🪲┃bugs-web

[willow pollen]

Surprise Glee Choir hits again

soooo .... hepcat .. .still lulz'ing?

[gleaming halo]

They're a small company and low on resources. It helps to provide detailled feedback that they can reproduce

[willow pollen]

I did ... even burned around 1000c trying to find what triggered the choir/chorus and whether there was a way around, and I posted everything here. Which doesn't mean I think hepcat's "lulz..." was ... "appropriate" for a problem that has been documented by multiple users over the course of the day.

[craggy charm]

Just wasted 200+ credits trying to get suno to construct a simple 12 bar blues, tried various prompts, but EVERY time the vocals sing the lines TOO QUICKLY and fit sometimes two full verses into one 12 bar sequence, tried commas, splitting lines, but it seems incapable of doing anything that is usable. So frustrating.

[gleaming halo]

true that tone doesnt get anyone anywhere.

[craggy charm]

Voices are WAY too strident on V3, and there's always a choir effect, sounds like a crowd singing sometimes, unusable.

[astral adder]

good for you but do realize the breadth of the input space. This model is an attempt at a one-size fits all and just because it works for your music and standards doesn't mean it works for all. The easiest way to settle this would be to be granted access to both V3 Alpha and V3, then Suno sets and communicates an adequate retirement date for V3 Alpha. Until then community has enough time to compare outputs and give Suno enough examples.

Imo that is the least they can do since in another post we're lowkey asked to be beta testers on our own money. I'm not even considering re-running old songs on this new V3 model to give as examples unless they come out to show some transparency and make testing this easier for us

[gleaming halo]

i did just that for them in the above example, re run old songs same prompt.. V3 Alpha back is best and will give them time to look at the issues and prevent users becoming dissapointed

[mortal tide]

This is so wrong, synthwave comes out so bad now compared to in alpha

[loud lotus]

No, it's not only rock. I know that some people here believe that the issue is genre specific, I found it in any genre I tried.

[willow pollen]

which comes back to what Morgiana said : the width of possible genre and styles is so broad that you can't say 'it sounds good for me so you're all just imagining things/prompting wrong'. I tested a couple of genres and it seemed to me to be rock-related ... The point is : there was quite a regression for many people between v3-alpha and v3

[mortal tide]

The sound quality now is shit, its not clean as it was before.

[gleaming halo]

exactly, my list of over 20 sampels included 10 different genres

[willow pollen]

To be honest I stopped generating anything at the moment (just as my sub was renewed and upgraded) because it's just a waste of credits right now

[solar grove]

I do try to run the range of song types. I'm currently trying to roll something as a rock song but I don't have anything really ready at the moment and trying force incompatible lyrics is an exercise in frustration. Please bear with me, I'm actively trying to test this out right now.

[mortal tide]

But what does it matter even if you make it sound good, we are still a big group of people for whom it sounds so bad now

You ruined your own product, this makes no sense

[gleaming halo]

do you want me to DM my list of 20 comparing v3 Alpha to v3?

[mortal tide]

McDonalds wouldnt change their patties for puke and expect the customers to just take it lol

[solar grove]

Sure.

(But I won't listen to anything until I'm done with this testing.)

[willow pollen]

enjoy the Punk Rock Choir Collective

Gabba Gabba Hey
Beet on the Brat
Beet on the Brat
Beet on the Brat
Beet on the Brat With a Baseball Bat

Gabba Gabba Hey
Gabba Gabba Hey

[Guitar Solo]

Gabba Gabba Hey
Beet on the Brat
Beet on the Brat
Beet on the Brat
Beet on the Brat With a Baseball Bat

Gabba Gabba Hey
Gabba Gabba Hey

[Outro]
Cause Sheena is a punk rocker!
Enjoy the Punk Rock Choir Collective!

[craggy charm]

Spunked another 300 credits on trying to get a 12 bar blues, I officially give up today.

[willow pollen]

Next try ... .I still had 'beat' in there and discord freaked out

(Punk-)Rock Lyrics for Hepcat, so that they don't need to come up with their own

[mortal tide]

Make sure to email billing and ask for a refund ([e-mail])

Maybe they will listen

[solar grove]

https://app.suno.ai/song/507cc576-9316-46fa-a7fd-b40133cc868e

Following the Sun Around lyrics by Joseph McCarthy, (c) 1926, Public Domain.

I think this is decent. A bit of a surprise to me too, because I always saw this as an adult contemporary song. Seems to work as a rock song too...

[mortal tide]

@solar grove

[solar grove]

For comparison purposes, here's the "original" version: https://app.suno.ai/song/7f303f12-9dcf-4780-b46b-cc44c4ee01a1

Actually that's not correct. This is the second version. The first was done in beta testing v2 and it suffered a bug so I had to remake it.

I still miss the chorus on that first one.

[grave folio]

can someone help with it?

[loud lotus]

What's your point? The lead guitar is thicker than a quadrupled Boston track and the vocals deteriorate straight into a choir.

[willow pollen]

1:43 single voice, 2:23 full blown choir

you can hear the additional voices creep in at 1:46

[solar grove]

Hmmm...

I mean I heard it on the last chorus and didn't think too much of it because, last chorus you get to break any rules to finish it out with. But yeah I start to see your point.

I may have been making complimentary music, so I didn't notice it as much before.

I'm going to listen to some examples and test this out some more with some of the rock\metal songs I have to see the comparisons.

[willow pollen]

to be frank, on my first gens I didn't think much of it. The occasional chorus was normal with V3-alpha too. But it's really every single generation with V3. It's basically vocal Neon

[mortal tide]

I dont even do vocals and I just noticed that the sound quality is way lesser now

[vestal thistle]

Vocals and Instrumental seem to have gotten a hit compared to V3 alpha.

[mortal tide]

I only do synthwave and retrowave and I notice a huge drop off, cant speak for any other genres

[vestal thistle]

I do rock/metal and I noticed the drop off immediately when the V3 model launched.

[mortal tide]

I can imagine, my credits had ran out so I actually upgraded today to the biggest package. Guess if I am a happy customer or not because last time I had alpha, now I have whatever this is...

[vestal thistle]

Upgraded to the Premium Plan too because I was really happy with the alpha and yeah, now this.

