📚┃suno-school

[restive marlin]

But its okay how it is i feel even proud its flagged as fanatical becuse this song is about imperi guards of Cadia so should be fanatical 😄

[final hazel]

You should be proud.. 😉

[restive marlin]

Yea its duch a banger i want to play dark tide

Becuse of this song

[final hazel]

Or.. you can create more bangers 😄

[restive marlin]

I created many songs but most of them have some mistakes errors and this one is perfect 🤩

Its harder to make songs in polish i think

[final hazel]

Practice makes perfect though! Anyhow.. Time for me to sign off, it's 3 am for me (isn't it for you too?).. Have a good one!

[restive marlin]

i work until 7 am 😄 and i work on computer so i can make songs :d while working 🙂

[slim geode]

Sooooo I made a 22 minute song, and when I got the whole song, the audio quality gradually gets worse the longer the song goes on, with noticeable pops each time the song extends. Am I pretty much just boned, or has anyone found a workaround for this?

[slim geode]

For anyone who runs into the same problem, it seems to kick in for songs longer than 20 minutes. As long as I get whole song at 19:50 it's fine, but any time I get whole song from 20+ it's busted. At least most of it's salvageable!

[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 prior to the continuation point to learn/mimic the styling you had (which is why extensions 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.

The fact you made it that far is very impressive. I know a few people who resorted to making short clip start/extensions at the start just to ensure that the quality remains high (while starting off at a great quality)

[rough finch]

Anyone else having issues getting extened versions? Tried multipule prompts with no luck, most credits get refunded, although some are gone forever with no audible result

[midnight crown]

Try to extend from different minute / seconds, even just one sec can make the difference

[rough finch]

Thanks! I'll give that a try!

[slim geode]

Ah thanks, I actually like this style of creating a "mix", coming from a DJ background. What I'm running into isn't the gradual distortion/audio quality "drift" - that's been a fun challenge to work around while making this mix, but an actual failure of the engine to snap all the clips together when it compiles the entire track. Fortunately it only surfaces for longer than 20 minutes so it was just a matter of cutting the last couple tracks out of the "mix"

[untold briar]

That's right, the 'Get Whole Song' process introduces artifacts or glitches between joined clips once you get past ~20m. It's a known bug/limitation - I think it's just not very high priority since so few people generate songs that long, relative to other things they are working on. But you could still mention it in bugs or suggestions to make sure Suno knows some people are interested in going that long (I'm one of them too.)

[jaunty sundial]

I can't generate music full song same someone else
What should I do

I extend it but I lost credit around from 5 credit, so I don't want lost credit

Can somebody have new account and pay it, can I borrow your account just one year it's enough

I'll try to see full song or not

$24 I think so it's expensive for you, actually I choose $8 it's enough, I wish it'll full song

I'll trial play on this website when pay already

[hollow hare]

There is no difference between free and paid other than amount of credit and ownership/monetization.

Extending DOES cost credits, doing Get Whole Song to merge the extended parts to the original is free. See the following: https://www.suno.wiki/faq/making-music/how-do-i-make-the-song-longer/

[jaunty sundial]

Thank you for recommend so much

[untold briar]

I'm just noticing this old but interesting discussion, is this message talking about SDXL or Suno? If it's Suno I can't understand the meaning and am curious to know more

[alpine venture]

both; its a general observation about machine learning

so in Sunos case, for example, adding broad tokens will improve vocal quality for males

so I've been using years for this more recently as a way to test

specifically, strings like 010203040506070800

it's also why post- helps in many genres: it is a highly broad and popular token, thus it is highly generalized

and with this "non-specificity" it can help dilute tokens which may be overfit

without polluting the style too much

it's essentially the same method I've been using recently to generate native 1600x1600 sdxl images

in 20 euler steps

with like 70%+ wr

in the example above that you replied on the part that's key is "post-post1,2,3,4,5,6,7,8,9,0,00-vocalist"

[untold briar]

Strangely I agree with the end result (what ends up in the prompts) but my mental model is almost reversed: I think of tokens like that as atypical anti-general narrow tokens. (Partly based on how they worked in versions earlier to V3, where the 'narrow' effect was apparent, but I kept it in my head model I guess). I wonder what spamming empty commas or semicolons is doing under the hood, just a random thought typing this now.

[rough bluff]

Can you search

on suno?

[untold briar]

In V1 sometimes a random string was almost like one-specific-voice, if you found the right one

[jaunty sundial]

Everyone
I use along you told me, why my credit left 5 credit. What should I do

[untold briar]

Free credits reset to 50 once a day

[jaunty sundial]

Oh nah friend, I'm sad about it

Everytime Suno's must credit for generate it right

Always Use credit in everytime right

50 credit per day
It's not enough for me

[jaunty sundial]

Is there any way that doesn't require credit?

[final hazel]

Computing power cost money. It's already a huge gift of Suno giving you 50 credits per day. Not many services do that.

[final hazel]

Only the sun shines for free... 🌞

[jaunty sundial]

I'd use credit a limit like that?
It's okay at least I like this website, I wish develop this website will make really Free and no credit anymore

[autumn lion]

Hi everyone

Does suno provide API to use Suno for developer using at other platfom?

it better if suno provide API to use it

👍

[final hazel]

As I mentioned earlier, nothing in this world comes without a cost. If something isn’t paid for with money, it’s often paid for with data or other resources. The 50 free credits Suno offers daily are quite generous considering the substantial costs and the hard work of the real people behind Suno. Just as you wouldn’t work for free, the Suno team needs to be compensated for their efforts to continue improving and maintaining the platform. I hope you understand and appreciate the value of the free credits provided and the option to purchase a plan if needed.

[final hazel]

Suno team is working on that. There is not ETA at the moment.

[bold current]

First off, I'm a fan of SUNO and love the app, but with that said I've notice a degration in output this past week to the point most of the songs I make are of extremely low quality. I've only been using the service for about a month now and even between that time and now I can notice a difference in the outputs. I make ALOT of songs btw. In the past the low quality entered about on my 2nd or 3rd Expand to make a longer song. But this was on the very first try - https://suno.com/song/b115ce65-7e99-4533-aeeb-09a79c292cae

Just curious if there is a known problem and as a paid subscriber is this being looked into?

Here is another example of the same song on another attempt that I think sounds okay to start, but quickly dissolved into Muddle, of what others sounds like it would of been a Epic song I would of liked to keep.

https://suno.com/song/9f61a90c-e420-49b9-8099-fdec20c195b0

[lilac garden]

i have a song i created on the free plan, if i re-extend it on the paid plan do i now own the rights to distribute? I don't think the faq answers this question.

[lavish latch]

i think mostly if you try to generate a song with style of power metal / epic power metal / symphonic metal, will end up messy and strange

[bold current]

Thing is.. 2 weeks ago I was not having this problem. And when I did it was later in the song after the 2nd or 3rd extend. Never had it this bad right out the gate on a multiple Creates

[lavish latch]

yes it's the same as me, the first try was successful

but when you try it again several times, you often get strange results. Such as irregular tempo, chaotic pronunciation of lyrics, or even instrumental sounds that don't clearly sound like what they sound like.

[bold current]

That 2nd song I posted up there is usable to me, if it didn't go muddled a little ways into the song. It's like murder.. hearing an epic SUNO song destroyed by reverb. Not to mention the Slot Machine effect of just dumping credits for a hope of getting a Jackpot of a working song.

[hollow hare]

They did address it indirectly, said that the song still counts as a free one because the base was made during the free plan (and probably has an audio watermark denoting it as such)

Hm can't find the entry about the extension info itself, it used to be there... I think or maybe I am experiencing a Mandela Effect x.x (edit: more likely the latter).

[unique stone]

If you haven't already, flag the song (use the ... menu) for poor quality audio and post some examples in #🪲┃bugs-web so the developers can take a look (you can put < and > around the address so it doesn't show the preview, if you have several examples)

[bold current]

I think the Devs might have me flagged to ignore I've flagged so many songs now 😄

[dusky marsh]

Hey, what if using Suno to create a track, then working around with a real human voice. I'd also like to modify the instrumental but retaining the lyrics. Would it be possible to distribute this as my own song with full rights?

[cold zenith]

if you have the paid subscription, it's all ready your full rights

and you can do what ever you want after with the song you have generated during " the paying subscription ".

[paper imp]

My songs have disappeared. The one I was working on yesterday is still there but all the others are gone. Has there been a change to the menu or is this a bug?

[barren plaza]

When is the official suno api coming? Didnt found one

[dull isle]

I’ve done the same exact thing. Is anyone able to help point us in the right direction as to if we own the rights to it or not. I would assume, since we’re extending using paid account and are paying for extending, technically we should have the right. Please let us know. Thank you 🙏🏻

[cold zenith]

if you're extended your songs with free credit, it means, you'didn't finish it in subs...So....The whole song is not yours.

[hollow hare]

I can't find my song in my library! What do I do?

First, reload the page and ensure you are in the library in case there was a visual glitch. For cache-bypass reload, see #📚┃suno-school message

For usual causes:

• You are in the create screen, not library, and the piece was down voted (which causes the entry to be hidden)
• Could be on the next page (blue circle for pagination on the picture at the end of the post)
• A filter is enabled (yellow circle), please disable.
• Accidentally trashed (red arrow to access the trash bin)
• The text filter can be finicky. Try to use as much of the full title as possible or unique keywords if the first attempt failed.
• Accidentally overlooked the entry (with text filter disabled). May need to reload the page first.
• Logged in to another account (can double check by going to your profile)

https://media.discordapp.net/attachments/1102676217687195679/1246851079182880808/image0.jpg

[cold zenith]

but the contrary....I don't know...

[dusky marsh]

I have a subscription, and yeah the song was generated during the subscription. Thanks for your answer 🙏

[errant python]

whats the best mm:ss to start the verse from the instrumental?

[tropic valley]

what's with the subscription logic of not having daily credits after
eg: use case where you'd run out of credits a week before renewal, so you are considered less than a free tier user? or is it related to free daily credits being the non-commercial use, while subscribed ones are paid upfront commercial use?

[fluid robin]

man its a bit a tad confusion to start upload the songs to spotify
i can find the local file thingy on the phone but not on pc

[hollow hare]

You need a distributors like DistroKid or Dittocto publish your song to Spotify and other companies, cannot upload directly onto there

[fluid robin]

ah got it

thats a bit annoying if u want to be a independent artist

[hollow hare]

Likely they done it so the company does not have to deal with as much legal issues and shoves that to other companies to deal with.

Found the get started page: https://artists.spotify.com/get-started

Make sure you read the ToS of the distributor (go over some videos that Top Music Attorney did on YouTube) because it WILL affect you. Especially how royalty will be distributed to you (and if you ever leave/kicked off a platform).

IIRC, there was an issue with TuneCore's policy she had pointed out as being quite problematic: #🗣️┃chill-chat message

[wind island]

SoundCloud I think you can

[fluid robin]

( Replace these with [ and ] easy solution when chat gpt being annoying

can be put underneath the posted text in the program and it will fix the issue fast

[opal storm]

im curious as well

want to integrate suno via API within our Real-time AI UGC system

[paper imp]

Hi. Thanks. I have 2225 credits left on my monthly subscription and I am in the correct account

[glacial scarab]

Did you click on this?
"Library"?

[paper imp]

It was the page navigation. Thanks

[lilac garden]

Yes it sort of doesnt make sense u end up paying for a subscription after hearing something u really like(i wrote it too) and are unable to claimnit as yours after u pay. I dont really think the faq specifically addresse this scenario of extending it can go both ways with the way is written. If someone from Suno can address it that would help

[elder glen]

hello my name is Panda im new to the server and i use suno ai for a few times now but when i try to hear the music theres no sound coming out what shall i do ?

[runic dock]

And what are the steps you followed?

[elder glen]

what do you mean with steps ?

[teal raft]

Always appreciate your research and sharing what you find, thank you.

[elder glen]

the things what ive done is restart the website plus my pc

[final hazel]

check your right corner, if you didn't mute the music

[elder glen]

no i didnt mute it its completely on full 100%

[opaque canopy]

refresh the page ?

[final hazel]

Do previous songs that worked before, still work?

[elder glen]

done that too unfortunately didnt work 😦

[unique stone]

Have you muted the browser tab?

[elder glen]

no i tested my songs what tried and nope no sound

[unique stone]

Do other websites play sound?

[opaque canopy]

can others hear your song?

[final hazel]

Well at least you are bringing the community together to help you @elder glen 😻

[elder glen]

i didnt made it on puplic its on private since i started suno ai xD

[elder glen]

glad i could help with that i appreciate it very much ^^

[final hazel]

Please check what @unique stone said here: #📚┃suno-school message

[opaque canopy]

i will make a bad joke have you try reboot your computer lol

[elder glen]

no

😦

[final hazel]

Ah.. well then it's your browser.. Are you on windows?

[elder glen]

yes

[alpine venture]

to be clear, Suno is a but harder to Guage, and I cant do rigorous testing like i can with sdxl

so I could be tooootally off base

it wouldn't be the first time

[teal raft]

We're all machine whispering shamans here

[alpine venture]

i once thought titles influenced musical output

[final hazel]

type in your windows search "sound settings" and got to "advanced sound settings"

[alpine venture]

for real haha

it's half-mysticism with closed models

[teal raft]

I know and I uniorincally love it

[alpine venture]

but I get it personally, if I was them if keep it closed

[teal raft]

Even if they were wide open we'd still be grasping in the dark

[alpine venture]

they deserve the payday

[elder glen]

okay will do

[alpine venture]

well I mean, a tokenizer tool would help

so I can at least know what its parsing

rather than guessing based on output

[final hazel]

There you should be able to change the volume back up.

[alpine venture]

mayhe ill make they in suggestions

[elder glen]

it works again

[teal raft]

What ended up being the issue?

