it does, for each extension, is study the style ~1 min (unsure if this got extended to 2 min for v3.5) prior to the continuation point to learn/mimic the styling you had (which is why extensions in v3.0 are 1 min long as the other min was reserved for studying).

So if you had a loudness or degradation issue within that minute, it would start becoming emphasized as that is what the AI picked up on and all future extends builds up the issue as long as it exists within the 1 min of continuing.

It also explains why the style can change when you vary the starting point by a few seconds

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

[scenic geyser]

hi thanks. it should match tho... got my 3rd part now...need to manage to get the 4th then im done

[charred merlin]

Not able to give explicit instructions, but a helpful tip is to make sure that you "get whole song" to extend from.

For instance, you have clip 1 and you extend it, creating clip 2. If you want to extend clip 2, you're better to choose "get whole song" so that the extension has the full context of the song, instead of just "knowing" what the most recent clip is.

[scenic geyser]

nice this sounds legit. thanks

when you doing your own lyrics, which i always do is it harder for the ai ?

[hollow hare]

In the worst case, if you do not want to do the voice swap method and unable to get the correct vocals after doing the above, is to extend from the part that had the voice you like (right after the correct melody, maybe with a short interlude) and use a DAW (like Audacity) to append it manually.

[charred merlin]

Absolute worst case, generate the full song that you want, focusing on the instrumental structure.

Afterward, use a stem separator like the free OpenVino tools in Audacity to remove the vocals & add your own

[hollow hare]

The AI does its own thing. It is not necessarily harder for it but sometimes it will omit certain parts of the lyrics (or even lines) to help fit the flow.

The AI doesn't really understand what you are saying, it learned various song patterns and trying its best to fit what you want together with the tracks generated

[scenic geyser]

ok used extensionfrom whole song...got something ENTIRELY different now. seems like it ends the song and then makes another afer it. the other one was 5 seconds long haha

[hollow hare]

Yeah, that happens uncommonly. Try changing the timestamp (forward or back) by a few seconds, that is often enough for a change. You may need to redo a portion of the lyrics if it is a tad bit early

[scenic geyser]

ok

you a programmer of this ai?

or a mod here?

[hollow hare]

Just a regular lurker member who saw people complain about this issue several times in the past (and experienced it once)

[scenic geyser]

i extended from the end of whole song...another 5 second one with different song...

got credits refunded

[lone abyss]

Can I select the singer voice in suno?

[scenic geyser]

something is happeneing...lul i cant make it to the last verses

[lone abyss]

Like, I want a woman singing it, not a man

[hollow hare]

You are giving it some lyrics/tags right when extending?

[scenic geyser]

this is it

https://suno.com/song/40fe3d5e-7745-476f-8732-8c1fa0e22608

maybe the end is wrong?

[charred merlin]

Agreed, the extension point can make a huge difference, especially it "sounds like" the melody is concluding, or is at the end of a "musical response". Extending from a point of silence often just gives you nothing, so there's a lot of experimentation.

You won't be able to perfectly predict how songs are created or extended, but in order to really get a feeling for the quirks and how best to get what you want, you need to be willing to blow a few credits on experimentation

[scenic geyser]

hmm so better not from 1:51 but from like 1:50 or 1:45

[charred merlin]

1:49-1:51 is silence

[hollow hare]

You can enter it as part of the style prompt to influence it, like "clear female vocals". If it is a genre that does not commonly have it (e.g. heavy metal), you will want to repeat female a few times in the prompt to give it more weight.

Alternatively, you can use voice swapping AIs like kits.ai (see #🗣️┃chill-chat message ) to replace with another voice or your own voice with RVC models

[scenic geyser]

aah okay maybe thats why then

[charred merlin]

1:48 you can hear the drums start to signify the track is over, I'd extend from 1:45

[scenic geyser]

can pay only for once...im using midjourney too sometimes. now i have this for a month. i like to switch maybe later

[lone abyss]

bro just dropped gold

[hollow hare]

If you are using the free plan, the daily 50 credits refreshes at 0300 UTC+0 (<t:1713236400:t> in your local timezone)

[lone abyss]

btw

there is a way to say to the prompt that you want a verse to be a man singing, and another verse a woman?

or just using your off-topic hyperlink would be easier?

[scenic geyser]

having the most expensive one... i like to do some heavy work in a few days and edit after

but 10k aint that much i think they could raise it

ok i extended now from 1:34 but still doesnt work

[hollow hare]

I heard doing duets is a bit easier in 3.5 as the AI pays more attention to the metatags (so [Male], [Female], and [Male Female] tags are more likely to occur) but I do not know how good are the odds.