[mortal tide]

I bet you also feel completely cheated

[vestal thistle]

I do, yeah

[mortal tide]

Same

[vestal thistle]

Would love some transparency

[willow pollen]

things like this happen ... no need to overdramatize. I'm actually quite sure that they'll find and roll out a fix now that the problem is known.

and communicated

[vestal thistle]

I hope they do

[mortal tide]

While we wait our credits are unused and they dont roll over to next month

So what am I paying for

[solar grove]

Have you emailed [e-mail] regarding this?

[mortal tide]

Yes

[solar grove]

Then please be patient and wait for a response.

[mortal tide]

Do you happen to know how long that might be?

Like on average

[solar grove]

I do not.

[mortal tide]

Aight

[solar grove]

I am only a Discord moderator. The team handles that side of things.

[mortal tide]

Yeah fair, ty tho

[merry sorrel]

I've been writing lyrics for years, no issues before, now all the rap is offbeat, no matter what I do and how I try to change the lyrics lol

[craggy charm]

Agree, I have tried pop, disco, blues, punk, every one the words go out of time with the beat most of the time

[halcyon violet]

I don't know if this is a problem that everyone is having but everytime i generate a song, it seems to take more than 10 credits per generation. I have 50 and everytime i generate a song, i can leave with 30, 35 or even less. What is this issue

I generated 2 patches and i started with 50 credits so i should have 30. But now i have 15.

[solar grove]

Were there any refunds? Does that number stay the same after a browser refresh? Save your work in a text file before you refresh or reopen your browser!

[cold iron]

guys why v3 i cant continue from this song its not working good any help????

[wary pumice]

At the lowest volume, when a sign appears that the music is turned off, you can still hear it.

[half cliff]

V3 always has every song with background singers, the voices mix too much with music now making it difficult to distinguish what's being said, and the quality of sound in general is like it has a lower bitrate or something.

Yeah. I'm getting choirs on fucking jazz romance songs.

[opal idol]

v3 sucks compared to v3 alpha. can't continue songs, the vocals are buried. will cancel if not fixed

why does everything have a choir now. If i want a choir i can ad that with antares. but I can't take the choir away

[spiral forge]

lol, the prompt "The most stereotypical song about New York state" always fails.
I was able to generate songs for the 6 New England states without issue:
https://app.suno.ai/playlist/c4b759ac-d1cc-4793-bf30-c27604df9088/

[cosmic stag]

can we get a word that flagg the dumb content moderation filters because I'm doing black metal, it's fn stupid that i can't write lyrics about a war

wtf is wrong with yall content moderation on music?

have you heard what is played on mainstream radio? you babies cannot be serious

if i see this stupid pop up one more time I'm cancelling all of our paid accounts this garbage just pisses me off

[solar grove]

Maybe this says something about New York? LOL

[cosmic stag]

changed lyrics to the point where it doesn't make sense, the moderation isn't making sense for serious people.

fix this suno

[low cypress]

have you tried pasting your lyrics in chatgpt, telling it is a lyrics that is been flagged for moderation? It will tell you the problem usualy.

[cosmic stag]

have you sterilized your writing with repeated words in the tapestry woven of the woven tapestry we find other worldly insanity of unspoken words for the absurdity of these content restrictions don't make sense

Like it

chat gpt sucks at writing I don't care what anyone says

sadly I'm not sure I'm going to support this level of censorship

[low cypress]

I understand your frustration, but continued ranting in this channel isn't productive. If you have specific issues, please email [e-mail]. We're volunteers and here to help, but let's keep this channel focused on bug reporting. Thanks for your cooperation.

[cosmic stag]

the bowels of the abyss, where evil festers and thrives,
We summon our legions, and sharpen our knives.
A vile crusade, a malicious endeavor,
To storm the gates of heaven, and forever sever.

Arise, you demons of the night,
With hatred burning bright.
On the wings of despair, we shall ascend,
And bring about the heavens' brutal end.

Through the tempest of malice, we shall soar,
With the roar of the damned, and the screams of the poor.
With the strength of the dark, and the coldness of steel,
We shall invade the heavens, and make the angels feel.

Yeah with a bit of prodding for chat gpt it's vastly different from my original lyrics and yet still continues to flag it so explain

need a list of things that flag the content moderations.

it's a bug obviously, no swaring no sexual content and it's still flagged.

[low cypress]

This is what chatgpt tell me about this lyrics:

Now again.. just trying to help you @cosmic stag
I understand your frustration, but I cannot change it.

[cosmic stag]

My staff is laughing at that... like isn't that the concepts of music now? what is this?

[low cypress]

I don't understand why you want this discussion with me? I have no answers in that regard. Please email [e-mail]. Again, we're volunteers and here to help, but let's keep this channel focused on bug reporting. Thanks for your cooperation.

[cosmic stag]

what should i tell them? chatgpt says it's lyrics is bad? lol

[low cypress]

Hey, I get where you're coming from. It can be frustrating when your creative expression feels limited by rules and guidelines. Trust me, I've been there too. But at the end of the day, Suno's platform operates within certain boundaries, just like any other service. While it might not be perfect, they do have to enforce rules to maintain a certain standard. It's about finding a balance between creative freedom and responsible content.

[cosmic stag]

again have you heard whats on the radio lately? lol what standards

[low cypress]

Again.. you still want to pick this fight with me? Me? Miss Nobody? I have no say in Suno.. ZERO.. I only help.. and it seems you feel like I am your lighting rod. Please move on.

[cosmic stag]

(I watch it all burn) - flagged?
im not trying to pick a fight im trying to find out lol if there might be a way to bypass this. It's gotta be a bug at this point.

i put one line

[low cypress]

Maybe some other volunteer has more free time. But I have some other things that need my attention. Good luck!

[spiral forge]

Is Suno's billing system down? I tried to add credits and it is just spinning forever.

[neat bloom]

I just destroyed over 300 tokens trying to generate a chorus after the second verse like the one after the first............ ZERO percent similarity.
Go "love" yourself with this approach, focus on your neural network and get it to a normal working state before we get to paying a hundreds bucks for one crappy audio track, this ain't "loving" funny!
V2 made identical choruses, so what the hell is wrong with v3?
From when did old versions become better than the new ones?
I don't have the words to describe how stupid this situation is.
I'm not gonna pay for this shit anymore because creativity shouldn't bring anger, grief, disappointment, and so on.