[elder glen]

my second email on the microsoft account it brought me with no sounds at all and couldnt even hear my game sounds when i was playing something

i accidentely used my gmail on my microsoft acc

bc i wanted a second minecraft acc

[sly mortar]

How can i create the song? When I click Create button, I see "An error occurred. A system error occurred".

[teal raft]

Glad you figured it out. Stuff like that is deeply frustrating

[elder glen]

Yea it is definitely i thought i can make with that a second Minecraft account but i guess i can only use my main one

[final hazel]

Have you filled in all the fields?

[sly mortar]

yes

[final hazel]

Can you show a screenshot please?

[final hazel]

I would relog.. I don't see your credits

[sly mortar]

what is credits? How can I get it?

[final hazel]

You get 50 credits free per day. And each generation give you 2 tracks which cost 10 credits

[sly mortar]

so why I do not have credit

[final hazel]

You should see them here:

[final hazel]

I don't know.. but that is why I asked you to relog

[sly mortar]

oh, you are right. I can see it

[final hazel]

If you can see them now, it "should" work 😄

[sly mortar]

yeah, I catcht it. Thank you

[final hazel]

Enjoy making music with Suno!

[sly mortar]

thank you. I will try relog

[sly mortar]

I have to sign up a new account and now I can see my credits

[broken lotus]

Confirming Yolkhead's observation that
post-male-<AnyRandomNoneSense17> vocals
is giving very good results 🤔 🤷‍♀️

[alpine venture]

yay!

[teal raft]

The other things they've said about noise in general seem to be A++ as well

[broken lotus]

even on 'shoegaze-rock' which should theoretically sound awful.

meanwhile just
post-male vocal
does nothing that I can tell...

[flat robin]

I think it would be interesting to exchange prompt ideas. I have a prompt I use with Copilot that gives me 5 style prompts (or at least 5 starting points) from a song title and artist name. I can post it here as soon as I go back to the office.

[errant python]

Its probably just useless or annoying to them to differentiate between the purchased credits that fall under a different terms of usage and the free credits.

[tardy iris]

Hi, I'd like to delete my Suno account. What do I do ?

[proper mirage]

Will there be alternative ways to pay besides CC?

[final hazel]

You will have to contact [e-mail] for that

[uncut pumice]

Why is it my songs sound great at the beginning and slowly get worse as I add clips? Is there a way to stop this?

[tardy iris]

ok, thx

[neat nexus]

The thing is, even if I gave you my exact prompts it would still come out differently for you. And that's kind of a good thing. We all get our own music and style.

[flat robin]

My only intention was to reason about the best way to have an AI output something feasible for a style prompt. For example, I give it guidelines such as "start with the genre but omit the word 'genre'". Stuff like that.

[neat nexus]

Eh? No that's no problem. I'm just explaining why I do it the way I do.

There's nothing wrong with sharing what questions we use to interrogate the GPT.

[flat robin]

Great! I'll post my prompt as soon as I'm back in my office.

[neat nexus]

For instance with the above, I asked this:

What can you tell me about the song "Name of song here" by "Name of Artist here"? Particularly the playing style and musical notes and techniques?

I then follow up with:

Can you give me ten genres that this song style would fit into?

And try to clarify that with:

Good. Thanks. And if you were to describe the music in less than 90 characters, how would you describe it?

See I know I'm never going to replicate the song. I'm not even trying to. But I'm hoping to be edumacated a bit and figure out some good ideas for metatags and for style prompts.

[alpine venture]

OK, I just saw that we can make custom images

so that's awesome

anyone know what model it is?

[final hazel]

casually drops a new link: https://suno.com/search

[fervent pebble]

Hey er'body! Having a lot of fun with Suno but I noticed a change recently I'm bummed about. I was creating lots of authentic, old, retro sounding songs with Suno using descriptors like "1960s, old, retro, vintage, vinyl" and this was working WONDERFULLY to make songs that sounded like old records! And then recently something changed and everything I've been generating is modern sounding...or trying to be modern sounding. Has anyone else had issues with their prompts not being interpreted properly?!?!

[spiral star]

any staff on? i cant seem to logg in my account, i click the proper email account in the google sign in list but it puts me in a diff account new paid for and empty

never paid for, em,pty

[final hazel]

Support can be reached via [e-mail]
Only advice I can give, is to check your email box if you are absolutely sure you are using the right login.

[spiral star]

i been in contact with the mail. been playing tag for like 20 hours, my times running out

i wann ause and get my stuff off then make aq new account

only like 30 hours left and i have 1/2 my points left

im doing it right but a no voice cust suport is killing me

the email suport is like watching apint dry

paint

i got a life to do also, cant sit and wait forever

[final hazel]

So you are not paying for the plan?

[spiral star]

im paid

it weont let me log in

i got A HUGE L8IBRARY I CANT ACCESS

oops caps

[final hazel]

I am really sorry to hear that, but we have no support here.
What plan are you on?

[spiral star]

i got a dozen messages in email and it takes hours each responce

the basic paid 2500 each month

i cant even use that in one month heheh

i think i got 1300 left and like 30 hours left

i have no prtob making a new account after

[final hazel]

If you shoot me a message in my DM with your email address, I will see if I can send it for you

[spiral star]

ok thanks

love your prog, and ive used them all over last 3 years

[final hazel]

I don't believe Suno is that old lol

[spiral star]

its newr, i used all the others first

i beeng releasing music with AI on disrokid 2 years and a diff service before that

[final hazel]

Good to know.. Let's go to DM 🙂

[spiral star]

it a hobby to me but now im makeing money and breaking even

ok

[ancient maple]

Is there any way (or maybe planned) to remake a certain part from the beginning of a generated song? In my case the very beginning (first 2 secs or so) are messed up but the rest is perfect :/

[fluid robin]

should i make [Narrative Voca] in order to get a more story teller part ?

[cloud pilot]

any updates coming to lax the censorship?

im a rap artist and i dont wanna be constrained

[sage stump]

is there a solution for the loading screen, when i click on create or is it just the high amount of people want to join too?

[hollow hare]

I really doubt the censorship system is going to go away. Too many people wanting to do stuff against the ToS.

Despite what you want, the ToS does clearly state that vulgarities is against their terms:
https://media.discordapp.net/attachments/1102676217687195679/1236146831902244874/image0.jpg

If you do not agree with these terms, your option is to go to a service that does allow it... which is pretty much no company really AFAIK

[cloud pilot]

when i spoke to the Dev he said that they would do that... i guess hes a liar cool

[hollow hare]

You just got to sign in. When the sign up screen appears, there is a small sign in link at the bottom left of the screen

[hollow hare]

Then wait for them to do that but there is only so much relaxing that can happen.

I can see them relaxing rules against things like having the artist Rico interfering with using Puerto Rico in the lyrics.

Obscene vulgarity though... hmm, maybe to a small degree, but will likely leave discrimination, etc. left untouched

[cloud pilot]

hasnt stopped me from getting around it and making the most unhinged songs of all time. its just such a pain in the phuckin azz to have to scrutinize every single way to spell something and even when i use words i know will work it will still read context clues sometimes and stop me.

[hollow hare]

Obviously getting around the censorship is possible but still does not absolve you from the ToS. If discovered by a member of the team that violates the terms, your account is liable to consequences of knowingly skirting the censorship.

For small stuff like swearing, it likely won't be much, but other topics such as calling for violence or murder against a specific person or group would result in - at least - a suspension of your account (if not getting forwarded to law enforcement, depending on the content used)

[cloud pilot]

i am talking to a bot. ok peace out

[glacial scarab]

but muh violence

[cloud pilot]

lol idc about violence i just wanna make funny songs

[final hazel]

Please show a little respect for all users. Paimon is just trying to answer your questions. Just because you don't like the answer, doesn't make Paimon a bot.

[cloud pilot]

i didnt say i didnt like it

[final hazel]

Then why call Paimon a bot?

[cloud pilot]

sounds like 1, its like a chat GPT response

[zinc stirrup]

guys, I generated a song, and I like the melody, but not the lycris. Is it possible to adjust the lycris but remain the melody

[teal raft]

They are a real person though so

[final hazel]

No, you are wrong. Sorry. But it does come off you don't like the answer, which is understandable because you want more freedom. @cloud pilot

[teal raft]

Coming soon: Playground insults revolving around calling kids you don't like 'bots'

[cloud pilot]

yeah i also just dont like the assumption that im asking to make "death threat suicide note the song" like the most extreme edgecase lmao

[final hazel]

Not atm.. In the future with V4 you can with inpainting.

[teal raft]

Maybe it already is idk I ain't no kidologist

[zinc stirrup]

when would v4 come?

[final hazel]

Well let's just drop it then for now.

[final hazel]

We do not know.

[teal raft]

"soon"

It's always "soon"

[fiery prairie]

I really hate to interrupt, but...

I generated a new image for a song, but when I open the image in another page to save the large version, the address for the new art is just a never ending string of numbers and letters and I don't see anywhere to add 'large'. I checked a normal image and it still works fine. Not the end of the world, but it would be nice to still be able to get the full size. Anyone know anything about it?

[teal raft]

See

[unique stone]

What's an example of the subject matter your songs are about? (that are causing you censor issues)

[teal raft]

Especially reading the latest announcement

"coming weeks" seems pretty swank

[final hazel]

I would drop this into #🪲┃bugs-web since it is a new feature and might have some bugs atm

[zinc stirrup]

all right, some times the melody is very good, but there is some small problem of vocal or lycris. Cannot edit it is a pity

[hollow hare]

Depends on how long the melody goes for. IMO, the easiest thing to do is Extend from a portion you like which has no vocals and use a DAW to trim off the portion you do not want in the end.

If you really like a melody, what I had done when I encountered the same situation, is extend the portion as an instrumental (using various instrumental tags) and used the resulting instrumental part for extending for new lyrics.

[teal raft]

and the devs are currently real keen to hear how we are all dealing with the pictures thing

It just launched, always a huge deal with a big change for users

[fiery prairie]

ok, I wanted to see if maybe I was just missing something obvious before submitting as a bug, because I've done that before

[teal raft]

Never feel bad about reporting something. At worst they've heard it before and you'll get told how to fix or get around it. At best they haven't heard of it yet and you'll be the first

[zinc stirrup]

wow, it sound great

I will try it

[final hazel]

It's brand new.. and sounds like a bug 😄

[fiery prairie]

thanks 🙂

[cloud pilot]

mainly just any swearwords, sexual inuendos, criminal activity in general, references to any type of firearm, references to being gay and its ALL SATIRACLE!!! the music you hear on a regular basis on the radio would be striked as "too sensative" by the AI. Its just so stupid. i completely understand you dont wanna have people making actual harmful material but i personally dont care at all and (this will sound dumb) but if i were building this i would not wanna limit anything you could say. plus im paying money for this. guess ill just stop paying and wait for a free model of this that doesnt retrict me at all. because some freaks could "possibly" make a song that is way to over the top it means the rest of us just trying to have a laugh have to carefully script every single line just to have a good time. anyway thats all i gotta say. im not here to justify why i like making unhinged funny music that hurts literally nobody and only me and my 2 friends hear it ever. if you wanna tell me im an EVIL PERSON dm me.

[zinc stirrup]

are there some tips for creation? I mean, I can't control when it's time to go to the drums, when it's time for the interlude, when it's time for the climax of the song and all that.

[hollow hare]

Have you checked out https://suno.wiki ? The song structure entry will help you out (do check out all the entries though, such as the ending tag entry which introduced me to [Big Finish])

Surprisingly, the [metatags] can do a lot more than you think... when the AI wants to listen to it though 😛

As an example, check out one of RayanAIR's song and the tag they use: https://suno.com/song/a98da392-ee0d-4fad-ba9b-937f0c9c5a91

[zinc stirrup]

well that's useful, I didn't see it on official website before, thank you

[hollow hare]

It's an unofficial wiki made by the moderators here. There were rumors of the official guide/wiki being dlated for an update but it may be some time before that happens as it is likely a low priority.

The old Guide had some interesting tags for Chirp (aka Suno v1) and Bark (their text to speech system), several likely defunct now, but some still work like the Girl Group style (https://suno-ai.notion.site/Using-Metatags-34944efe09ec41d693e314f13af44695)

[clever bloom]

I am trying to start off a song with an opening narration that is spoken, not sung. But no matter how I set up the prompt, the words are always sung, not spoken

[zinc stirrup]

Thx bro

[hollow hare]

Hopefully this helps you: #📚┃suno-school message

[graceful condor]

is there a way where i can make it where male sings then a female sings?

[hollow hare]

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

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

Again, getting it working on purpose is hard.

Alternative methods:

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

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

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

[swift lichen]

Is it possible to make a song using some of my own samples too?

[edgy river]

Is there any chance my views are stuck at that number? It's been a while and I haven't seen the numbers go up even though I shared it with my friends. I'm not sure, but could it be a visual bug? How would I fix it if that's the case? Or is it just that no one played the song again lol.

[clever bloom]

Is there a way to proc a chorus when you're generating lyrics in Custom Mode?

[hollow hare]

There is a [chorus] tag. Heck, I think something like [Gregorian Chant Chorus] can work, gotta try that one day.

If you haven't checked out the wiki, be sure to read all entries. This one sounds like https://www.suno.wiki/faq/metatags/song-structure/

[graceful condor]

im trying to extend a song, but the extension only goes up to 10 seconds and just repeats the last sentence and ignoring the entire paragraph

[hollow hare]

Did you have any lyrics when extending?

I highly recommend adding some instrumental tags or other song structure and expand the lyric box as much as you can so it forces the AI to continue on:
https://www.suno.wiki/faq/metatags/instrumental-tags/

Another issue I found with repeating in the past was if the song structure were the same (e.g. you used just Verse instead of verse 1, verse 2, etc.)