For 3.0, I had a spiel about it at #📚┃suno-school message

But, basically, yeah - voice swapping AIs would be the easiest alternative method.

Another alternative method, albeit still credit consuming, is making an instrumental part 1 (to get a very good tune) and extending it twice, one for the Male style prompt and the other using the Female style prompt

[scenic geyser]

lol...look thats my addon to the song...

https://suno.com/song/47d3a014-6018-4bce-99a5-55bc69b785ca

wtf

the rest is good but that first verse??? come on

[hollow hare]

At least you can clip it out / substitute it easily with a DAW, but yeah, sometimes the AI can be infuriating

[lone abyss]

bro just dont get tired of dropping gold

[hollow hare]

Oh, and if you are using 3.5, using dashes instead of spaces for a compounded terms will help get the correct genre (e.g. hard-rock)

See Jonathan Fly's post about 3.5: #📚┃suno-school message

Edit: Oops sorry @vocal cosmos, forgot to remove your tag 🙇‍♀️

[scenic geyser]

ok well it just dont work with the whole song...

i only get whispering stuff like that

[hollow hare]

Maybe add a modifier like [angry scream] or [loud] for the transition

Though probably not worth testing that out on your paid account. Test it with your single free-plan account

(Having one free plan alt. When you have paying account(s) is typically fine. Having multiple free plan accounts are against the ToS)

[feral surge]

My face when suno can't do percussion only, after many many many failed attempts: 😭

[hollow hare]

Did you try (or know about) instrumental tags?

Has a good example for percussion on that page. Also check out RayanAIR's Novo Lands and how he used those tags

[feral surge]

No I mean like drumline, stomp, that sort of stuff. Trying to get suno to do -just- drums for an entire song is impossible, it always adds in some kind of melodic instrument - usual some kind of piano or string.

[errant meadow]

so I am using version 3.5 for main body and version 3 for ending. version 3.5 fills in lyrics and keeps trying to add to it.

its annoying

[scenic geyser]

never had so many problems creating a song now...

2k credits in still no chance... before i made with 2500 5 songs

is a refund possible due to the destroying engine now?...for real man thats not good

[final goblet]

I 've just used the 3.5 version. It's so cool. However, I don't know about premium users, but for non-premium, is there a way to CUT the length of the song I've made? There was a ONE-MINUTE SILENT EXTENSION in the song I've made. Thanks.

[scenic geyser]

maybe extend it from before the silence

[final goblet]

I see, I guess there's no feature yet wherein you can cut it off.. Thanks!

[novel pike]

Hi guys. Is it possible to change the vocals within one generation? I want to make a male and female duet, my idea was to generate one part with female vocals then, extend with male vocals and so on but i'm sure i can use something like [verse 2, male vocals] or something but so far it doesnt work. Thanks.

[untold briar]

Sometimes formats like a script can work
https://suno.com/song/759c7e05-bb3a-4c53-ab34-545be0a241a0

[novel pike]

haha thats interesting.... and funny. And really... unlimited credits from 2 to 5 AM? 😄

[untold briar]

I just asked ChatGPT to write a Seinfeld script, too bad it's not a real policy lol

[novel pike]

hahaha oh well!

[still peak]

note for the devs: 80's hair metal does NOT equal country

someone put fuckin jackyl in the algorithm ffs

80's hair metal
80's glam rock IE motley crue everything's coming back fuckin travis whatever or nickelback

i thought changing the song title to "fast love" would help but nooope

also, what's the deal with no intro even though i type (intro)

also, 80's metal does no t equal SA-BA-TON!
i love sabaton but 'm not trying to make it lol

[lilac gulch]

Yesss you are right thank you 🙂

[wintry saddle]

@novel pike with v3.5 100% possible

Still may take a few trys to get it to behave tho.

[novel pike]

I tried it with female vocals in the prompt and Male: before the verse that I wanted with a male voice. It kinda worked out. I need to try a few more times but since it's not unlimited credits from 2 to 5 am I will have to wait tomorrow 🙂

[winter escarp]

is 3.5 turned off ?

i cant seem to select any models now

[untold briar]

3.5 is free and in top menu

unless therre is a random site outage at this exact moment, but that would be a glitch

[winter escarp]

ooo i just saw it position changed lol

[untold briar]

3.5 is also better than it was at launch

[idle jacinth]

So, I subscribed on the page and joined discord after and I can't use the bot? 🤔

[lone oak]

Does Suno have a new bug? The Playlists I created have been disappearing these past two days.🤣

[untold briar]

The bot has been replaced by the web site.