[neat sail]

The bug is actually that the volume looks like it is at the lowest but it isn't, so if you drag it a little more to the left, it will shut off.

[sinful silo]

Wondering if this is a bug. A few times (rarely, but still) I've generated a song and when I hit play it went well over 3 minutes. Then when I come back later, the song's cut off around 2 mins. Some screenshots attached

[neat bloom]

Ahah, read the comments here and came to the conclusion that people need to stop paying for this crap and only then will start real work on SUNO AI, improvements, bug fixes, users will finally be heard, etc.

[humble cobalt]

I have had this a couple of times. What happened in my case is that it repeated the first 50 seconds or so again, so there wasn't really anything "lost".

[sinful silo]

Thanks, that's what seems to be happening.

[neat sail]

Can you describe how to reproduce scroll bar stops at 1:00? On desktop or mobile, which page, etc?

[neat bloom]

Why is there two minutes for the initial generation but only one minute for the continuation?
I have a slow tempo of the song, I can only fit a verse in a minute, so I can't make the same chorus as the first one.
I just don't understand it, if you can make the first generation for two minutes, why the hell can't you make a two minute continuation????

[formal tiger]

What did you all do? After the update it won't pronounce some Spanish words correctly It tries to do a mix of Spanish and English.

[spiral forge]

Did you try having everything in the lyrics and prompts in Spanish?

[lucid spoke]

It's just loading. I'm in the middle of a rap battle 😦

[neat sail]

How long have you been trying? Probably just a temporary glitch?

[spiral forge]

I reported it earlier, so billing hasn't been working for ~1h20m at this point.

[formal tiger]

Something is going on, is taking forever to produce and I want it to start with the instrument but it goes straight to the voice, ugh

[wary flare]

So, Suno just deleted a bunch of the stuff I was working on, and I cant seem to find any of it even in trash, is there a way to see where all of the songs went? One was done, and it deleted that off my list too

[neat bloom]

I hope that you guys are actively fixing all the stupid bugs at the moment and will make 2 minutes to continue generations instead of 1 because at the moment there are some works in progress

[humble cobalt]

Try using "get whole song" after the second verse, then continue from the "whole song" it generates. This might help the AI to "see" the previous chorus (otherwise I believe it just looks at the part you continued from, and wouldn't see the first chorus.)

[formal tiger]

It did that to me too but I was just staring, I would no work on songs tonight, they are doing updates. It will just waste your tokens.

[wary flare]

I may have found the solution, apparently in library there are pages as well, dont know why it compartmentalized the stuff I was working on or completed though, I literally just started using it

[wary flare]

^

[cloud python]

Dang and here I was worried about the lyrics in my song taking pot shots at the government, corporations, etc LOL. My song is a lullaby by comparison

[neat bloom]

That's exactly what I do

[lucid spoke]

Tried now again, still not working.

[neat sail]

You likely have to email [e-mail] but you can click on "subscription" then "edit billing details" and see if there failed transactions on that screen

[lucid spoke]

Okay, I will mail them. It still says "Open" on those 4k credits

[neat sail]

Oh yeah email for sure to make you sure you those all don't suddenly go through and you just spent $200

[lucid spoke]

Yeah, that's what I'm worried about 😅 But even if they go thru or not I'm sure that I would have spent them anyway.

[neat sail]

Well the credits you buy this way don't expire, but even if they went through, I'm sure Suno support would set things right

[stone isle]

On a few runs i've noticed that V3 lowers the volume of the vocals in some places where it becomes too hard to hear while making it loud in other places. This makes it hard to split into stems for the low volume parts.

[royal mason]

This should be fixed now

[idle pine]

Good Luck Suno Team, V3 isn't great, Please fix or revert to V3 alpha. Thanks.

[willow pollen]

have you voted on the mini poll?

[cerulean sentinel]

https://app.suno.ai/song/2a6b9e90-e54d-44d3-9b46-065b9ffc0a49

this song seems to be basically in the right ear

although I can't rule out hearing damage haha

but as it progresses it seems to become more pronounced

btw, I've done testing and I think the "it isn't great" is subjective, and the effect is similar to people's perceptions of gpt-4 degrading

it's almost the same to me. The voices degrading thing does seem to be an issue, but I can't honestly tell objectively how much so, as I don't have access to the prior models. From what I can tell, it's mostly related to adhering go previously good style tags, which may produce bad results when the model changes. Tweaking these style tags to find new useful ones seems to produce good results, and I absolutely have produced several songs better than my alpha results

with clean voices

I think there are plenty with dirty voices on alpha, we're just collectively biased towards believing this to be less since all of our public lists contain a group we specifically selected for, namely songs with clean voices

[willow pollen]

have you produced rock songs with good non-chorus'y voices with V3? @cerulean sentinel

because it's not a question of 'difficult to get'. It seems to be literally impossible. The songs start correctly, and after the first verse (sometimes even before that) a chorus starts creeping in. By the end of the 2mn the voice is just drowned out

notice how, even though additional voices appears in ... well .. the chorus, they always recede and, for example from 0:40 the singer's voice is well separated (albeit a bit autotuny) https://app.suno.ai/song/8f54b6fd-4a8e-4080-837e-807209db92bf

I dare you to create something doing that in V3

https://app.suno.ai/song/16252a79-3caf-46a9-9e1a-e80829058e13 slightly different prompt on that one ... also impossible in V3

those were generated with V3-alpha

[tacit geyser]

I agree, I think I spent almost 2000 credits, all songs have a choir sound at least in some parts (most are choirs from the beginning). I assume the devs are aware of this and hopefully until my next billing cycle (I'm being selfish 😂) they will release a fix.

[cerulean sentinel]

I think so ya

[cerulean sentinel]

lemme take a looksie

here's one
https://app.suno.ai/song/3a53cca0-a516-42b7-802d-168ea2bbbcd8

https://app.suno.ai/song/0c862712-1e2f-41e9-853f-dc13573adfae

and another

I hear a bit of reverb in 2, but it's thr same in v3 alpha, all voices can have that autotune

without careful prompting

most recently though I just used myself in the style, idk if they're recursively feeding our outputs to it, but it worked

https://app.suno.ai/song/2a6b9e90-e54d-44d3-9b46-065b9ffc0a49

so that's all rock, although specifically indie and slowcore respectively

[willow pollen]

that's pop

[cerulean sentinel]

everything is pop technically if you squint hard enough

Are you looking for classic rock?