[lilac garden]

are you finding that the longer you wait to work on a song you created and extend it (days,or weeks), the harder it is for the engine to produce consistent results --- if i wait to do some changes ona song sometimes i get a copletely different voice, and completely different direction

[tropic fox]

Has the prompts syntax changes in v3?

Seems like whatever's in the make square brackets are being spoken.

Is there a current source?

[azure crane]

Why have you removed the ability to see the player when Expanding a song? It must be fixed asap…

[quiet estuary]

Hello friends, do you also have the problem with scratching in the songs in V3? All my created tracks have been scratching extremely since yesterday.

[fair cloud]

hello y'all , can someone help me, i need to know wich options are aviables actually, ty ty in advance 🙂

[cobalt spear]

nice to see the improvements on the site, but when will the subterranean audio quality finally be fixed?

[peak plank]

I'm having an issue today. I have a clip which ends after the second verse. I'm then trying to extend and make a bridge and chorus, while transitioning from the acoustic sound of the verse to a more metal sound with distorted guitars. The issue I'm having is that 75% of the time, the first line of the chorus is skipped entirely. Most of the time the chorus is started with "Though in need of a bath", see lyrics below. It also frequently replaces "Neat" in the first line with "Sweet". Why could that be?

[Heavy metal riff]

[Chorus]

Neat king Eryndor!
Though in need of a bath.

Sweet king Eryndor!
But do not cross his path.

Leet king Eryndor!
Or you'll suffer the wrath.

WRATH OF THE LICH KING!!!

[uncut pumice]

LMFAO. I will cancel my sub. Udio is better anyhow !!! Fuck censorship !!!

[fleet edge]

did I see some announcement of a meeting today ? I can´t find the info anymore

[hollow hare]

It is your money, do as you wish, but be sure to take a look at the ToS of the service you are using.

Because, surprise (well, not really), Udio also has something similar! Shocking right?! A company trying to limit their liability 😮

(Forgot to highlight the lower half and there are other similar terms in section 5.12)

So even if Udio does not implement a censorship/content filter system, which is mentioned after the yellow box, you can still be held liable for the content you generate and the consequences resulting from it.

[fleet edge]

Oh I love this topic. Just so you know, a lot of people creates songs to bully people. So yeah, there has to be some limit or control over what people can or do create, because there has to be some limits in life.

[bold current]

I hate that those same censorship items will be how people can tell if a song is AI or not in the future

[fleet edge]

thats a good point yeah

[bold current]

It will also propell more extreme content for people to prove they are not AI generated

[fleet edge]

I´d reduce that list tbh. I´d approve obscenity and gross stuff , just transfer resposibility to user. Just focus on difamation and bullying

don´t see the point on creating Disney songs, if people wants songs about drugs and sex

but defo I´d ban defamation and bullying

[bold current]

I do a lot of Cyberpunk. so sometimes going hard there is needed for the genre.. but was a musician before SUNO so I can just make my own songs if I need that

SUNO is more about exploring and experimenting for me

[fleet edge]

yea. its tricky topic. the issue is where to draw the line, and who decides where the line is drawn. so, there will never be agreement on this

[unique stone]

In fact, Udio users also frequently complain about the moderation being too harsh. The grass is not always greener on the other side.

[glacial scarab]

I heard the quality went down as well since you can pay 😏

[unique stone]

I've read this. Supposedly they want you to waste credits even when paying 🤔. It's the same as it always has been, to my ears. Anyway that's a topic for off-topic so that's all I'll say here.

[broken lotus]

These are the people who refuse to eat sushi because the rice is there to fill you up.

and dinner rolls are also there to prevent you eating a full course.

lol

[pliant comet]

this site has been a great tool for me editing suno songs - allows you to split the vocals, drums, and bass from the music and download them each as separate tracks so you can edit them individually in Audacity, Adobe Audition, etc. https://vocalremover.org/splitter-ai

[soft lintel]

GM, not sure who to reach out to but it looks like our daily credits have not been being added. Does the Free plan still get 50 daily credits?

[broken lotus]

yes, haven't heard any problems...
they reset for everyone at 1 specific time each day: around 10PM EST

[edgy roost]

Any idea how I can manage to recreate this sound with a promt? No matter what I write, I can not get the effect to come back. Half female singing, half growling.

[glacial scarab]

@edgy roost
you will have to prompt it part by part change the styles and hope it will catch on after a while.. basically just the same as with every duet just even more difficult
But possible I did something a bit similar here where it switches vocal styles:
https://suno.com/song/405f4e0c-c038-43be-b138-3e1e4470d8d0

[edgy roost]

Thank you. It came up randomly for me with aggressive alternative metal promt.

[glacial scarab]

yea, I also just rerolled until it happened at the start after the first "Ah"

after that it more or less kept going, few retries because sometimes it stopped doing it.

best bet is probably chuck a bunch of conflicting vocal styles into the prompt as well as some genre where that is a bit common

[soft lintel]

Ok thanks for the info. I didn't see any added yesterday and don't believe the day before did either but I will keep an eye out tonight to see.

[nimble sundial]

is there a prompt guide anywhere? I've noticed (lyric) is an echo type effect. is there a list of things like this?

[final hazel]

There is a community wiki which you can find here: https://suno.wiki
This wiki is not complete though, still much to discover. The devs know what they have learned the model, but it doesn't mean they know all the ways how to communicate with it (if that makes sense).

[nimble sundial]

TYVM

@final hazel sorry for the bombardment but is there a way to increase the lyric length text box if needed or is it a hardcoded token limit?

I'm hoping to try and push this to it's limits to see if I can make a fully custom structured song with lyrigs and production techniques but a lot of metadata will take up lyric slots

[final hazel]

1250 is a hard limit, but you can just extend a song with the button "extend", add in more lyrics and keep going. Maybe this video will be more help for you. It's my personal workflow: https://youtu.be/lj8utBd_IUM

[nimble sundial]

oh thank you so much

ah so I can split the song up with the extend, add new elements then combine at the end. great!

[untold briar]

As a substitute for not having 0 second Extend From capabilities, toggling to V2 for the 1 second time seems to wipe out more of the musical memory. For example if you want to gender switch the singer voice in the middle of word.

Interestingly, a few of of the more abstract sounds (things you wouldn't really describe with a musical note, but are content in the song) seem to be interpreted completely differently in V2 and V3. Sort of like how in Bark the same semantic token can render as a guitar chord, piano chord, or synth chord, or even a voice sometimes

[untold briar]

some of the test tracks are mildly amusing, at least the short bunches of clips in between the long normal clips https://suno.com/song/2de0daf5-a0c9-452e-a802-f48bbb4b4452

[glacial scarab]

this needs so much more attention, it even works most of the times in my "blender" style prompts where I just throw 5 genres or something together to see what suno does

[worthy yarrow]

possibly off-topic

but have the devs shared any sneak peak of v4?

any features they teased

[untold briar]

I'm curious to test against my 'through nonsense into the prompt' strategy and see if they are essentially the same thing. (id, do you even need "post-male" or do you just need extra chaos in the style prompt?)

[glacial scarab]

hmm I think "post-" does stuff on it's own so it might work just differently without it.

[untold briar]

Try using other words that have an effect, connected by dashes. Being a compound word for sure has an effect versus otherwise.

oatmeal-reel-AnyRandomNonsense, etc

[glacial scarab]

And I think many of the solutions go into the same direction as @alpine venture noticed by the earlier iterations of the "noise removal" prompts. this one just seems to not influence the style from what I understood like other solutions (like me using singer-songwriter and then a bunch of slightly related genres and styles to the main one to steer it away from actually being "singer-songwriter" genre)

I really need a bigger subscription to test all this extensively 😅

[real plume]

Hello, when creating a foreign langage song (french), and we wanna specify the "style of music" do we have to specify in french or english ?

[untold briar]

Usually English works best, but not always. In fact sometimes using completely different language is best, like German style words for an English song works better for a specific type of music or an instrument. But probably start in English first.

[cerulean pagoda]

Can I redo the beginning of a song?

It jumps into the lyrics too fast and I want to add an intro

[thorny cypress]

No, but if you want to influence it to make a longer intro, you can add nonsense words to the beginning of the lyrics like oooh or la la la.

Or if writing in French, ooh la la. 😛

[untold briar]

You can guarantee an intro by starting with instrumental (no lyrics to sing) and then Extend From with your actual song. Nonsense words a la oohhh and la las are a nice middle ground as well, yeah.

[old tartan]

I am hoping that suno will implement inpainting and reverse extensions. It is so fruatrating to have an almost great take and have to burn it because I can't alter it at all. Either because its a syllable off or just starts too soon.

[teal raft]

v4 is 🔜

soon

where infill/inpainting will be a thing

Or at least, has been long talked about by lots of the devs

[old tartan]

I really hope so. I prefer suno, but ive been dipping over to the other guys because of these fantastic features. The longer it takes, the more time I spend over there.

[graceful cliff]

maybe sth else will come out first...

[final hazel]

https://tenor.com/view/let-cat-out-of-bag-who-told-you-who-revealed-secret-tell-secret-cat-in-bag-gif-24638665

[unique stone]

Intriguing! I reckon either a revision of v3 or perhaps offering "v3 alpha". 🤔

[cobalt gull]

Hey folks, just joined after messing around with suno for a little while. I'm a dev looking to build an app around suno's api / maybe on top of their model ... anyone know if this is possible or been done before? Maybe who to reach out to about this? My client has $$$ lol

[hollow hare]

There is currently no API publicly available (something you would find out easily if you searched because a LOT of people were asking about it).

IIRC, the devs did hint that it will come eventually

[cobalt gull]

Sorry, yes, I did see there isn't currently a publicly available api yet, that's why I am here asking about if anyone else has been able to get private access or if anyone has successfully been able to build another app on top of the model. Sorry if this is the wrong place to ask about that

[hollow hare]

Well, I mean *points at all the unofficial apps involving Suno and Udio in the play/appstore*

I do not think anyone will talk about how they achieved such a thing.

Not that it isn't hard, it's just most of them are designed to scam money from people who do not know about the site, and also poor taste about talking about it on the official server... and their methods break the service ToS and are actually illegal in various locations.

[cobalt gull]

Absolutely not asking about doing something unofficially! Definitely trying to give Suno money here. Your feedback seems really combative, if I've stepped on a historical minefield or something, again my apologies

[hollow hare]

My bad, just the number of times I have seen people who has been asking with those intentions, ugh...

As far as I know, there are no private access or anything like that at the moment.

The devs do have API access on their road map but plans of when to implement it is not known.

Likely around or after v4 would be my guess, especially if the team decides to make it so that the API commands can control smaller details that would not normally be available through the site.

In the mean time, what you can do is prep the app for its eventual release (e.g. setting up the UI, optimizing the UX, setting up a folder/sort like system, etc.)

[cobalt gull]

All good, appreciate you taking the time to share your thoughtts at all. Makes sense there's a lot of folks trying to rip them off (I wasn't aware of that). I work for a huge multinational fortune 500 corporation (not bragging, just saying there is a lot of $$$ here) that wants to partner with suno. My thought was that perhaps a private integration could be valuable to them, not only for the money it would bring but also potential exposure. Sounds like this is maybe the wrong place to reach out for that though?

[hollow hare]

Yep, this server is more for casual things to discuss among the public using members such as myself.

I believe you will want to get in touch with [e-mail]

Edit: see 3Daizy's message below

[gusty mango]

Hello, is there a full list of Suno commands somewhere?

[hollow hare]

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

[fleet edge]

I agree with you. I don´t think private integrations are doing any damage to suno. In fact I believe its totally the opposite

Developers with private integrations are putting own moneyand effort into these integrations, advertising , etc. which only brings revenue to Suno

its like saying that developers integrating with OpenAI via API are damaging openAI. thats nonsense

but I guess it´s an internal decission wether they want to officially open API for developers

[final hazel]

Feel free to send an email to [e-mail]. I know there are more companies interested and although it might not be available yet, it might help to already establish a connection with the developers about this.

[cobalt gull]

Makes sense. Our client's goal is to try to get something up in 2025 so who knows, their existing roadmap may be aligned with that anyway for releasing the API. But ye, we'd be super happy to provide with them with cash in exchange to be one of the first with official access. Hoping it could be a good mutual back scratching thing, I will definitely reach out, thank you again

[midnight oracle]

Why I can't put my own lyrics?

[still talon]

Use Desktop mod on your browser

[midnight oracle]

still can't

[still talon]

Ok, do it again refresh, need to be popup message after you press add my own lyrics

The popup is message you need to agree

For me is working I sure is working on all browser but I only try and use Google chrome

[midnight oracle]

i will try with chrome

Tysm

[still talon]

💖 hope it's be ok 🙏 update me lol

[still talon]

O and, is very cool system, try write some words and press make random lyrics, it will make song use your words and is make amazing stuff

[glacial urchin]

What prompt should use to generate a very fast rap song

[edgy river]

...

[nova ledge]

Is there a trick to getting a female vocal in Suno? I'm writing female vocal/woman vocal at the top of the prompt and getting all male vocals, I tried 'male vocal' to see if that would conversely generate a female vocal, but all males either way... Thanks!

[final hazel]

Just add it in style of music. Something like female vocals

[untold briar]

You can try repeating "female, female, female, female" as well, though it can sometimes influence the music in a not good way if you over spam a word. Better can be adding extra adjectives that are typically used to describe female voices.

[warped furnace]

Something like that: gothic rock, female vocals (in the style of music section). Female voice should work, too.
Sometimes it switches the genders without prompting. Don't know why. And on the rare occasion I get a duet. Would be nice when I want it intentionally. 😅

[cursive jetty]

i had that happen it switches back and forth - but turned out better in the end https://suno.com/song/248979c3-206c-4b8c-9d7a-3b996cfd5aed

[lone oak]

After the update on the 22nd, the previous bugs were not fixed. On the contrary, the sound quality became worse, the volume was also lowered, and the 20 songs generated were all with high noise. Why is it getting more and more outrageous with updates?