[idle jacinth]

Thanks

[lone oak]

3.5v can now be connected, but the melody of songs with lyrics is still more stiff than 3v

Is there something wrong with the suno server? The playlist and uploaded cover disappeared after a day.

Sorry, I found that the playlist was folded. What a strange design. It went to the second page even though the first page was not full.😇

[cold zenith]

i don't know why, but i've put this : " alto female Vocal " in style. But i always got a duet...

[dusk shale]

@versed valley I wanted to know what General commercial terms means, can we use on our social media account? Can we use songs and create videos and post them

[undone star]

what changes did you notice compared to launch?
Not that i dislike this complete lack of any details ever and we basically have to pull all the info out of our butts all the time 😄

[undone star]

if you have a paid account - yes. see the FAQ. https://suno-ai.notion.site/FAQs-b72601b96de44e5cacd2cd6baa985448#f0f1441180484d6094206e84e334ba36

[proven gull]

Can we separate music from singing

[charred merlin]

Search the chat history for previous posts I've made about "OpenVino" or "Audacity"

[broken lotus]

it needs a 3rd-party app like UltimateVocalRemover (free) to create 'stems' from the original song.

most DAW and many DJ apps now do this too.

the search words you need are:
'demux' the process of separating the audio elements
'stems' the resulting audio files.

[proven gull]

Thank you so much @charred merlin @broken lotus

[cold zenith]

i got a problem, i don't know why, but i've put this : " alto female Vocal " in style. But i always got a duet... After 50 tries, it still give me a duet..

[broken lotus]

I never got soprano tenor alto etc to work...

I wonder why alto would trigger a duet 🤔

only thing that 'works' for me (when it wants to) is describing the voice in either genre words or emotional words... I've explored a lot of the male voice (I use 'depressed' to make the voice very deep), but the words don't seem to transfer to female voices... 😓

[cold zenith]

what would you put then if you just want a female vocal in a accoustic instrumental ? ( because its keep give me pop and other stuff )

[broken lotus]

🤔 'female singer-songwriter' has gotten me some beautiful voices and acoustic guitar

[cold zenith]

thx you. i just sent to you in private, the prompt i use.

[still isle]

so if we have full rights to the music where do wer get the licence for each song?

[cold zenith]

the same way when you are creating a song on your keybaord.

[still isle]

@cold zenith I'm not a musician, I write lyrics and just found Suno to be a cool way to use my old stuff, so I made a few good songs now and a lot of people want to either have them or distribute them. So IF I was to distribute or monetize on a song I need a licence for it, how does that work?

[dusk shale]

Subscribe

[cold zenith]

No i mean, you just got to proctect it.

[cold zenith]

Someone that, create a song with his keyboard, or Logic Pro. Just have to protect his music.

[cold zenith]

This is the same things, you have to protect legally your property.

[broken lotus]

there is no 'Wizard of Oz' certificate to show. By the time you get other corporations involved in your dispute, the song will have been milked.

WHat we do is upload our song to distributer, they lock the song with a fingerprint.

Music doesn't last long enough to go to court – by that time it is millions of dollars.

we have to fight other scab wannabe/thief producers. To do that, we need to use the tools of the distributor, and be smart about never sharing the music until it is in the stores.

[still isle]

but distributors dont accept ai generated music

[broken lotus]

DistroKid does. Ditto does.
Who is your distributor? Switch.

[still isle]

soundcloud

ok il check those out

wich one is bette than the other? anny experience with them?

[alpine venture]

https://open.spotify.com/album/6cZm0ClZxWagQSnw2UA8ug?si=kDe7lvqJRwWDGqy1giXjSA

that was the fastest they've processed one

[alpine venture]

I mean, they clearly do lol

See above

Idk how yall think this wil work, I saw a lot of organized posting similarly oriented on the suno sub, but like, almost all of us have published stuff on there bc why not? Minimally, it makes it a bit easier to listen to when I'm going about my day

[still isle]

is that soundcloud?

[alpine venture]

but also, people like it

no Spotify

[still isle]

who did u use as distributor

i cant put music on spotify without a distributor it says..

[alpine venture]

yea, you can use any of them

wet circuit above uses distrokid

so I think that works for most

but any of them will work

bc it's just music

[still isle]

so you pay like 140 bucks a year then

[alpine venture]

I have no idea what anyone pays and I'm really not concerned with it lol

personally i have no vices except machine learning

so I'm willing to spend my expendable income on it, especially bc I enjoy publishing, in general

and I strongly encourage others

I mean, there's no loss in creating something, even if down the line you look back and aren't satisfied. At the end of the day, you'll never regret seeing something like that through

[broken lotus]

pretend it's not about AI for a while, and do some research just from pov of a musician/producer who wants to compare distributors...