I don't do that, it's too bland

[willow pollen]

any kind of rock?

[cerulean sentinel]

the Mother one is slowcore rock

[willow pollen]

1:25 .. .full chorus

i nMother

[cerulean sentinel]

you should see genre sites, everything gets labeled pop

it's just a way to throw shade

[willow pollen]

to be honest I find those clips of yours VERY bland ... it's all a matter of taste

[cerulean sentinel]

yea, I hear that, I thought it sounded good though for that portion, but ya

[willow pollen]

https://app.suno.ai/song/c738f8ed-8f5d-4afb-89a0-710c561b02db <- pop ?

[cerulean sentinel]

I feel ya

[willow pollen]

(v3a btw)

[cerulean sentinel]

Ah I see, ya, I don't like metal much honestly, I'm an outlier by suno standards lol

it's so boring to me

[willow pollen]

no .. .you're not.

[cerulean sentinel]

OK, well let's stop this is the wrong place

to discuss taste

[willow pollen]

which is why the poll looks the way it is

the problem with the choir is most evident in rock and metal. A big chorus is wanted in pop to alleviate the overall bad singing abilities of the singers 😉

[cerulean sentinel]

ya, I see that, I was just reviewing

ans I think therss a way around it

specifically, I would test out continuation

below 1 minute

on your gen

and see if it is replicated

I suspect it isnt

bc I was looking at some of my songs which use over 2 and they also include that chorus, although in such positions it happens to have been selected for

but it goes away on a later continuation

[neat sail]

I brought up the idea of adding a #model-feedback in the Feedback category separate from #bugs, so there's a specific place for model comments which are currently being put in bugs, but adding channels is up the Suno folks.

[willow pollen]

yeah .. the chorus generally starts around 10-30 seconds in ...

[cerulean sentinel]

which makes me think it is worth testing if it's something in the 2 minute implementation specifically

[cerulean sentinel]

oh wild, I hear some degradation now and then but nothing like that. Try adding some rags like "intimate vocals" "underproduced" etc

[neat sail]

A rather crude way of dampening down chorus/choir/reverb is to intermix some V2 clips

[cerulean sentinel]

the voice is strongly affected by that

[cerulean sentinel]

yes, I use v2 in many gens

it's a great bridge in certain situations

and for some genres is produces better openings

[willow pollen]

yeah ... there are ways to circumvent the issue .. .but they are extremely costly and definitively a regression compared to V3A

[cerulean sentinel]

thus was in v3a

at least, on mine, some of the songs I just examined are before the switch

[willow pollen]

was Mother?

[cerulean sentinel]

keep in mind, your collection is going to be selected for already

[cerulean sentinel]

no

that's today

[willow pollen]

cause that's the only one I heard the chorus in

[cerulean sentinel]

also leave my mom out of this

[willow pollen]

¯_(ツ)_/¯

[cerulean sentinel]

😜

[willow pollen]

https://tenor.com/view/spy-tf2spy-no-mother-team-fortress2-gif-22434453

[cerulean sentinel]

Idk, best advice I have to to try acoustic and less produced tags, the issue is thsi may pollute your metal. But what you can also do is jist start you riff, and then ignore metal tags altogether. Just progress to limited tags that focus on vocal clarity. The metal cadence will likely be preserved anyway just bc the momentum of the song

[willow pollen]

it's spoiling jazz too ... https://app.suno.ai/song/582f95e8-f408-42e8-953d-b573dcfdac87

[zenith karma]

Whenever I set a time in the "Continue From" and then switch between v2/v3 then the time reset to 0:00 everytime. I was expecting it to not reset.

[verbal glade]

Theres so much feedback on v3 vs v3 alpha. Can they just make both available in the dropdown and let users pick?

[strange fjord]

I do not think this will happen. Even my V3Alpha processions are now tagged as V3

[willow pollen]

the good thing is that I know except for 5 songs all the rest are V3a, cause the V3 clips all landed in the trashcan ... along with ~1500c 😭

[strange fjord]

Same here

[white vigil]

the pre-existing default song ate 10 credits even while tossing it into trashcan 😁

[proven socket]

@neat sail
~~Has this been marked as a potential bug/quirk:

Style Tags can leak into the lyrics?

• Vocalist shouts "2 Step" at song start which is in the styles box (#🪲┃bugs-web message)
  Example: https://app.suno.ai/song/092365e2-1866-4ea0-9d52-15414d8898cb/
• Vocalist shouts "Banana" at song start which is in the styles box (#🪲┃bugs-web message)
  Example: https://app.suno.ai/song/10c99889-4d1f-4664-ae70-ecc9338d921b/
• In thse examples, you can basically change the last style tag to whatever you want and in many generations the singer will sing it.

A slightly similar quirk is where it will read out metatags: like you might have [Big Finish] written or [Action Section] and it reads that out. Though maybe when this happened to me I was missing a bracket - I don't believe so though. I can go through and try to find examples of that, because I'm sure I have some clips where that happens.

I guess these things are rare, but yeah.~~

[proven socket]

Or if this is just part of how the AI naturally works - that makes sense too

I'd rather have funny quirks than too many edge case work arounds that mess up how the AI works 😆

[verbal glade]

The underlying model though. Call it whatever they want, just make it available.

[neat sail]

Depends on the cause. If the cause for example was a website UI bug that is messing up the prompt that is a bug for this list. I think it's likely an AI model quirk (it for sure can happen as a quirk) but if it happens very reliably that would be a hint maybe there is a more mundane cause. TLDR: ???

[proven socket]

i don't think it could be a UI thing (you can replicate this over and over), the AI model quirk definitely makes sense as you said. it must be the part where it combines the style data and lyrics data

[neat sail]

My best wild guess is the cause is actually pretty simple.

Look at the style prompt in your examples. What word is in front of the word being sung from the style prompt?

That could be a confusing situation for the AI right?

"Oh chorus so I now I start singing!"

[proven socket]

yeah fair, but it still shouldnt do it. I get what you're saying (I'm a programmer myself), but there should be some kind of containment so it doesnt happen. However if containment means messing the model up, then its probably better for it to be a quirk lol

i actually got the example from @wary flare above, I was just reposting it. I didn't actually notice the chorus thing till you mentioned it 😆

though now that I appreciate it is probably coming from "chorus" it doesnt seem as bad lol

[neat sail]

Yeah it's still a bug of course. But it might be a tradeoff that you choose to accept. the model is freeform and loose, so it gets mixed up, but the upside it maybe all kinds of weird prompts and suggestions and [tags] have cool effects. (This is all my uninformed personal speculation, just fun to theory craft a bit.)