[unique stone]

What did they update on the 22nd? I don't see any updates about the music generation, just website features.

[cursive jetty]

the gate filters on vocals is now awful, vocal patterns on custom lyrics is awkward - seems a real step bak...

[remote flume]

This new update sucks. Please convert back to the one prior. Making song on mobile have been a nightmare so far

[lone oak]

Not sure, I haven't used it for half a month. I saw an update yesterday, but today the experience is very bad. The vocals are constantly distorted, accompanied by a heavy background noise.

[muted jay]

I've spent almost 1K credits on a song because the AI does not seem to be recognizing the [Chorus] tag. It keeps singing the chorus exactly the same way as the verses. Anyone else had this?

[lone oak]

What problem happened to you? This update is so outrageous

[cursive jetty]

ditto - same here killed 1k of credits just trying to get 1 track - last month every track on point

[muted jay]

Yes! The quality of outputs last month was so much better

[cursive jetty]

tried re using prompt from last version now and its utter dog shit

[remote flume]

I can't see the time on the song, when trying to see where I need to extend the song from. So I have to listen to it first and make sure I know exactly where I want the extended time.

[lone oak]

The songs generated by this update have such poor sound quality that they are unusable. It’s time for the official to wake up.

[warped furnace]

Not that. But it cuts out the first two lines of the chorus the rest of the song. Chorus 1 ist complete, the next one without lines 1 and 2. 😅

[remote flume]

Before that,I was able to see the song timing at the same time.

[muted jay]

I've had that too

[cursive jetty]

amen to that

[remote flume]

Also, if I decide to extend multiple song, only the first extension start at the time I pick, anything after that start at the end of the song I'm extending

[cursive jetty]

did you retry a ppropmt from last month in this version to bench? its so way off??

[warped furnace]

It's infuriating when the rest is what I wanted. Would drive me mad if I was a paying user. 😅 Just discovered Suno so I'm trying out. So far I'm reluctant to get a paid plan.

[muted jay]

I did. Huge difference

[lone oak]

Congratulations on finding a new bug🤣

[cursive jetty]

youd think with all the new revenue and subscriptions they wouldnt degrade really

[lone oak]

I'm a paying user. I've been going crazy for more than a month. I finally waited for the update, but the new one turned out to be a mess. I still have more than 9,000 points. It's really infuriating to face the generated mess.

[unique stone]

The AI model ("v3") has not changed since it was released toward the end of March. This is what generates the music and determines the quality etc. They have not updated this at all since then. Poor quality has been an ongoing problem for a while. I have started posting examples in #🪲┃bugs-web but this really isn't a new problem (but it is a problem, no doubt about that.)

[muted jay]

I used to be able to create 1 good song a day last month when I wanted to and my 2500 credits would go a loooong way. But this month I've been stuck on 1 song for a week now and can't hack it because of the bugs. And this after I upgraded to 10K points 😦

[lone oak]

It’s too outrageous. How about SUNO officially change the points to permanent? This update is simply a critical blow. Are you ready to make old users quit?

[final hazel]

It's correct what Furrybeans says. The updates were on the UI and not on the model. Model will change soon though.

[cursive jetty]

feel you... im same here, doesnt seem to make any difference - out of the last 20k credits the quality and consitency is off, gate, levels, noise passes, vocal flows are all worse - wether thats the bass model swelling with hallucinations and glitches or bad customisation flows - i dont know - but product is worse

[cursive jetty]

out of the last 20k credits the quality and consitency is off, gate, levels, noise passes, vocal flows are all worse - wether thats the bass model swelling with hallucinations and glitches or bad customisation flows - i dont know - but product is worse

[lone oak]

However, the sound quality of more than 20 songs generated today is extremely poor, and some volumes are even only half the volume of normal songs. Such poor sound quality has never been continuously generated in the past.

[cursive jetty]

feels like the model is warping

[lone oak]

Are the officials using users to train models?🤣

[remote flume]

Haha, but do I get refunded my credit for those wasted credit 🤣

[final hazel]

Once a model is made, it's not changed. Ai is very unpredictable in general. I like to share you guys a post of me where I have the same feeling as you do now. And here is the kicker. It was with V2.

[muted jay]

Ditto

[unique stone]

And here's a developer responding to me to confirm that they do not change the models.

[final hazel]

and another one.. again.. this is V2

[muted jay]

Stayed up until 2AM to get the AI to sing the chorus lol

This is ruining my life

[final hazel]

Ai is unpredictable, but it becomes better with each version. This is just the way it is atm. Nobody is making it worse. The model is the same.

[lone oak]

Probably not, the official is not that generous. My advice is not to get too addicted. Experience tells me that the same set of lyrics will only get worse if it is constantly generated.

[final hazel]

I personally was FULLY convinced in the past too, that they changed something. As soon as I paid, I felt the quality was worse. I even had conspiracy thoughts. And funny thing, you see it now happen to competitors too. First they were free, now they cost money too and people have the same responses as you guys have. But it's really not the truth. You cannot expect perfect outputs every time. It's a mixture of luck with the right settings and lyrics.

[cursive jetty]

hmmm having been in since the start - -the customised lyric fits, rhythms, flows were spot on - maybe 20 renders to get a full track that was on point. now ive blown 1k and cant even get the first vocals to flow on the lyrics. the model is ongoing, changing, training.. but right now its off

[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 prior to the continuation point to learn/mimic the styling you had (which is why extensions are 1 min long as the other min was reserved for studying).

This is probably why the first voice could not be picked up, it may have been too long since the 1 min reading from the continuation point

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

[final hazel]

Have you tried to redo old songs? Just to check how it responds then? Because lyrics can do a ton too. I often have to tweak lyrics because it doesn't do what I want. But a good test is redoing old song.

[lone oak]

Do you also experience frequent vocal distortion?

[hollow hare]

More like who doesn't 😭

[cursive jetty]

aware of that, quick reset purge clears all - the vocal flow patterns are off though, even when giving it easy phonetic rails its out. all tracks with any level peak, degradation get bad reports..

[hollow hare]

IIRC, a user resorted to making very short clips (like upto 30s) because they noticed that it keeps the quality high.

In some of my songs, I noticed clear degradation by the 2 min mark, many typically started around the 1 min mark (though that is likely do to the chorus happening combined with the genres I was testing)

[cursive jetty]

sure did - first thing to do.. latest version cannot flow the lyrics back in. i ran a series of test last month using 1 set of lyrics and then built 12 different track styles to see how far it could stretch. all outputs were able to flow... its not happenong now - really weird

[lone oak]

The sound quality today is simply unlistenable.

[cursive jetty]

https://tenor.com/view/in-pain-dr-gustav-zoomer-abbott-and-costello-meet-the-mummy-pain-in-the-ear-my-ears-hurt-gif-18797890

[lone oak]

Maybe it's a little better than the tape I've listened to tens of thousands of times🤣

[final hazel]

The luck and predictability is also a thing. And like I said.. even competitors who are now also on paid plans suddenly have conspiracies that this is a fraud, scam, and done to "sell credits". I know it is not, because I see it here too. Luck is a huge factor. Is V3 perfect? No.. I made a meme about it here: #😜┃suno-memes message
But I know in a very short amount of time there is an update coming that you all will appreciate. I know it, because I heard and worked with it.. just stay tuned. 😉

[cursive jetty]

look forward to it.. keep me posted. suno.com/@tokaris -we took the last set out and played it live in Brazil, NY, London, Italy

[restive marlin]

Any one know if its posible to play music from suno on discord bots ?

[low remnant]

Did rename vanish or get moved?

Oh, probably in details.... lol I swear I looked several times for likely candidates

[final hazel]

Edit details will do the trick 😉

[teal raft]

My muscle memory is slowly adjusting

[maiden token]

Do I lose my credits if I don't use them all before the rebill date?

I had almost 2k still and I canceled my sub since I didn't plan on using more than 2k at once and I just logged in after a month or so and all I have are the 50 free credits. :C

[final hazel]

Yes, they do not carry over. As it says here (on the subscription page)

[maiden token]

Oh no, that's super unfortunate. :S
Thank you for responding so quickly!

[low remnant]

LOL, ok @final hazel , give me the information you don't have. Wen inpainting? LOL I have a perfect chorus in a song except for one mispronounced word and no matter how many credits I throw at it, the RNG does not want to give me a good revision. I know you can't answer, but blink twice if it's before I my credits renew. 😛

[final hazel]

😬 Keep in mind. In-painting will be for V4 only..

[low remnant]

Well poo

[final hazel]

😐 I know

[low remnant]

Time to spam extend then lol. No matter how many times I'm like "no one but you will notice". I notice and can't unhear it

[final hazel]

Clumsy drops a link #📚┃suno-school message

[low remnant]

Grrr. Hmmm, I wonder if I can shove her hard enough in the second chorus to get the pronunciation right there and fix the first one in post. So far she's locked onto messing it up every time. But I have my ways.

[final hazel]

Sometimes you got to put it in a playlist for the next day, and do something new 😐

[low remnant]

LOL, this is my third return to this song. I had to browse through 30 pages to find it again because I forgot to drop it in the "In progress" list.

The answer is "Yes!" I can bump her into the correct pronunciation in the second chorus. Not every time, but much higher ratio than getting Suno to recreate half the first chorus the way I liked it from scratch! I might just take that W for the night and come back to it with some fresh eyes tomorrow.

[rose crane]

Okay so I have a question for ya'll. Maybe it's just me, but has anyone else felt that they got better results when v3 was in alpha? I'm experiencing a lot of...tinnyness? Or just simply degraded audio the longer a song goes on now and I don't really remember having this issue before v3 went live.

[low remnant]

Yes. I loved v3a. For the music I generally make, it gave me better results.

But I'm hoping I love v4 even more betterer.

There are strategies to avoid the loss of quality, the reverb, the tinniness, but most of them rely on vigilance and aggressively avoiding any extend with a hint of them no matter how good they sound otherwise.

[rose crane]

then maybe i just suck at my prompts. I just can't figure out how to avoid it

and it's just really killin my vibe, man

[low remnant]

I feel ya. What type of music you make? Some genre's are hit harder than others. Dreampop tends to be at the bad end of the spectrum IMO.

[rose crane]

mostly rock with synthwave mixed in. just finished a Titanfall 2 album and there are a couple of songs that could've been....well shit they could've been a bit better

[graceful condor]

how do i make it where the beat will get softer for the outro for instrumental

[low remnant]

I'm really hoping for v4 soon. I'm on the 10k credit plan, paid for a year. And finished last month with 7000 unused because I was tired of murdering good ideas when Suno kept turning them to poo.

[rose crane]

yasssss

i mean still it's amazing that we can even do something like this in our lifetime, but like....c'mon.....

[low remnant]

100%. I've found trying a new sound allows me to still create some stuff. But some of my v3a creations, I was really proud of. It give me ideas above my station or something.

[rose crane]

it's really making me wanna pull my hair out....and i'm bald....

[low remnant]

LOL, I tried everything to convince the devs to leave it as an option. Like I'd pay double credits to use v3a again. I might even pay 5x. But it's gone and never to return.

[eager topaz]

Most definitely. I haven't generated a good song for more than a month. I am only doing Instrumentals because the voice quality in songs is so poor. I have just tried again today with an instrumental and the 1st try is bad. Distortion and 2nd gen was just 18 seconds and music was totally different to 1st clip.

[rose crane]

so wtf....

like how do we address this?

again i'm super grateful for the opportunity to do stuff like this, but how can a version degrade with coming live?

[eager topaz]

IDK. Devs are aware of the problems, but not sure if they are trying to fix or just concentrating on V4.
I'm just wasting my credits and I signed for 12 months.

[low remnant]

TBH, I think v3a slept with one of the devs wives.

No idea why, but they seem to really not like it

[rose crane]

Lolol

[low remnant]

Originally I thought it was just inefficient and cost too much compute to run. But if it were that simple, charging more to use it would be a logical step.

[rose crane]

I mean I get it. I really believe in the project and I want it to succeed.

so devs if you're listening...like...what's up? what happened? are ya'll okay?

[low remnant]

Absolutely, all the shit I give Suno comes from a place of love. I never knew I had the desire to make music before I had the ability to do so. But to briefly feel so connected to the universe and have it taken away is very disheartening. I know others are unaffected. Or make better stuff with v3.0 but not me

[eager topaz]

Pre 3 I created some great songs, ones I am proud of, and some amazing instrumentals that I love listening to and they are crystal clean quality. Now, all has changed and I am with opposition AI creating once again, some amazing quality songs and instrumentals.

[rose crane]

see but I don't hate it enough to stop paying for it 🙄

[low remnant]

I cancelled my sub, but it's mostly symbolic, since I'm in until Feb 2025. But if things don't improve for whatever reason, or I legit run out of ideas, then I didn't want to get charged for another year.

[eager topaz]

I'm going burn through this month's credits as fast as possible to produce lots of reportable clips/tracks

[hollow hare]

I recommend generating a number of instrumental tracks so you can extend when you got some really good ones.

(I recommend using various instrumental tags rather than the instrumental switch option as you can customize what/how the instrument plays, and importantly, easily stretch the first gen to 2 min)

[graceful cliff]

This is some real sh*tposting

[eager topaz]

I've never needed to do anything like that before V3 . I have many amazing instrumentals with many mixed genres. Excellent quality, release ready, with minimum mastering. Along came V3 and for me, at least, changed everything.

[surreal oxide]

Hi, guys!

Is there any way for Suno to replicate a song he has already created, but without the vocals?
Or, is there a way to produce a continuation of a sung song giving instructions to do the same thing, but without the voice?