[alpine venture]

In psychology, specifically DBT, the skill is referred to as "building mastery" and it legitimately helps improve subjective psychological health

[still isle]

i have no problem paying that eather just asking. Im trying to learn what others do,,,so 1. get a distributor and pay for a year to be able to register songs properly 2. then you can get licences for your individual songs. am i missing something?

[broken lotus]

there is no actual physical 'license' to download
it is a legal concept...

[still isle]

well, my opinion is, is it good to listen to, do i like it?, then i don't care who or what made it, but that's me. I do see the problem for professional musicians but hey now they can make music better and faster using these generators as additional instrument .

[sharp kraken]

Just do a lil FTT - not the same timbre not the same wood not the same house

[cold zenith]

Can someone ask the developer to put the 4 min on V3 too? that would be useful (it's unfortunate that it's not 5 min). that would considerably reduce the number of continuations.

[charred merlin]

Not sure it's quite as simple as that? I would imagine any improvements to the model result in a different model, like 3.5.

[valid flax]

yeah they can do that for sure ... for double of the amount of credits

i assume v3.5 is twice efficient at music gen than v3 and it requires less resources

but that's all all assumption since we dont know how their backed is working

[alpine venture]

I figured it's the same, but you could be correct. My assumption was that it's a simple fine tune, while v4 will probably be more parameters, or a superior architecture

a good fine tune on our data will steadily improve the model much more than most people realize

[grave wraith]

Question about the beta procedure:
3.5 is the beta... do the devs just make changes to it in the background or will a change for the user only be obvious when a new version comes out?
As it seems beta bugs are not fixed, taken from comments here and in bugs channel

[remote dome]

I feel like i get better results with v3 than with v3.5 - I just spent 1000 credits and not getting any results I was looking for with v3.5 but getting decent results towards what I am looking for with v3 within 200 credits using the exact same prompt

[valid flax]

3.5 is not beta

[undone star]

you might need to adjust the way you prompt to get the results you are looking for. restructure - think about the parts you want to emphasize on or use multiple prompts of the same keywords. And yes - even that might not give you the result of 3.0. If that is the case - start with 3.0 and extend with 3.5. works fine.

[grave wraith]

😄 Well...wasn't on the website yet. Just saw it was released

[broken lotus]

I think with 3.5 we saw SOME kind of update around 2 days ago.
My limited understanding is that 3.5EA is not exactly a whole new model, it is v3 with extra training on top... (I use the analogy of the overtrained stable diffusion models we can download from Civit.ai... they are the base model pushed with extra training to create a certain aesthetic....

when they talk about v4... it is some stages away still under development. A whole new model. (not so far away, about 4 weeks according to a Dev comment, but take with a grain of salt)

Info they learn from 3.5EA is being used to influence how v4 is being made..., but they do not go into specifics.

Also whatever the mystery update was 36hrs ago.... seems to have been done because of feedback on 3.5...

but Tony didn't say WHAT exactly was changed...

[grave wraith]

I just created one song with style rave to test and it is everything but not rave. Now I have to complain and ask why one would release this to the public as we read since days/weeks that 3.5 is worse than 3.
And it only got worse after being released to pro and premium.
I had some real great results on the first day. Only downvote was that annoying static sound after 3 minutes.

Now it isn't even respecting the style prompt AND has the static sounds and hiss all over the tracks.

A little bit disappointing

[broken lotus]

Are you somehow unable to use v3?

[grave wraith]

No, why?

[broken lotus]

You seem upset that we have a new model with different capabilities, as though it is the ONLY model we are allowed to use..., and there isn't a simple drop down menu to choose the model with the capabilities you need in the moment.... 🤷‍♀️

[grave wraith]

V3 is just not the topic. You argument like someone build a shorter road from a to b but it is awful slippery and instead of fixing the problem you say, you can use the old road. That's not how development works

The old road takes more time and more credits

[broken lotus]

I reject your monotheistic approach.
Older models are limited by the technology under which they were created, but we have the ability to switch model mid song.

I don't care to lose old models just because you complain.

I want MORE models.
not one.

[grave wraith]

WHAT are you talking about??? I complain as a paying customer that a model was released which is full of failures. I don't complain that models exist! There was a test phase with 95% bad reviews and they just released it to the public with no changes. THAT is what I complain. Do you read your channels?

[unique stone]

v3.5 early access regularly degraded audio quality after about a minute. Most of my tests exhibited this. Instrumentals would end very quickly - almost every time.