[proven socket]

yeah i like this perspective also. Makes 100% sense 🍻

now that I realise it says "chorus" just before, it does change my mind about it lol
(dont know why i didnt notice that)

[neat sail]

Yeah my wild theory was "chorus" in the style prompt, with a word following it, sometimes confuses things (after a chorus are words I should sing!) and so it sings the word.

[rain lodge]

HELP! I invested two hours crafting a captivating song, yet upon its public release (whol song), it fails to appear in the new-area.

[proven socket]

yeah after testing around I think you are 100% right, once you remove "chorus" (and "verse" too) from the style tags it doesnt occur

[rain lodge]

I already had some songs inside the trending-area with more than 10.000 views, but it's not funny to work on a song and then it doen't even show up in the "suno-charts"

would be important to remove this bug. Cause once it is public you can't release a song again!

[proven socket]

yeah this has annoyed me too lol. According to what has been said previously, it's the AI filter that they use before the songs go to the "New" list, it can be a bit sensitive, it might flag some songs incorrectly

[rain lodge]

there's not one bad word in it 🤨

[low cypress]

Hey Gin, I understand your concern but it's important to know that the visibility of songs in areas like the new-area or trending-area on Suno is influenced by several factors. These include the timing of the release, the number of songs being processed, and the algorithm that curates content for these sections. While it's disappointing when a song you've invested time and effort into doesn't get immediate visibility, it doesn't necessarily mean there's a bug. Sometimes you have to "push" it a little by posting it in #🐣┃suno-showcase and sending it to friends or family.

[proven socket]

i have had songs without any bad words flagged, and even instrumentals that only had a few words for the sake of the song art 😅

[low cypress]

The moderation looks at the meaning of the song, not at bad words in general.

[rain lodge]

Then there should be a status-update like "flagged" so you can fix that and release it again. Two lost hours for a song that would be top 10 again 🤨

[proven socket]

yea though ive never written a song that would require that moderation 😆 . I understand this is something we have to put up with for the sake of practicality. I'm only saying the AI's sense of moderation isn't always correct

[proven socket]

yeah i've thought the same, like i have one in Trending atm too, and there are a few times where I think i've had better songs that didnt even go to New

[low cypress]

I know.. I have the same problems in general with chatGPT.. but I also do not know a better or faster or cheaper solution. If you know, I am all ears 😉

[proven socket]

fair point, that makes sense. I don't have a better suggestion lol

[low cypress]

Just so you know.. I made 3 songs yesterday and they were "clean" too, but still not added to trending. Dissapointing? Yes.. but not a bug, and there was nothing wrong with my songs. All my babies are beautiful just as yours! 😉

[rain lodge]

I understand your point, but it's essential for platforms to ensure transparency in how they showcase content. It shouldn't be akin to a lottery, where the outcome is left entirely to chance. However, considering that the ranking system is only in its initial stages, I'm confident that Suno will soon improve it.

[low cypress]

They explained as I explained you. It's an algorithm. And if you know how to influence it, there is no challenge anymore. I think at least. I mean, I don't know the exact algorithm of Google either how to make my page appear at number 1. I see it like that. And besides that. There are only 50 spots for both pages, while there are already more than 280.000 people alone in this server of Suno. So not everyone can get on the trending even if they would want that... But against.. just my personal speculation and thoughts 😄

[spiral skiff]

any reason why Suno belatedly said "fuck this song in particular"? did the violin go so hard it destroyed the image? 😂

[rain lodge]

Alright, cool, here are some suggestions that this algorithm could consider in the future:

• Whether someone has already had one or more songs in the Top 10 / Top 20
• The amount of effort put into a song (number of variations used)
• Prioritize Premier subscriptions

[low cypress]

Ah yes.. but it's in the wrong spot then. Next time please place in #💡┃suggestions-web 🙂

[willow pollen]

3Daizy, if you happen to talk to one of the devs, could you just ask whether they are aware of the Choir/vocal v3a Vs v3 thing and whether they are looking for a solution or feel like it's user-generated drama for nothing? I mean the question completely neutrally, it's just to adjust my expectations.

[neat sail]

Yeah if you browse around you can find suno devs like tony popping up and chatting about various things including model feedback

[low cypress]

I can only point you to this: #💡┃suggestions-web message

[willow pollen]

and this : #🪲┃bugs-web message

well ... that's a pretty clear statement I guess.

'user created drama' it is

[low cypress]

Yeah, it's just how things roll. Every model's got its quirks and room to evolve, you know?

[willow pollen]

hmm ... no comment.

I'll resubscribe if they have a model that can actually generate anything that doesn't turn into a musical

twas fun while it lasted ¯_(ツ)_/¯

[low cypress]

Yeah it really depends on what you want to create. I made 3 songs yesterday of which the quality was pretty good. 🤷‍♀️

[willow pollen]

I made one song that was good ... but that's because the genre 'supports' a chorus effect. All other generations were complete crap, even stuff like 'gypsy jazz, string guitar and violin, male voice' turned into a (instrumentally actually great) Broadway event with a complete symphonic choir

all female at that 😉

[tacit geyser]

Everything resolves itself at some point. Just give it time and of course feedback is important so the devs know what to fix.

[willow pollen]

the thing is : the devs' answer was ' well ... you don't like it, too bad'

which is okay. It's their product they can do what they want.

[neat sail]

hope you stick around even if on free, I've been checking out your feedback, appreciate the detailed samples, good complaints are gold

[willow pollen]

but that means for me that I cancel my subscription (what happens with top ups btw?) and wait till a better model is released

[willow pollen]

yeah ... I think the next model might be better again. It's just a waste of money for me to stay subscribed at that point

[low cypress]

Top ups stay

[humble cobalt]

Top-ups should stay. You're not supposed to be able to use them on the free plan but I was able to when my sub expired and i took a break for a few days.

[dawn radish]

keep getting error occurred. have to hit create quite a few times for it to work.

[shell yew]

error after error now 😦

[dawn radish]

so its not just me then.