[teal raft]

To uh, country anecdote whatever the strange, almost deity-like reverence v3a had, v3 itself has been fantastic

Not that I want to say it, but I suppose I must: skill issue

[teal raft]

No, either with a song that exists already outside of Suno nor inside Suno itself

If it's your song, download it then run it through any number of vocal remover tools that are out there

You have to do this outside Suno, of course, but its an established, common thing

[surreal oxide]

It's how I'm doing it at the moment, but the quality reduces a lot. Can you recommend some process to me?

[teal raft]

Unfortunately no I personally am not well-versed enough to be any sort of authority on it

I do know lots of people like https://www.bandlab.com/

Though I haven't used it. Let me see if I can find the one someone was suggesting in this chat, give me a moment

[teal raft]

https://vocalremover.org/splitter-ai

I think this one is free

I don't think these should be reducing the quality

[eager topaz]

I use RipX DAW Pro, reducing to stems (individual tracks) but the instrumental tracks remaining without the vocal track are not really good quality. Not really good enough to use to add a different vocal track.

Not at all comparable to studio-produced tracks, if that is what you are wanting.

[surreal oxide]

I've heard of both. I used Audacity's OpenVINO plugin, and X-Minus, but the results are questionable.

Thank you very much!

One last question...
Would you know how to continue a song by making the entire band stop, except for one instrument? I want to repeat a chorus, but only with the drums.

[teal raft]

You're welcome.
Getting that specific control is, right now, essentially trying to prompt toward it and hitting generate to see what you get

Which is to say, no, there is no way to realistically get there through prompting alone

however

v4 should be soon

I'd guess in the 'coming weeks' according to the announcement #📢┃announcements message

and with v4 it may every well be possible to get what you are after a lot easier, but I can't say for sure if it would

Since even with infill/inpainting I would imagine it would just be prompting a specific section, but still just prompting

Unless we get blessed with more fine controls somehow

tony ain't saying outside of doing a mens warehouse "you're going to like it, I guarantee it" pitch

[surreal oxide]

I'm dreaming of the day when it's possible to download instruments separately.

[teal raft]

from what they've said before they'd like to, but since Suno doesn't work in stems it would be some sort of seperation they do internally

But the quality isn't up to snuff to put 'Suno Official' on it

But again, who knows what is coming with v4

[wise coral]

is the how to make music channel down?

I cant read it, there seems to be no content that it can be loaded

[untold briar]

It's out of date in any case. Check the page linked from the Suno web site, and for more advanced help, can look at https://www.suno.wiki/

[quartz marsh]

Hello, I have a feature request. When making a song public, I would like to hide the prompt that was used to generate it.

[teal raft]

Why would you want to do that?

[quartz marsh]

Why should I have to disclose the prompt?

[teal raft]

A core tenet of Suno is collaboration and sharing

But a suggestion is a suggestion, and you should request it here #💡┃suggestions-web

[quartz marsh]

Thanks I didn't have that channel. We already have the option to either keep the generated song private or public right? Anyway I will move the conversation there.

[bright whale]

I noticed that SUNO has a really hard time changing genre for the extension. For example if my initial part is grunge and I want the extension to switch prog metal, it usually fails (badly)

[untold briar]

The best way to do this is to Extend From a time a few seconds before where in the song you want the song to switch, then Extend From that clip from 2 or 3 seconds from the start of the clip, with the new genre. This should get the genre to switch. Then you probably have to Get Whole Song on the latest clip to try and recover as much song memory as possible, before Extending any further in the new genre. Though if if you want the song to go off in an entirely new direction you can skip that. If you are going from grunge to prog metal, maybe you can get away with a little longer than 2 or 3 seconds actuallly, since it's not as hard of a shift.

[true night]

things 1 can look forward to into thus future , i think they pump out stuff at an very dast rate regardless them devs of suno amazing constant at work

[true night]

i did that on 1 of the mis typos tracks i have from last week its not pitch perfect i def hear diffrence im not ultrea pleased with but it dont have the typo so another alt was made amoung the sea of alts i have 😄

[wispy nebula]

Suno's quality is far worse than ever, now little more than useless tin can renderings. Time to quit and cancel this rubbish.

[raven hawk]

Does anyone know how to effectively force the Bark API to sing a text? I tried adding the song notes but this does not work:

data = {
"text": "♪ In the jungle, the mighty jungle, the lion barks tonight ♪",
"voice_preset" : "v2/en_speaker_4"
}

[untold briar]

Maybe try:

[music][music][music][music] ♪ In the jungle, the mighty jungle, the lion barks tonight ♪ [music][music][music][music]

Maybe even double up the ♪ to ♪♪

**Oh, don't use v2/en_speaker_4. **If you want Bark to sing, you will have to use either a singing voice, or just a random voice (don't pick a voice)

This server has kind of moved on from Bark, so probably Bark goes in Off Topic more accurately now. But close enough for now.

[raven hawk]

Do you have examples of "singing voices" IDs?

[untold briar]

There aren't any singing voices in the default prompts. Actually I think a few of the foreign voices will sing (in English!) but I don't remember English voices doing it. So you have to kind of make your own.

A Bark voice is created every time you don't specify a voice, so you just randomly retry until it works, then save that voice. You can DM me later, I am minutes from sleep.

[raven hawk]

Thanks, I'll play with it and see what it delivers. Cheers

Well, after a few tests I gave up, it's too unpredictable... I can't create something consistent with it... unfortunately.

[untold briar]

For music that is likely the case. But for TTS, regular speech, it can be consistent with the right voice.

[raven hawk]

Yes, but I'm looking for a consistent "Text To Sung Speech" solution...

For A Capellas

There are gazillions of outstanding Text to Spoken Speech APIs

but no good Text To Sung Speech one

[untold briar]

I'm pretty sure I could get a consistent "Text to Drunk Terrible Karaoke Singer" to work in Bark. I think somebody who really knew what to do could get it do sing in tune, because sometimes it does. But even in that case you will quite limited by the 14s generation time. So it is probably not possible to have long term musical coherency.

But pure singing with no music in the background, is more doable than full music/songs. You do get a good sample of that from time to time.

[raven hawk]

"Text to Drunk Terrible Karaoke Singer" 🙂 TTDTKS

[untold briar]

Haha, I mean it is kind of a novelty, I sort of thought about making that. It really can sing out of tune better than Suno or Udio, loool

[untold briar]

Even when you get it singing, it often really wants to do something else: #🐶┃bark-showcase message

[raven hawk]

It's easier to force Suno to do an acapella and then separate the stems if there's still some music...

Bark is a previous generation, far too experimental for professional use.

[bleak bobcat]

440 / 5.000
Any way to keep the choruses the same? I'm trying to extend a song and the chorus is totally different, just like the bridge above. Sometimes it gets it right, but almost always something changes. I know that with version 3 things have improved a lot and I just discovered it. I don't even ask about duets anymore because I've done tests without success, although on YouTube people extract the vocals and then change them to edit it later.

[hollow hare]

See #🪲┃bugs-web message

[low remnant]

Yes. And I apologize. I just never did understand the anti-v3a sentiment even as an option, even as a premium. Those of us who loved alpha really loved it. And I'm sure we're a really vocal tiny minority. But it's real and v3.0 has presented some real challenges to continued creation for us. So, while waiting for v4, what else we gonna do? I still don't enjoy that competitor that I won't name. So passive aggressive discording it is.

[hard grove]

hello, I make instrumental music and I would like to know if we can indicate an end to a piece of music, for example my latest creation I am apart of 7 and still no end and I would like to finish it... 🙂 Thank you for your reply

[restive steeple]

Guys, has anyone noticed that now it started to generate as if from half of the track. It used to generate from the intro, everything was cool.

[autumn jolt]

I mean, it's not great to passive/aggressive S******, but I agree that 3a was better than 3. That's what it felt like. I've actually paused creating new stuff waiting for v4. It's just too frustrating churning through so many more credits than 3a to get the quality I felt I was getting with 3a.

[cold zenith]

what was better. ?

[low remnant]

I don't even care about credits. I think I said I'd consider as much as 50 credits/gen to use the v3a algorithm again. What I hate is taking my few precious beautiful ideas and sacrificing them to the v3.0 engine. And that's what it feels like.

[cold zenith]

what is the main difference between the alpha and release oneV?

[low remnant]

Besides, it was just a joke. I thought it was actually funny myself but I apologize if it wasn't taken as such

v3a was quirky. But

1. it never (for me) devolved into the reverb mess that I tend to accidentally trip and fall into with v3.0
2. the performances were (again in my gens, all I can speak to) more emotive, I could really wind the singer up

Really, those are the big ones from my perspective with 1 being the biggest. It's really sad to have a song you thing is going to be beautiful and then 3 extends in find some echo you didn't notice in the first segment has amplified and fed into itself and now you can't complete the song.

[cold zenith]

Yeah, the dev knows already the #1, i and others talked a lot about it. I think they gonna try to resolve it in the next update.

I'm not gonna lie, first one has crushed a lot of good songs. But we got to remember that, that product isn't at his first year. We have to be patient.

[low remnant]

And I'm waiting for the next update. Just less than patiently. I wanna create me some music! I need my fix.

[cold zenith]

I trust them for resolving that problem in the futur.

[low remnant]

And it's hard not to feel like they are withholding the one thing that they already have that would uncork my creativity. I just want to use v3a again. It's still there somewhere, fire it up and charge me more for my weird engine fetish.

[cold zenith]

yeah but it was an alpha, so it is a non sense in a logic to bring it back. They won't paid other server for 3a. They have to be focus on 1 thing at the same time.

[autumn jolt]

Fair question. This is a little bit vague but 3a felt more expressive/creative. Basically, I could get the initial 2-min clip I liked within 10-20 generations. Whereas now I feel like it can easily take 50+ generations to get what I'm looking for. I accept this could be bias, or being more picky, or even overcomplicating prompts/tags compared to before leading to a worse outcome.

[low remnant]

I'm not asking them to improve or maintain v3a, just leave it up until v4. ... and yes, I understand it's not going to happen. And that businesses don't tell people why decisions are made. But I would really love to undewrstand this one

[broken lotus]

Alpha couldn't end songs. And only knew 2 genres Pop and Country Pop

This eulogy is bordering on science fiction.

[low remnant]

But I love me some pop

[broken lotus]

It is 100% possible to PROMPT v3 for Pop

but whatever.

[final hazel]

I said it before, and I will say it again... If they would bring V3a back, there will be complaints saying "**it's changed! this is not V3a! **😠 ". You all are romanticizing that model. There were a lot of problems with V3a too.

[broken lotus]

As long as I am not forced to have every prompt turn into lyrical pop...

[cold zenith]

Because the way to prompt has changed. If they add more genres etc, you have to be more precise etc. If you put rock ( how many types of rock are there ), the same goes for gospel, soul, jazz, etc. So the A. I have the knowledge of all those things and if you don't know how to talk to it, he'll choose something that's close. ( but sometimes it goes totally wrong it's true )

[low remnant]

I'm also NOT asking v3a to replace v3.0. I'd just love to be able to toggle into it.

[broken lotus]

Agree. more models is... MORE MODELS and 100% a good thing.

[cold zenith]

she meant that, u'll say that it has changed, even though it has not.

[low remnant]

I say a lot of stuff. No one pays attention to me, nor should they. LOL

[daring tusk]

Is there a way I can tell the ai which verse to sing in a girl's voice and which one in Guy's Voice?

[cold zenith]

but more server than...and the price could rise.

[broken lotus]

money BOOOOOOO

xtra models YAAAAY!

[cold zenith]

put a mike and laura

[daring tusk]

If that's a joke then I don't get it

If not please explain how lol

[cold zenith]

[laura verse], [ mike √erse]

[low remnant]

That's why I suggested making it a premium model. That could even open up that business path for Suno in the future, having models at different capability levels and different costs.

[daring tusk]

Okay thanks

[cold zenith]

for it can know this is a male and female

[daring tusk]

Imma try it out

[cold zenith]

or the most common male and female name

[cold zenith]

hum

[broken lotus]

Any guess how big the models are?

[low remnant]

Honestly, no clue. (trade secret, lol)

[broken lotus]

🕵️

[blissful rampart]

I stumbled on a song awhile ago on suno that had a 'live audience' singing along. I've tried putting in similar prompts of "live music with live audience" etc but nothing generates where an audience is included. The one i heard even had the voice say "Cat hear you!" etc. I have tried to replicate it and it just doesnt do it. What's the trick?

[autumn jolt]

Yeah, it's not so much getting the wrong genre, I just felt like the output was better, sooner, without wading through so many clips and burning so many credits. I made a whole album with lyrics i really liked on 10k credits and now it feels like I need to use about 2k+ to finish a song I like. This is for songs with vocals, though, instrumentals don't take as many.

[blissful rampart]

I'd like to do a "live version" of one of the songs i created in a play list for a bit of a laugh

[glacial scarab]

10-100GB+ 🤷‍♂️

[blissful rampart]

This one https://suno.com/song/99aad3b2-0ff5-4042-bbf9-b39c4ed580ce

[low remnant]

Probably post processing. I managed to get an audience participation song - in v2 of all things - but the quality was terrible when the audience was singing

[blissful rampart]

i notice the lyrics on that one have ** at each end

Is there a list of commands that generally work? I have tried things like (fiddle solo) or (drum solo) but it doesn't seem to eventuate

[low remnant]

Oh, it's an actual straight suno song? well, in that case it's just luck, prompt, and credits. I've burned so many on those quirky ideas.

[autumn jolt]

I'm definitely not romanticising, just sharing my feedback that 3a seemed to work better for what I wanted to use Suno for. Like I say, I accept this is subjective, vague and very possibly biased. I agree you can't please everyone, but right now we can use V2 or V3, I guess some people would like V3a as an option too.