These issues were addressed with the final v3.5

[charred merlin]

Similar to ChatGPT's 3.5, just an optimized version of the existing model that doesn't quite qualify as a whole new model. Just a "turbo" model if you will-- some people genuinely preferred GPT3 over 3.5 because while 3.5 was optimized for speed, the results were different. Then GPT4 came out and everyone shut up.

[grave wraith]

Different results may be acceptable...but having a style like rave or opera and out comes a rock song is just a bug, isn't it?

[charred merlin]

Out of curiosity, is there any documentation regarding the test phase you're talking about?

[unique stone]

If anything I'm surprised at how quickly it went from early access to being released, with the issues I faced being resolved in the process. I expected it to take longer.

[unique stone]

v3.5 early access announcement - #📢┃announcements message

v3.5 released yesterday

[charred merlin]

There's absolutely nothing regarding a 95% negative response in this. I'm asking for documentation of the wildly negative response to 3.5

[unique stone]

Ah sorry, misunderstood

[grave wraith]

Go back through this and the bugs channel 🤷

[prisma wraith]

I personally was at the very least 82% happy with the early access quality, although miffed about the credits spent on 12-24 second long instrumental extensions

[unique stone]

I did see a lot of comments about how it was bad in one way or another, a lot weren't constructive and just of the "this sucks, I'm going back to v3" nature. I posted some particularly poor generations to highlight the issues I was facing, as did others.

[charred merlin]

Not really interested in doomscrolling through narrowminded echochambers of "this sucks because it's not exactly what I want". Unless there's something specific you could share, I'm not sure where you're getting the statistic of 95% disapproval?

There might be different expectations about the specific capabilities of generative AI-- there's amazing breakthroughs being made every day, but we're still in SUPER early days.

Just because image gen from Stable Diffusion wasn't itself capable of hyper-specificity, the community took those advancements and built upon them to make ControlNet and whatnot. (I know Suno is a closed model, but the logic still applies for the community at large)

[broken lotus]

you are welcome to your opinion.

You are NOT welcome to abuse other members because they don't agree with your opinion.

Have a time out.

[alpine venture]

It's so crazy how gaslighty the internet has gotten. Like, it's a legit strategy to just bombard places with blatant disinformation that every user will know at face value isn't actually true

this statement is absurd

95%? The real takeaway here to me is your statement "do you read your channels" which is intended to create the impression that this is common knowledge

when its absurd

like, on the plus side, I hope soon people are inoculated by false concensus, much like how standard advertising became less effective over time

[prisma wraith]

But in their defense, if all the data you're collecting is from a place where negative parts about a product gets put, then of course the end result would be overwhelmingly negative

[alpine venture]

right, but even in such a case, it's 95% people posting their songs with the caption "bangers and mash" ie subjectively it's successful

ya, we getting botted

[charred merlin]

https://tenor.com/view/james-franco-hello-127hours-can-anyone-here-me-stuck-gif-9845404

Echochambers are a helluva drug

[lone oak]

https://suno.com/song/ed5023b5-59f9-469d-a081-b98f984c9d5c

[violet relic]

When will Suno Pro users get Early access to Suno 4.0?

[unique stone]

Hopefully soon(o)

[undone star]

once the devs feel its time. soon (tm).

[lone oak]

Didn't the moderator say it would take another month? We can save a month's fee.

🤣

[broken lotus]

the estimate I heard was 4 weeks... but idk if that is 4wks and v4 is out of the oven... not necessarily when we can use it...?

If they have the in-painting working from start... I'd expect another week or so to finish the UI for new features...

but a 'beginning of summer' release sounds about right...

[charred merlin]

You mean you haven't gotten access to Suno V6 yet? Oops, not the memes thread.

[cursive nest]

Hey guys! Is this where I would ask for prompt help?

[lone oak]

I think I will be able to scratch all the cards before V4. In V3, the repeated feeding of lyrics has already caused the AI ​​to have language barriers.

I hope the language model of V4 can be smoother. I can already hear the feeling of translation software in V3.5.

[alpine venture]

oh my god I would have never guessed that soon holy cow

I'm losing it

[neat field]

try contralto ?

[charred merlin]

@developers thank you for making 3.5 the default model when opening suno (or saving previous model used as default, not sure which) ❤️

whoa Bark is open-sourced?!

👀

[hollow hare]

Yes, available at:
https://huggingface.co/spaces/suno/bark
https://github.com/suno-ai/bark

The list of old metatag doc page, which includes some for Bark, can be found at https://suno-ai.notion.site/Using-Metatags-34944efe09ec41d693e314f13af44695

[faint granite]

i am not able to use chirp