[vernal harbor]

is the suno website having issues? cannot seem to continue from a song

[dawn radish]

just seen this message pop up now 'We are doing backend upgrades. You may experience some downtime'

[shell yew]

nope nothing working 😦

[mortal tide]

Maybe it is overcrowded atm. give it some time and it will probably work better

[shell yew]

dang always when I got a spare 5 mins to do something as well LOL

[mortal tide]

Now we know who to blame at least

[low cypress]

Just give it a few minutes.. they are working on it

[shell yew]

Yep it was ME 🙂

[low cypress]

https://tenor.com/view/bunny-trouble-spanky-gif-9970011438135193066

[spiral skiff]

so this is what the devs are doing when they say they are "working on it" .... 😏

[low cypress]

lol

Nah.. they are kicking the server instead

[vernal harbor]

can anyone recommend a good way to end a song? 😄 kinda get just cutoff in the end

[dawn radish]

[instrumental solo outro] is what i normally put

[low cypress]

remove your lyrics, remove your style of music, and then put in end in style of music is mostly working better than metatags

[spiral forge]

Adding "finale" to whatever style I currently have works for me quite often with the latest v3.

[solar grove]

3Daizy is a guide not a developer though. I am only a moderator here myself.

[dawn radish]

how do i empty my trash without doing it 1 by 1?

[spiral skiff]

got a reply in #💬┃general-chat from the dev mc about it earlier

[solar grove]

Oh. I missed that. Sorry.

https://tenor.com/view/alright-carry-on-continue-resume-keep-going-gif-16771637

[bold marten]

trash can can not always trash the music, menu oversize fonts

screensize :1920x1080

[shell yew]

what ?

[low cypress]

What is unclear?

[shell yew]

and then put in end in style of music

[low cypress]

Yes.. put “end” in style of music field

[shell yew]

ahh OK

[vernal harbor]

hmm, when downloaded the full song, their is a silent gap between part 2 and part 3... how to remove / avoid having that?

kinda part 2 faded out.. then like 1 sec after part 3 faded in
😄

[bold marten]

@vernal harbor you search the tune for the beginning of the silent bit and added it to the left side timer, to continue from there example [0.30] then left side timer [you input [0:29] and click continue

you can't avoid either, it just ai have a hairy one.

[vernal harbor]

ok thanks for that

[broken cloak]

https://app.suno.ai/song/ecd3dacd-c5ba-444b-ab1c-2f772aaeeeac I tried this prompt twice and got the same message :/

[spiral skiff]

oh boy lol

[untold eagle]

too many people using it

[fierce path]

why is anyone else having to wait extremely long to load the create tabs

its been aorund 3 minutes and its still loading, and my wifi connection is working fine too

[tame latch]

website is down

[honest mason]

how did u crash right after i did a yearly subscription

this sucks

[mild mango]

damn

[lean estuary]

backend is down, vercel is good

[wary flare]

too many people using it

[lean estuary]

https://cdn1.suno.ai/85eee00d-0b45-447d-99e1-c2718fdd179d.mp3

but cdn is fine!

[mild mango]

牛逼

[honest mason]

他妈的刚付完钱就垮了，

6

[mild mango]

我刚想付钱

[wary flare]

今天国内来了一堆人

估计扛不住了

[boreal isle]

i still loading here

its*

[pine lichen]

Hey I can’t login

What’s going on?

[manic stream]

too many people using it at once probably

[wary flare]

website is down

[honest mason]

get a larger server or stop free service what so ever

[pine lichen]

But god damn I do love them

Been obsessed last two days

[wary flare]

this isn't a bug report and isn't unessecary

[crimson osprey]

please be patient and don't assume things. the devs are obviously working on bringing it back asap.

[pine lichen]

Any update I was just doing my yearly subscription

[honest mason]

it‘s nessecary when i paid my money

[wary flare]

#🪲┃bugs-web message

[wary flare]

okay buddy. Many are paying more than you and are being patient

[honest mason]

are u one of them？

[wary flare]

try to enjoy anything else for a few mins and let them fix and we good to go 🙂

oh i'm for sure one of them

[shell panther]

I am one of them @honest mason just chill and wait

[boreal isle]

can you stop complaining? every site can be down.... even my site, even the largest

[low cypress]

Suno staff is aware... they are working on it

[honest mason]

well i mean ur not the one to say whether its nessecary or not

[craggy charm]

Hope this downtime is an update

[boreal isle]

uhmmm sorry? but downtime can always happen... its not really to complain about it, even your paying €100 a month

and no im not paying for suno.ai

[tame latch]

Easy folks, we are working on it

[boreal isle]

we know, thx

[tame latch]

Standup and stretch, walk around a bit

[wary flare]

it may be worth changing this to thread styles posts to handle conversations more easily and quickly delete non bug posts 🙂

[boreal isle]

ok

[honest mason]

whether im paying a cent or a grand i have rights to report when there is problem，this one just to you，not the website

[wary flare]

If you guys are ever looking for some discord help, feel free to reach out. be happy to help

[tame latch]

easy

don't blame other users

please

[honest mason]

im sry？who blames first

[tame latch]

just drop the thread

plz

[boreal isle]

yes, you have that right. but over and over agian.... (my last message about this)

[wary flare]

save bugs for here and move to general to share things you have downloaded from suno! 🙂

[low cypress]

Look a DEV is talking to you while they are having problems.. just drop this now before I have to put people in time out..
Thank you for your understanding!

[lean estuary]

I sent a link just to prove cdn is working, not just sharing things

[sonic viper]

Hello everyone,I have a question, i made a song with suno but there are lots of artifacts in it, is there a way to remake the same song without artifacts?

[solar grove]

Can you turn on slow mode in server settings?

[olive glade]

Probably need to fix your timeout system, some mod timed me out for asking what domains were ok to link in here. And as a result I couldn't access the bug channels or ask questions about billing and support. Is there an e-mail we should use instead ?

[solar grove]

I meant for Discord.

[gaunt quarry]

In every metal song somewhat like background noise sounds along all composition, hope it will be fixed soon

[lean estuary]

human voice might also have this problem, maybe denoise should be enhanced

[gusty chasm]

Hey @wary flare check your friend requests DM's!

[prisma orbit]

@tame latch Do you know when this failure will be fixed?