[cold zenith]

You're right on this.

[blissful rampart]

@low remnant i pay for suno and yeah, will burn 300 credits trying to get a decent mix lol. I wish i could take a song ive already done and then say "make this live"

[low remnant]

Time for a new conspiracy theory. I'm going with "legal issue". v3a was too good, copied some performers style too closely and they got a cease and desist, and now we'll never speak of v3a again. 😛 Yes, more trash posting, but my trash is so tame.

[autumn jolt]

On the subjective of versions, I've always worked on the basic that software goes alpha (test), beta (public test) live. It surprised me when V3 went straight from alpha to live (and that an 'alpha' was available to all subscribers in the first place). Is this just a Suno thing or a trend/change to how software dev is labelled?

[low remnant]

Maybe in v4, you can pull that off with careful inpainting. Hey! we get to keep the before and the after with inpainting, right?

[blissful rampart]

I would like the ability to empty the trash bin with a select all and delete. I wonder how much wasted storage/server space it's taking up!

[autumn jolt]

I definitely wouldn't say V3a was 'too good', I generated a whole load of garbage with it! But I got some really great stuff too.

[low remnant]

To have the studio and live songs if we did @blissful rampart 's idea?

[blissful rampart]

like ive just done 10 'songs' to simply delete them because its not what i actually wanted

[low remnant]

💯 it was hit or miss, but towards the end, I learned pretty well how to communicate my intent to it and got more hits than misses

[blissful rampart]

I am having a laugh with Ai. dont get me wrong. But i have a personal joke goal of pulling together a funny album and having it listed on spotify LOL

[autumn jolt]

Yeah, it's annoying how that static background grows and vocals get worse across extensions. Especially if you've finished a song you really like. Only thing for it is to go back to before the background artefacts started and extend from there and rebuild. Or if there's something later in the song you love, live with it. Or export clips and do some work editing stuff together if you can.

[low remnant]

... and I'm going to admit, I'm a bit bitter over the rollout of v3.0. I had a song I had been working on for days, I was one continue from completing it. Went in to work like that, came out and it was v3.0 and it would not match the style to finish the song and I had to toss it.

[blissful rampart]

yeah ive had a few oddities too

Things like i have part 2 and 3, but it cuts off. so i extend and type (outro) and it then has a minute of silence. Or will do something completely different. last night i reported 12 'extends' as a bug

[low remnant]

Yeah, I'm tired of having to be constantly vigilant for the arrival of the artifacts (noise, reverb, echoes, choir, etc). That competitor I hate, I made a 6 1/2 minute song on there to prove to myself I could do it, and the last segment was as clear as the first, with me not focusing on cherry picking in the middle at all. I hope that is the experience in v4. I know it's possible to do long songs in v3, but having to be on guard all the time against the quality fall off is tiring sometimes.

[autumn jolt]

It makes things easier if you're using Suno tracks and sharing them to somewhere like YouTube. Because if you have a good outro that's 15 seconds, then silence, then it starts up again, you can just grab the whole thing and end the song in the video editor where you want to.

[blissful rampart]

i could use audacity and mix/edit them

tech question: the Ai model it uses, would it have been trained on real artists somewhere along the way? Or is it purely mathematical/algorithmic? I generated a purely random song and the vocal sounds like Bird of Tokyo. I simply stipulated "catchy country with male vocals"

[broken lotus]

well, music is basically algorithmic... AND every AI has been trained on real stuff that exists.

both of these things can be true.

[blissful rampart]

so then, if i release an album, stipulate it is 100% ai generated, even state what prompt was used, what is the comeback if a band says "hey that sounds like us"

[broken lotus]

you tell that band to try harder because an AI has just replaced them.

[blissful rampart]

LOL

[broken lotus]

🤣

[blissful rampart]

the terms and conditions for sites like spotify and youtube say you cant post Ai music if it copies an artist, and that you have to state its Ai. Ive gone so far as to generate an artist name with Ai. It's fascinating. You could (and easily) create an entire persona of an artist, back story, imagery and music. And they'd never actually exist

[azure saffron]

that was hilarious!

[alpine venture]

sounds like is meaningless. Certain chord progressions have been used over and over. Entire genres effectively have exactly the same beat in every single song

if you copy their song, that's different

like, lyrics, progression, etc

I would argue a variation would also be IP theft depending on a few factors

But like, if the words are different and someone is just saying stylistically it sounds like them. Tell them to gitgud

[blissful rampart]

yes 100%. What is also interesting about ai, is (in my opinion) it tries to sound realistic, and yet its criticised for not being a human. However it seems okay to have humans so autotuned and processed they dont sound real...... yet thats blasted over the radio.

[low remnant]

We're training the AI at this point. It's trained on data from itself with us 👍 or 👎 the results. So especially by v4 it's going to be almost all synthetic data. I believe, I'm not Suno LOL

[alpine venture]

https://suno.com/song/15df55f6-e62a-46cb-ad00-ef905ac3da02

[alpine venture]

bingo. Synthetic datasets are the future.

[blissful rampart]

So my challenge is to have an album, 100% ai, lyrics ai, artwork and band name/artist name and get it on spotify/youtube music. Thats why i am loving suno because so far i have 8 songs that i won't lie... ive listened to in the car and go "that is actually catchy and singable" Showed work mates who seem perplexed at "thats ai?" "yes... all of it."

[broken lotus]

The worst part of managing a pretend AI Artist is all the pretend AI assault lawsuits from the pretend AI groupies..., and the pretend AI hotel bills....

[blissful rampart]

and the pretend cocaine habit

[low remnant]

OMG they're the worst!

[strong falcon]

Just a question 🤔 Did your credits disappear at the next month if you don't subscribe another month ?

[blissful rampart]

ive made my artist a quirky country artist because why not. Its cringey right. So what addiction should he have?

[broken lotus]

gambling

[low remnant]

Actually the worst part is that I mostly cut myself off from social media, but the damned band has to have them and now I have to learn how to use them again

[blissful rampart]

hmm gambling. On what? make it funny. Gerbil races?

perhaps he's addicted to nasal spray

[broken lotus]

bidding on Etsy 1-of-a-kind knitwear

[blissful rampart]

LOL

well i did ask an ai site to generate a country singer in short shorts.

apparently thats explicit. *shrug

[low remnant]

Others might disagree, but I'm going with it's 2024, who cares about albums? Singles for everything!

[unique stone]

Pretty sure they only did this for the v3 alpha, to help train v3 (and evidently a lot of people thumbs-up'ed songs with layered vocals/choirs.) They might be using this still as general feedback for future model training but it won't change the existing models.

One of the team asked me some questions a few weeks ago and one of them was whether I use thumbs-up/down and what for. I mentioned that I rarely thumbs-down anything and I tend to use thumbs-up to "bookmark" something in my library so I can find it later. So I hope they aren't using this as training guidance.

[zinc stirrup]

guys, can I remove a song from trash?

I accidentially delete one

[blissful rampart]

i do notice the ai seems to lean heavily on harmonised/multiple voices for choruses. Is there a command to enter to stop that?

[broken lotus]

yeah, Suno expects us to use the thumbdown in some meaningful consistent way... 😏

[broken lotus]

you should still be able to open the trash and pull it out...
unless you deleted from trash, then it's gone-gone

[blissful rampart]

eg i dont want a chorus where the artist all of a sudden has 5 voices harmonising it

[low remnant]

Maybe they use which segments are extended. I'm sure there's valuable data in all the songs we make somewhere. AI people know how to tease it out.

[unique stone]

If it's in my library with no reaction: I don't like it
If it has a thumbs-up: I want to find it again later to explore some more
If it has a thumbs-down: Whoops, I must've misclicked
If it's in my trash: I've finally got round to deleting some stuff

[blissful rampart]

yes i agree, I think if its in trash it should auto thumbs down to the ai model.

its trashed for a reason right

[broken lotus]

to clear the screen?

[blissful rampart]

well you wouldnt trash a keeper lol

[unique stone]

I usually try to be very specific about the kind of vocal I want (pitch range like soprano or whatever, soft/powerful etc.) and split it across multiple tags and keep hammering it with more "vocals" stuff so llike "soprano female, soft vocals, female vocals, clean vocals"

[blissful rampart]

i also wish, if we pay for suno, you could copy lyrics in for a whole song and it generates it. makes it a song to cover all the lyrics you put in. Free, sure cut them off at 2 mins.

[broken lotus]

there's not a great way rn to work with a lot of clips on 1 project...

I try to generate and dump everything into a playlist..., then sort from there...

I end up temp thumbup/down to remind myself what I've heard so far, then move to trash as I go when I don't need the clip ....

it's not 'folders' but... idk any other way except to just flood the library.

[blissful rampart]

But i find it frustrating to see it cut off mid verse. extend, repaste in, rinse and repeat

[zinc stirrup]

I'm sry for ask, but I didn't find trash

[blissful rampart]

icon up the top right

[broken lotus]

it is the tiniest icon in the siberia of the website, LOL

[blissful rampart]

it must take up so much hard drive space for suno.

i wouldve thought maybe the trashed files would auto delete say after a month

[broken lotus]

I haven't actually deleted the trash...

[zinc stirrup]

wow, thank you bro

[broken lotus]

because issues.

[blissful rampart]

me either. there's probably 5 pages of trashed stuff. Imagine that times what few hundred thousand consistent users?

[low remnant]

My library is total trash. I should be embarrassed

I found it easier to just make a playlist called finished and never look in the library LOL. We don't talk about the library

[broken lotus]

it makes it so hard to find stuff...

Like I just made that song last week... how can there be TEN JIBILLION CLIPS since then...?

[blissful rampart]

im so pleased they added a search function

[broken lotus]

Searching our own unlisted clips is coming soon™

[low remnant]

I had to page through 30 pages of songs The other day because I forgot to stick it in the "in progress" list LoL.

[blissful rampart]

website polish like that should be straight forward for a web developer surely

[broken lotus]

I should spend some time searching for my usual prompt words... see how other people are doing...

[blissful rampart]

good idea

another question in terms of custom create. when you want to include an outro or intro, or solo, is it square brackets or curved?

aka (Outro) or [Outro]

[unique stone]

I started renaming anything I cared about to be a word and then just a number at the end. Nothing fancy, just 1, 2, 3, 4, none of this "ext. 1-1-2-1-2-1-1-2-1" business that others have 😄

[low remnant]

I've got one in the works you might appreciate. I was especially proud of the use of the words pickle and jar.

[blissful rampart]

ive managed to get suno to sing the word butthole actually

[low remnant]

My system is similar. I use a -number if it's an extend, and a .Number If I get whole song before. But then I can backtrack the whole tree when I need to

[broken lotus]

the () are for sung ad-libs... usually backup singers or the main singer interjecting at the end of a line (ooh baby baby)

[square] are for metatags... that's usually 3-word max instructions for the AI...
[fade out, end]

[blissful rampart]

ahh gotcha

so you could do [instrument solo] and stipulate an instrument? kazoo for example?

[low remnant]

Sino doesn't do kazoo LOL I burned so many credits trying!

[blissful rampart]

Well that needs to be rectified LOL

spend your $125M investment on more cowbell and kazoo

i wonder if cowbell is in there. hmm

[broken lotus]

Priorities dammit!

[blissful rampart]

i really want to have the ai vocal say "everybody now" and they sing along with a cowbell in the background 🤣🤣

[broken lotus]

I have gotten cowbell..., but not by prompting for it. 😓

also 'theramin' is not a word it knows, but it will play the sound under 'Halloween'

[blissful rampart]

is there a wiki of commands/instruments/things to try?

[broken lotus]

here's a big ol' list compiled under v2...
https://www.suno.wiki/faq/style-and-lyrics/styles-and-genres/

and lots of other stuff on the community wiki.

[low remnant]

Three is. But I never add anything because so many things work exactly once and never again

[blissful rampart]

ive noticed that

[sacred socket]

Don't know about you, but the more iterations/extend I do the poorer audio quality I get. After six to ten extend/get whole song efforts, the audio isn't passable to what the song was at the beginning.

[blissful rampart]

itd be nice if you could select particular voices for consistency. Aka if (like me you are making an album) you could 'reuse style and vocal' and then generate new lyrics etc. But it'd keep the same algorithm for consistency

[blissful rampart]

yes! i was getting annoyed last night. kept trashing the 2nd and 3rd parts because they would get worse and worse. distorted, errors, chopped words. Its like the Ai gets tired

[sacred socket]

👍

[low remnant]

It uses the previous segment to make the next one, so every error that creeps in is replicated in each extra segment, like photocopying a photocopy.

[blissful rampart]

yes but i have noticed say by the 3rd or 4th attempt of generating an extend, the vocal is way different, the music tonality is far off.

so last night i was mucking with a sea shanty. Just needed 1 more verse and chorus and done. I spent 20 mins and blew so many credits because all of a sudden it turned into a rock song

[low remnant]

I dropped it in suggestions (hear that mods) but never saw that any dev noticed. I wish we could use a reference point in the song that mostly locks in the voice generator and instruments, like the eyedropper in photo editing apps for picking colors but for voice and instruments to keep consistency for the whole song

[blissful rampart]

yes thats a good idea.

[low remnant]

That'd be a killer feature

[blissful rampart]

or just a simple "use this model" tick box

so it locks in the voice style

That way you have consistency between songs if you plan to be an idiot like me and create an ai artist 🤣

[sacred socket]

Seriously, it's like sitting at a one-armed-bandit/ poker machine, You keep putting your nickel in because you had a good win earlier. I really hope we aren't the victims of Suno discovering how much more money they make by NOT refining this ai.

[blissful rampart]

maybe. but like anything, they will have investors expecting development and changes to keep the business going

otherwise they will stagnate quickly

[sacred socket]

Investors care about returns. Is there a 'white paper' attached to Suno's ai implementation? Suno is currently the perfect model for attracting creatives and keeping them spinning / gambling for that next win.