[low cypress]

@prisma orbit Don't do that.. If they have news they come by themselves. I warn you not to unnecessary ping them

[sonic viper]

@low cypress can i ask you a technical question?

[low cypress]

Sure.. if I can asnwer it I will 🙂

[sonic viper]

i made a song with suno but there are lots of artifacts in the vocal part of it, is there a way i can remake the same song without artifacts?

for example if i continue from the same song is there a way to let the ai "remaster" it?

[undone bloom]

I used the (/chirp) but dont open the window

[low cypress]

Remastering a song to reduce artifacts isn’t something Suno can do. Each generation is pretty unique. Continuing from the same song might not necessarily 'remastering' it and will also completely change it. You can try a few strategies to potentially improve the quality for future songs you make:

1. Continuation with Adjustments: If you choose to continue a song, tweak the style tags or settings. This won't guarantee a remaster, but it may produce a version with fewer artifacts.
2. Recreation with Variations: Start a new generation using similar prompts and settings as your original song, but with slight variations. This could lead to a cleaner version.
3. Segmental Regeneration: Identify the specific parts where artifacts are most prominent (when you started a new song) and use the 'continue from' feature and specify the timestamps where the clean-up is needed.

Remember, due to the inherent variability in AI music generation, these methods may not perfectly replicate your original song, but they can lead to improved results. It often requires a bit of trial and error to get the desired outcome.

[sonic viper]

@low cypress thank you so much for the explanation, very helpful 🙂

[low cypress]

You are welcome

[thin axle]

@sonic viper Might I suggest you use tryReplay.io ... you can swap out the vocal models.

You would just need to find a good model on weights.gg

[sonic viper]

@thin axle thank you for the suggestion 👍

[half cliff]

Is the whole choir thing being addressed? Or is it not considered a bug? Because I can't make a single jazz piano+sax romance song that doesn't dissolve into annoying choirs.

I'm sorry is this not the bugs section?

[bold marten]

i put that in the wrong channel, not for you

[clever pumice]

How do I report a problem related to foreign genres that are different but have the same name?

[vapid field]

hi, anyone from the team here? the server downage killed my sub i just bought i guess...

[solar grove]

Definitely email [e-mail] but I will also mention this in the moderator's section to the Team members.

[vapid field]

ty!

[sinful sentinel]

there was a song here that now returns an internal server error, is there anything that can be done about it? https://app.suno.ai/song/5e2c00c4-d427-48b1-917a-a86448e29325/

[low cypress]

that probably means that the song is deleted. Was it yours?

[sinful sentinel]

a friend generated it and shared, I don't think he deleted it

[crimson osprey]

suno just genereated an almost 4 minute long clip. i thought the maximum were 2 minutes?

[low cypress]

Please check with him first

[sinful sentinel]

ok doing that now

[low cypress]

Let him also try to use that url

[sinful sentinel]

the url doesn't work for him, he says he didn't delete it, and the song is gone, it's also not in the trash

[low cypress]

I advise him to send an email then to [e-mail]

[wary pumice]

You can add an unlimited number of the same track to your playlist

epic

[mortal tide]

I keep putting "saxophone" in style window but there is no saxophone in the songs

[low cypress]

Spam it more times.. sometimes you have to force it. And if it is a genre that normaly doesn't have much related with saxphone, it can be hard too.

[mortal tide]

Costing me credits tho

I am making synthwave and before it worked basically every time so its a big diff

Could any other style prompt be blocking the saxophone?

80s synthwave, saxophone, steady clean bass, clear sound, epic, emotional depth, neon, cyberpunk

[low cypress]

Yes, we are currently in a phase where certainty is limited and full control has not yet been achieved. If this unsettles you, I would recommend revisiting Suno at a later stage of development. I personally embrace this stage because it gives me a head start. I am a volunteer guide and still pay for 2 premium accounts. Knowing it will learn me and support Suno.

[mortal tide]

Thanks I guess

Ty that was a more helpful answer, I will try this, appreciate it

Hopefully wont lessen the chance at least

[fluid wigeon]

Is it possible that Suno AI only works with AI-generated lyrics? A self-written text leads to bizarre songs that have nothing to do with the lyrics or the prompt. Is this in any way normal or intentional?

[low cypress]

No, sometimes you just have to spam.. even with sax sax saxphone saxphone

[mortal tide]

Thats new for me but Ok sure

[low cypress]

look above the lyric field. Do you see the toggle "custom"?

[low cypress]

that was literly the first thing I said 😹 "Sometimes you have to spam it " 😉

[mortal tide]

Yeah it hasnt been like that at all before

[low cypress]

Every model can shift a little... and then just ask the community. There is always someone who puzzles and play and spends more credits than others 😄

[mortal tide]

You seem to get so defensive when someone points out bugs or issues, Jimothy at least had a suggestion to add "sax" rather than say "hey it isnt perfect, live with wasting your credits that you pay us a lot for".

[low cypress]

No.. I am trying to set the right expectations. Too many think there is a whole list, and whole set of rules that work. And get upset if they spend money and it doesn't work. I am trying to be transparent.

[fluid wigeon]

Unfortunately I don't think so, it happens again and again. If a lyric is completely generated with ChatGPT, it works, but as soon as a verse or chorus is self-written, the generated songs are miserable. There has to be a reason for that. Sorry to be annoying, but this inconvenience annoys me too.

[low cypress]

Oi... 😅
So your lyrics aren't working? Maybe my video could guide you? https://www.youtube.com/watch?v=5uJw5kb05as
I am not sure if you are following the guidelines how to use metatags? Our community website could help you: https://www.suno.wiki/

[fluid wigeon]

My lyrics are like all AI lyrics, but I wrote them myself, they don't violate anything in terms of content, but the singing is always slurred, the language is changed, or words are changed. I just suspect that human work doesn't count for shit at Suno AI, and only AI-generated Lyrics work anymore.

[low cypress]

Hey Me140104, I hear you on the lyrics issue. It’s important to remember that AI, including Suno, is still evolving and sometimes struggles with language and pronunciation. It’s not perfect and can change or slur words unexpectedly. Sometimes it helps if you change genres or change your lyrics a bit.

[fluid wigeon]

Again, this isn't helpful, as other lyrics like this work once they're written by an AI. This problem is basically, and that can't have anything to do with the fact that Suno AI is still not fully developed... Seriously... I'm not imagining it, I've noticed it many times.