[blissful rampart]

no idea.

[low remnant]

Suno has a real competitor and more on the horizon. The devs are working their asses off. And probably don't want to announce much of it because that just helps the competition out getting started matching them. But I really can't wait for some hot v4 action.

[sacred socket]

you mean a fruit machine with ,,,,, eight reels😅 ???

[low remnant]

And mini games. Mini games are where the money is!

[sacred socket]

hehehehe, and buying tools

[blissful rampart]

and building castle walls for microtransactions

before the Ai zombies come

[sacred socket]

that's the word i was looking for, microtransactions. I don't like to sound cynical, but why kill the golden goose by upgrading the income model to something that is more expensive to run?

[blissful rampart]

it annoys me when companies do the microtransactions. Its a silly way of thinking. Offer a product, people buy it. dont make it 'free with a catch'

[sacred socket]

I don't think we need worry about that for a while with Suno and Udio. What I do find odd is these two leading audio ai companies have identical user interface setups and basic operation. I don't believe in coincidences. What's going on there/

?

[blissful rampart]

its probably the same somewhere. Doesn't bother me, the entire world is probably run by 10 companies

[sacred socket]

well, that's the sort of disclaimer that ends 'what does it matter, we're all gonna die soon'. It does matter that ai like this is transparent in its relationship with the public. We all know how powerful ai is in this phase. I'd like some reassurance I'm not the patsy in some pay to play monopoly.

[low remnant]

Wait until they find out those 10 companies are all subsidiaries of one ...

[blissful rampart]

Who's a sentient head in a jar. Lol

I just want Futurama. Sorry

[low remnant]

I'm waiting to upload to San Junipero. I think that's the best we can hope for at this point LOL

[blissful rampart]

You wouldn't upload anything via Aussie internet that's for sure. 😂

[sacred socket]

I'm in WA, what do you mean not upload via Aussie inet?

[blissful rampart]

Lol Aussie internet is some of THE worst in the world.

[low remnant]

I assume because you gotta cross so much water to get to any real servers.

[sacred socket]

oh, yeah, ahahahaha, and we're proud of it!!!!😅

[blissful rampart]

And our government stopped a national upgrade to fibre to homes. I'm lucky I was part of the test roll out. But we pay waaay too much for slow speeds

[low remnant]

LOL don't talk to an American about slow speeds in the home. There are people out there in the sticks on DSL speeds still. Thankfully I'm near civilization.

[blissful rampart]

250mb down, 25 up for $100 a month is ridiculous

[sacred socket]

That las liberal coalition made sure the NBN died a thousand deaths.

[blissful rampart]

Yep

[low remnant]

I mean like 20 down 2 up speeds. Y'all making me laugh

[clever bloom]

Anyone here have experience making Eurobeat style songs?

[sacred socket]

I'm on rural wireless bband with Vodafone. I'm meant to get minimum 7mbps dl and 3mbps ul speeds. The best I get is about 1.5mbps in streaming/general use. But run a speed test, Voda have worked out when I try to do that, and bingo, there' my 7mbps showin' like a champ.

[sacred socket]

try the Suno prompt wiki - https://www.suno.wiki/faq/style-and-lyrics/styles-and-genres/ check out the song structure tags page, too

[clever bloom]

I have been looking through those. I was hoping to find someone who had an example I could look at and listen to

[sacred socket]

try googling, Suno, eurobeat - https://www.youtube.com/watch?v=Wn-MlvbNPhA

[golden bolt]

are there any technical papers about the model

[sacred socket]

You might try asking something like co-pilot ai, it is gpt4, on ms Edge

[soft lintel]

GM Team, it looks like our account has not been getting the 50 credits per day as per the Free plan and hasn't for at least a few days. Any chance you can look into it or is there anything I can do?

[golden bolt]

I have no clue what you are trying to say

Are you telling me to ask chatgpt or are u telling me it’s using its api or what

[sacred socket]

Ask copilot if Suno has published a technical paper

[broken lotus]

they haven't.

they are a business, not academics.

(but we should poke them until they spill their secrets)

[broken lotus]

Suno is like Midjourney..., it's their own system, so they keep some things secret

[sacred socket]

A technical/ white paper often accompanies a tech business portfolio searching for investors

[golden bolt]

So what, it’s not uncommon for businesses to publish papers on their models, they just don’t open source it

[broken lotus]

m'kay you both proved me wrong but there's still no published paper, LOL

[golden bolt]

Ye lmfao

Well imma dip have fun

[sacred socket]

yeah, I asked the same question and didn't get a reply. A white paper with investors would tell us whether we're participating in an online gaming casino ( nickel fruit 'music making 'machine) or a bona fide ai development business. The fact that Udio exists with exactly the same public online interface and business model makes me wonder.

[glacial scarab]

it's probably a more advanced implementation of the AudioLM paper with lyrics added into the mix

[sacred socket]

Do you know where the audio LM paper originated? Who/how was it bought up?

[glacial scarab]

Before chirp there was bark and bark readme on GitHub references AudioLM

[sacred socket]

Two main companies , Udio and Suno have bought into the existing AudioLM tech, and are duplicating eachother's business online model. What that model currently reminds me of is sitting at a poker machine, putting my nickel in the slot and hoping for the next win.

[tepid widget]

Hi, I was wondering about the energy cost of running the v3 music generator algo on Suno servers, does anyone have an idea?

(I'm concerned about the CO2 emissions of this service)

[near tangle]

for those that edit their songs after generation, what Windows software do you recommend?

[neat field]

audacity probably has the most google search results for figuring out digital audio workflows; i use audacity and bandlab

[neat field]

Suno has not released any CO2 statements, however they probably rent from a cloud provider and many more of those are CO2 conscience

example:

90%
In 2022, 90% of the electricity consumed by Amazon was attributable to renewable energy sources```
source:
https://sustainability.aboutamazon.com/products-services/the-cloud?energyType=true

[tepid widget]

If they tell us their cloud provider, that answers part of the question, because it's proportional to how an inference (and all the training) will use it.
So basically we'd like to know how long and what resources are used for training and inference.

[broken lotus]

not sure they are training v4 under the same provider that runs v3... 🤔
both are interesting...

[sacred socket]

If you haven't used a digital audio workstation before, it's a bit of a learning curve, but Reaper is a popular one. https://www.reaper.fm/ it has a lot of free processing /fx/filtering/mixing plugins too.

[near tangle]

huh

[low remnant]

I downloaded RipXDAW. They have a really decent trial period so you can determine if it'll be helpful or confusing.

[dapper fulcrum]

in any neural network it is possible to establish the strength of influence of theses, so why is it not here yet?

[zinc stirrup]

Hey guys, If I want to add a intro in front of the song, is it possible?

[teal raft]

It is not possible right now, no.

You can only build songs going towards the end

Currently you can't add things before

[zinc stirrup]

and if I want to cut and delete some clips from the, is it possible?

[teal raft]

Also right no, no, not within Suno itself.

[zinc stirrup]

fair enough, I will use AU

[teal raft]

This is going to change with v4 release. Currently Suno is on v3

#📢┃announcements message

'the coming weeks'

[zinc stirrup]

oh, another question pls, I noticed when I extend the song, the lycris go wrong, but np, I accept it. But the lycris it shows is not same as AI singing, can I edit the Lycris Pad?

[teal raft]

Also currently no, you can't. I have a few songs where I took just the beginning hook and took the song in a different direction

But the block of lyrics from before stick with the song

[broken lotus]

🤔 "coming weeks" sounds definitive.

[teal raft]

Does indeed

[lapis cedar]

sorry if this sounds a bit silly but how would i go about to make something similar to this? https://www.youtube.com/watch?v=FCXLnLJymFQ
i'm making a heist game but i absolutely suck at making my own music

[low remnant]

Ok, serious question. Will v3 continue to be available after v4, like v2 is still to this day? I know I knock v3 mercilessly, but I do have a few unfinished songs I think will be completed well when I sit down and do it and would hate to not be able to.

[vestal rapids]

Is v4 imminent?

[bright whale]

Is there way, or a trick, to make the Extension lyrics be related to the original lyrics?

[fluid robin]

Trying to make ambient songs but with no instrumental best way to go around it ?

[vestal rapids]

you mean an ambient song with vocals?

[fluid robin]

mostly with Animal sounds and nature

got one but it only plays halfway and then instruments pops in

[vestal rapids]

ah ok, I haven't had much success with doing that sorry

[fluid robin]

https://suno.com/song/a28bfcd7-d144-4531-b416-91c823b1b1b3 as an example

trying to make the animal sounds stay the whole song

[broken lotus]

Tony confirmed today, yes. V3 will stay...
(he did not say 'forever' but yes, across any transition into the next model)

[vestal rapids]

is v4 imminent?

[teal raft]

#📢┃announcements message

Maybe, probably

Major updates in the coming weeks would lead me to believe that yeah, sooner rather than later

[vestal rapids]

fair enough, thanks

[teal raft]

If it's like before when they release a new model, it will just kind of show up one day

With an alpha for paid subscribers

But it might be that this time there are too many subscribers to make that a viable option

Who knows what they will decide this time

[vestal rapids]

we just wait and see 🙂

[broken lotus]

there are a couple of big things coming... one is soon-soon, like I've seen it working, it's unexpected.... another advance in the technology....

the narrower eta I heard for v4 is not my definition of 'imminent'..., it would be on my far end of 'coming weeks'... (but I'm an impatient person)

[vestal rapids]

I have endless patience, between ChatGPT, Aiva and Suno there is enough to keep me busy 🙂

[unique stone]

Hmmmm "unexpected" 🤔

Is this more "it wasn't expected this soon" or "nobody expects it because it's not something anyone asked for"?

Tony hinted a short while ago in the general channel with 2 words that suggest v3 might be getting updated.

[teal raft]

New vee three

[broken lotus]

I think it is kind of a sideways move, but maybe could also see it as a transition...

my analogy is when it was discovered the concatenated audio was a higher encoding rate... everyone guessed the next model would generate closer to that level...

the soon-soon is an expansion like that, that can partly be taken advantage of now... or, as soon as it is released...

[unique stone]

Starts excitedly gnawing on the leather pads of his headphones

I've never noticed the concatenated audio being a higher encoding rate. Tony has mentioned this before a couple of times, particularly about how the videos are encoded, but it's always 192kbps when I check. "Contatenated" = "Get whole song", right?

[broken lotus]

v2 generated at 128, but was concatenating at 192, so people guessed v3 would be 192...

[unique stone]

Ohhh gotcha, sorry I thought you meant v3 was generating at a higher rate

[broken lotus]

and it was a way to keep them backwards compatible

[unique stone]

Assuming it's also something related to the audio encoding, I'm gonna guess they're jumping directly to WAV (or FLAC) format. If an updated v3 is coming I'd also guess maybe the audio quality gets bumped-up in addition to trying to solve metallic ringing/reverb/choir.

That's partly wishful thinking though 😄

[teal raft]

Waayyy back but it was talked about doing generations before, even with v3, was 'more of a UI thing'

I think it was tony or georg that mentioned something

as in before the current track

[unique stone]

Ah, extending backwards

[teal raft]

yeah

It was just a passing comment and not a concrete thing and also I remember like I make music: I think it's good but ehhhhh

[vestal rapids]

I wonder if Suno will make the watermark reading tool available?

[unique stone]

Why would they?

[vestal rapids]

to enable consumers to understand where music has come from

[unique stone]

If they did, it'd probably be something via the website for you to upload a piece of music or paste a web address (youtube link) into. I doubt they'd let anyone have the actual software otherwise it'd be reverse-engineered. Even via a web page though it might lead to people discovering what's actually encoded in the watermark and lead to it being reverse-engineered and defeated.

[vestal rapids]

I guess so, however I think provedence is an important thing for consumers to understand in the age of generative AI

[hollow hare]

It would make more sense for YouTube, Spotify, etc. to identify and label songs as such rather than having a common user upload/check different sites themselves because there will likely be many more AI music services in the future (esp. from record label themselves)

[broken lotus]

the bar for making AI music is pretty low. I can singlehandedly out produce any so-called studio producer, lololol

The bar is for them to prove their legitimacy, not to label US. Genie isn't going back in the bottle.

Let them pass out Wizard of Oz certificates and pay for it year after year.

I'm over here, doin MY thing.

[teal raft]

In another community I am a part of, non-AI, we were talking about AI because hot topic

and someone said they "weren't creative enough to deal with AI"

as in they were used to working in accounting or whatever they did normally

and this AI stuff scared him precisely because you can get it to do so much more than before

I get handed the keys to machines that can make me whatever I want and I go "hell yeah" and run off to play

This guy, and I suspect many others, just sit there and go "oh no"

It was just an interesting and novel way someone expressed their fear

[broken lotus]

too much creativity...

too many options

[teal raft]

Yeah

Whatever the phenomenon is when you look at Netflix and have so many choices and yet you just kinda look at the same screen

[broken lotus]

it's real.

[flat robin]

Man, tell me about it, lol! I've been sitting on my newly-refreshed 2500 credits for a week now, unable to decide on a song to create. 🤦‍♂️

[graceful condor]

@unique dome 3.5 preview taking forever to generate rap songs

@unique dome actually it is working now but Suno sucks at making good hip hop and rhymes with rap

like it sucks major sounds like 12 year old

[crystal prairie]

workers are still spinning up under load, also please don't tag random team members

[graceful condor]

bahahahha stupid ai @crystal prairie lets get the rap sounding like a true fallen angel god

[crystal prairie]

in the works

[graceful condor]

i see as i need to promote my indie video game but the rap genre is just not there yet @crystal prairie @unique dome

[crystal prairie]

user was given 24 hours to scroll up for this post

[weary scroll]

use chatgpt to make the lyrics

[wintry saddle]

so the announcement said: Improved song structure and vocal flow - are their any new tags / markup / tricks or things it recognizes over 3.0?

or a case of consistency improvements for what we have today

[untold briar]

I'm not sure what Suno is specifically referring to but I would personally say there is an improvement in the effectiveness and expressive capability of tags in the lyrics field generally. For example you might be able to get away without a style prompt, which works less often (but not never) in 3.0: <#💬┃general-chat message>

The broad tag vocabulary seems mostly the same. 3.0 and 3.5 are much closer than 2.0 and 3.0 in that respect. With some exceptions: spacing and word seperators like dashes can be quite different. An example that came up is 3.5 prefers 'acid-house' over 'acid house'. Outlier tags - tags that aren't really musical terms or are even just random strings -- can be very different as well.

[teal raft]

i.e. prompting is different now

different enough

[wintry saddle]

Yeah seen that I often tried both - and no dash prior to see which I preferred

[untold briar]

I wouldn't really say that actually. It's not a big change relative to something V1 to V2, or V2 to V3, in prompts.

Specifically what I mean is many 3.0 prompts carry over 1-to-1 in 3.5 and produce the same style of music, without changes. This sort of 1-to-1 comparison was hard to do between earlier versions (even between V3A and V3!) because sometimes to get same result, you had to prompt differently. Not to say there aren't differences, but that a lot of prompts do just carry over, it's sort of backwards compatible in that sense.

[teal raft]

Ah, I see

[wintry saddle]

Ok, will have to experiment with bracket pairs like < >,{ } and ( ) and lyrics in " or ' for effects

I did notice in V3 rarely you could put squared instructions inside (backing style echo) and it would SOMETIMES pay heedence to it.

Will see if that success frequency has gone up.

[untold briar]

{} and <> are not supposed to work. (Though don't let that stop you, I am also a fan of trying things https://suno.com/playlist/6532a282-415b-4ab3-aa35-799e5bbc44d1). But just know it might waste credits.

[wintry saddle]

Sure I rarely use them, I expect them to do nothing (predictable)

But like to test occasionally

Thanks for the live feedback tho

I'm hoping one day we can do A counterpoint 2nd line in something like triangle brackets, and have 2 voices / singers overlapping, but it would probably be a UX and dev nightmare to handle it nicely.

But assuming not before V5 tbh 😉

[untold briar]

I know the devs are aware of the first bug. Actually if you have a second, can you check if the bug is actually as you describe, or if instead what is happening is the UI looks like you have deselected the song-to-extend-from (and the time) but if you hit the Create button anyway, the resulting clip will still be an extension (it will say Part 2 or whatever). When I tested the bug this is what was happening to me, and I'm not sure if this is the same for all users reporting the bug or if that is a different one.

The second item is probably best posted as a suggestion.

[untold briar]

When you hit Create again, you aren't actually creating an Extension, it is actually resetting the selected song to Extend? So you get a "Part 1" ?

[vestal rapids]

I am noticing 3.5 is giving me shorter generations. For the instrumental prompt upbeat catchy pop rock I am receiving generations of under 2 minutes, sometimes less than a minute. Whereas with v3 I would need to extend tracks.

[elder dune]

Hey guys, I joined here because over the last couple days I've been noticing that the AI will flat out ignore the first 3 or 4 lines of my lyrics, with the very first part 1 generation sometimes starting with the lyrics at the end of the song, even. And I can get it to give me an instrumental intro to save my life. 99.9% of the time it starts signing immediately and it's not Even the first line of the first verse.

Last month, it was working much better and I was able to put together a full length album [15 songs, 50+ minutes of decent music] .. but yesterday I spent half my monthly credits on one song and it never gave me anything close to what the prompt and lyrics were.

It almost feels like it's not delivering the whole track, but maybe it's chopping off the first several seconds or something. It's very weird and frustrating.

Am I the only one, or had anyone else experienced this?

[untold briar]

Have you tried in the last hour, there was a major update this evening, 3.5.

[elder dune]

Yeah I tried I'm 3.5 and, if anything, it's Even worse

I could be doing something wrong with the way I'm entering the lyrics, but it's the same basic format I was using before and getting good results with. For example, if I wanted an instrumental into, I'd do:

[Intro]
[Instrumental]

Or [instrumental solo] and I'd usually get it. Today after a dozen attempts with that, I spammed like 10 lines of [Instrumental solo] and [Instrumental] and it still started singing right away with the 3rd line from my first verse .. every stinking time

[worldly laurel]

I have similar experiences trying to reliable generate an instrumental intro for my songs. I'm aware of the method where you use the instrumental only option then extend from there but it's a waste of resources to generate a whole song (even more so with v3.5) only to use the first 10-20 seconds. I wish there was a more reliable way to get instrumental intros and not just have the song start on the first verse most of the time.

[untold briar]

Can you link or DM me a specific song prompt, the whole thing?

The other way to make intros is to start with 0 lyrics (or no sung lyrics, you can use [brackets]) and Extend From early in the song with your 'real' song with lyrics.

[elder dune]

I'm in my car right now [not driving] but let me see what I can dig up..

Right, I did that once before, and it worked ok, but then I still had the skipping lyrics issue. The funny thing was, I got clever and just typed in the first 3 lines [the ones it was consistently skipping] twice, and then, of course, it Sang them twice..

https://suno.com/song/42571dbc-3f9e-4e20-b827-422e71c641b4 here's tonight's example

Note the ridiculous intro that's completely ignored

[unique stone]

WAV downloads heheheheh

[elder dune]

How do you use the [brackets] to produce not-sung lyrics? Just enclose whatever in them, ie: [don't sing these lyrics you worthless pile of half baked silicone] ?

[untold briar]

Are you prompting literally 'nu-metal' with the dash?

[elder dune]

I'd settle for flac

[untold briar]

It's too early for me to have specific advice, but just vaguely I have had success using [genres] in the lyrics field to nudge the song when the style prompt doesn't seem to quite get there

Random guess, try 'nu-metal' as the only strict genre term in the prompt. 'metal' by itself there may be overpowering nu-metal. You can also repeat it 'nu-metal, nu-metal' but I would try removing metal first. Hybrid isn't one I've tried, could go either way.

There are some genres (country, k-pop, in past versions at least) that tended to blow away weaker genres, so if you wanted a hybrid you would to balance them out like 'country, polka, polka, polka'. Not a real example, maybe polka can fight country, but many genres can't.

[unique stone]

I would've done too, but now we have wav

[elder dune]

Works for me. That's what Audacity is for, ha ha

[untold briar]

WAV was always one of those features you're going to add eventually, it's probably trivial compared to other features, so why not get it out of the way even if the impact is low at the current quality. I was happy to see that in the announcement.

[elder dune]

Idk maybe I just have to give the AI a few days to settle in the the update and 'learn' how to parse lyrics again, lol

[unique stone]

Even if it doesn't add much quality it at least eliminates some of the quality loss.

[untold briar]

I don't think skipping here is a mystery. That's a lot of stacked [instrumental] tags that could increase hallucination chance by itself. Try using just literally one or maybe one + a bit of description. Separately the lyrics are atypical as well and I suspect might confuse most models. Is that latin written out in a sort of phonetic way to exaggerate the pausing with dases? Maybe start with regular latin, and if it's too fast, first try slowing it down with style and [bracket] prompts, and only if that fails, start spelling or adding punctuation like that. There are also options, CAPITAL LETTERS, `stars, etc, you might try. Basically that's a lot of dashes between 1 or 2 letters and I'm not too surprised it often gets confused.

[vestal rapids]

are the new WAV files compressed? they are very small in size

[unique stone]

Huh, not actually WAV

[untold briar]

WAV seems bugged, I'll make sure they know

Let me test if it's different with freshly generated clips. Seems to be the same problem.

[unique stone]

Good point, I was downloading a v3 gen from earlier

[real falcon]

up to 4min max clip length in one go, but how do I define the length of my clip? mostly curious about if I can define a length, like between 3-4min, of the clip it have to generate

[unique stone]

Seems it's generating the WAV but then sourcing the file for the download from the MP3. I've given some details in #🪲┃bugs-web

[elder dune]

Yeah that's Latin written phonetically. I've got a few more examples where I did as you're suggesting, tho, and it still did the same thing. And I've done the dashes things with transliterated Hebrew words and I got an absolute banger on the first try ..

Sidebar: I'm a bit of a language nerd, ha ha

[honest summit]

it's just putting the 192 kbps mp3 in a bigger container (wav)

lmao

I need the actual wav

there's no way I land the exact same dice to re-do the song in wav

[elder dune]

Anyway, it would help if there was an exhaustive wiki or something that told me what exactly can go in the lyrics and how it's interpreted, etc., with some examples. The programmer in me demands syntax, ha ha

This one got a lot closer, but still no intro - I even got the solos I asked for. https://suno.com/song/f52dfb55-23c8-4335-8b30-ca99ab0a3dba

Tried without the [intro] tag just now, and it still won't give me any kind of instrumental beginning. Argh

[real falcon]

I'm currently using, I just dunno if it takes all into account or not, but the 6 songs I've used this setup up for have an intro
[Intro]
[Guitar - Set the mood]
[Piano - Introduce rhythm]
[Build Up]

[elder dune]

I'll try that

Well that's progress .. now it's just giving random vocalization 'oooh, ahhh' type stuff

Wish I could tell it 'just. Don't. Sing for like 15 seconds'

[real falcon]

haha, yeah same here, sometimes I just want to smack it and other times it done something way better than what I wanted

[elder dune]

Yes

@untold briar you were spot on with getting rid of the dashes in the Latin tho. It's not skipping anything anymore

[real falcon]

try check if the vocals are singing out the content of [....]
I've had that happen sometimes, so had to either remove it or change it

[elder dune]

Ugh .., this last time, it sang "verse" before my lyrics 😦

[real falcon]

[....] should be like a command for it, if it don't understand it, don't do anything about it 🙂

[elder dune]

Yes

[unique stone]

Nah it isn't doing this. It's giving you the MP3 file but with the wrong name.

The actual WAV has high frequencies not present in the MP3. You can get the MP3 address from the debug console in your browser and just replace the end with .wav to get the real wav

[elder dune]

Got an into finally: https://suno.com/song/4977e983-6733-4ec9-a49d-1409b624bfc9 but the song is too 'happy' for the subject matter. Ha ha

[honest summit]

let me check

[untold briar]

It is obvious from the file size, the real wavs a lot bigger than the mp3 or mp4

[honest summit]

I don't see anything in the console

apart from ublock blocking tracking

lol

[untold briar]

It might be fixed now

Yeah all set. Refresh your browser first though:

[unique stone]

I had to go to the network tab, then press play, then you see the UUID of the song. But yeah, should be fixed now.

Well spotted @vestal rapids btw!

[elder dune]

Finally! May have a winner here. I used:

[Instrumental solo]
[Acoustic guitar]

[Verse 1]
Lacrimosa .. etc

And got a good intro

[honest summit]

well at least the shelf is at 20khz

so we good

[elder dune]

2 good intros from the same 'create' !!

[vestal rapids]

someone was bound to notice soon enough 🙂

[real falcon]

nice @elder dune you're on a roll 😄

[elder dune]

https://suno.com/song/51cfc4a5-a429-4fbd-b34c-8734c2d34967 not much of an intro on this one, but like it. We'll get there. Thanks for the tips!

[vestal rapids]

is there a way of encouraging the model (v3.5) to provide longer generations?

[chrome lily]

3.5 seems to be ignore genres

70s mellow Yacht rock for example is producing heavy metal more than half the time

[real falcon]

haven't had any luck yet with that... got songs anywhere from 1:45 to 3:43

[untold briar]

This is only day 1 so take this with a grain of salt, but I think in general 3.5 style prompts are better with dashes instead of spaces for style words like 'yacht-rock, acid-house' and so on. That has been my experience so far at least.

[unkempt spade]

that explains why reggaeton works fine 🫠

[unique stone]

.wav download appears to also work for old v3 generations too 😄

(in a meaningful way, i.e. high frequency data is present)

[cedar grove]

3.5 for me is awfull, I read that someone in 3.0 had problems with metallic sounds, All my prompts now had this "metal plate vibe", awfull.

3.0 had this voice weird chorus effect, but now the sounds are awfull with voices the instruments.. all of that

https://suno.com/song/47e85338-f312-49a9-b7fb-57467ed289d1 first 15 seconds are great, then WHAT DA HELL HAPPEN?

https://suno.com/song/4c3fee88-3192-4929-9094-47af19dccae2 same in this

[untold briar]

is that with Instrumental mode?

[cedar grove]

yes

but It happens to me with voices too

[untold briar]

I think that problem either the same as 3.0, or maybe slightly better in some more specific areas. But in the places where it's bad, it is still bad. There are sometimes prompt changes that help.

[cedar grove]

https://suno.com/song/899235a3-d837-45e2-a740-a846ff0a79ac

https://suno.com/song/3515b8b6-7b2e-41a7-b4e1-71c85bf9689d

both examples the voice starts better, thant it's awfull.

[elder dune]

Oh, noice!

[cedar grove]

In 3.0 in my experience it has this annoying like multivocals chorus , but this 3.5 is awfull, can't use it all

[unique stone]

Be sure to thumbs-down songs with horrible audio quality for v3.5 because this looks like it's going to be used to improve the model.

[cedar grove]

yuppp,

[elder dune]

Yeah, I mean, it's fledgling AI, so you gotta expect a few stinkers - but not 100 in a row, haha. I had wondered if it 'learned' from our thumbs-downs, and if it tracks those by account, etc, to try and predetermine what each individual tends to favor..

[teal raft]

It does not learn in real time

The thumbs down thing is for the devs to see when they train the model later