[half cliff]

Why is V3's default always to add background harmonised Vocalists? This should not be default behaviour.

[low cypress]

Me140104, I understand your frustration, and I want to be transparent here. As a volunteer guide, I've shared all the solutions and insights I have. Suno AI, like all AI, has its limits and can be unpredictable. If these issues persist, and none of the suggested methods have helped, I’d recommend reaching out directly to Suno's support team for further assistance via [e-mail]

[half cliff]

V3 alpha didn't always add choirs to songs. As it is, only 1-2 in every 50 songs do not devolve into choirs.

[mortal tide]

We are many that miss alpha, @half cliff 😦 Hopefully there is a new version very soon

[low cypress]

The DEVS left this statement of the current version: #💡┃suggestions-web message

[fluid wigeon]

This has nothing to do with limits and unpredictability. I think you're missing the point of the problem. It works as soon as I enter an AI Lyrics and then it works really well. But as I said, as soon as something of your own comes along, it bugs. And I just want to address this issue to draw attention. Sorry that, as I said, it was annoying, but as I said, it's also very annoying that human creativity seems to be trampled on, Have a nice evening everyone, I will not consider further support here now.

[half cliff]

V3 is worse in every respect I do not see how alpha would be a step back. Rather it looks to me like that the devs have taken the route of improving performance by reducing quality to maximise customers as when the alpha was running it was only being used by a smaller number of people (so less load on systems). As it is right now, you will waste 100 credits to find a single normal song that isn't a choir and the chances that this song is even decent is low as a result of it being 1 in 20. So this means you need to spend 1000+ credits just to find a good opening to a single song that isn't a choir fest.

[low cypress]

With all respect. I don't see the bugs. The whole day people make songs with custom lyrics and nobody else is complaining about the same thing you do. I make songs myself too, made a few today and don't have seen what you are saying. So I personally cannot help you. I ONLY use costum. But the Devs will look here too. If I made a mistake, I appogize in advance.

[fluid wigeon]

There are bugs here, but as I said, I won't consider any more support because it doesn't help anyway now. As I said, have a nice evening.

[knotty pebble]

I just wanted to drop in to say I'm really enjoying v3 and it's worked a lot better for me than V2. I thought the devs could use some positivity in their day.

[tame latch]

thx it is a rough day

[mortal tide]

Stay strong, hopefully next version kicks ass and us whiners turn into praisers instead. Thanks for all the hard work you and the rest of the team does on the daily. 💜

[sweet spruce]

I don't know if this is some kind of bug but almost all songs with a choir and a full band creates a lot of distortion. Especially the vocals are affected by a lot of distortion. Has this always been the case or is it something that has emerged as arrangements become more and more advanced?

[sweet spruce]

So, is there any way to remove the distortion when there is a lot going on like many instruments and choirs. For example in the chorus. I almost always get a lot of distortion in the chorus.

[gleaming halo]

the only way is to be vocal about it here and to ask for v3 alpha to be put back which did not have this problem

[loud lotus]

But they kind of made it clear that alpha won't come back. En-Tal summed it up pretty well. The people who like V3 are probably all non paying users, who only had V2 before. For them it's a big improvement. Everyone who pays and used alpha before is upset, as it was better. Let's hope they can somehow fix the choir issue in a different way.

[gleaming halo]

Upsetting all your paying customers and not having a plan to fix it in the short term is not a sustainable business model. It's also not a good sign moving forward. Hey we can break anything, won't fix it and don't care if everyone is upset about it.

[sweet spruce]

Ok, I was thinking to start a subscription but if the problem with the distortion is the same for paid and free accounts then I have to wait until it’s solved…

[vale cradle]

Couldn't generate that.
Prompt flagged for moderation

whats wrong xD "EDM- und Trap, Orchesterinstrumente,Duduk-Melodie aus der armenischen Musik,Shakuhachi-Sample"

[steep creek]

Hello everyone, how are you? Lately, I've been having some issues with Suno. I'm from Brazil and I make some rock-style songs with Suno. However, lately I've been facing a big conflict with Suno's language. Whenever I write a song, Suno most of the time randomly sings it in Portuguese from Portugal, even though I set the prompt to sing in my language most of the time it doesn't work out :( and the worst part is that this happens always at specific times, like at night. I'd like to solve this problem. If anyone can help me, I'd be very honored! I've paid for the pro plan and I don't want to waste my points for nothing. Thank you!

[willow skiff]

nothing I spent more than 1500 credits just to find a good track, but it seems that only the choral versions are available also he no longer gets the genres right and skips the words of the songs to his liking. The V3 Alpha was better and honestly it would be good to restore it under another name at least for paying users also as V3 .5. I am forced to suspend payment if this does not happen. thanks

[dire burrow]

hello I simply pay the most expensive it must be renewed today in fact I noticed at the moment that I have 2 active subscriptions the most expensive and the least expensive I was taken from the least expensive I did not receive credit my the least expensive I didn't want it and I was not charged the most expensive one while I want it and I can be reimbursed

I just came to maintain it so I canceled the cheapest

[whole verge]

Hey there. I've noticed that if you create a song without adding an [Intro] tag, then you're almost guaranteed to generate a song that begins with vocals. But that's a problem because, for me, the [Intro] tag only seems to work about 25% of the time - most of the time it's just ignored. In fact, tags in general are quite hit and miss. Songs will often begin generating a chorus a few lines before the [Chorus] tag, and other times the verse will overlap with the chorus, ignoring the tags.

[mental prawn]

I agree the same thing what you're saying, bro.

[sinful silo]

I'm getting inconsistent results when continuing songs? First it wouldn't sing the second verse although I clearly tagged it [Verse 2]. And now it sings it but it's different from the first verse and it won't repeat the chorus the same way either? Is the AI having a bad day? 🤔

[knotty mason]

Better than V2 but trash compared to V3 alpha.

[knotty pebble]

@fluid wigeon honestly, I'm getting better results with V3 than I did with a V3 alpha.

[knotty mason]

It must be genre specific. For me v3 alpha had clear vocals, consistently reproducible styles, more consistent results. V3 has distorted vocals, hard to reproduce the same genre, doesn't repeat choruses(sings them totally different)

V2 to V3 alpha was such a leap that I was excited and thought this was viable to make songs that I actually wanted to listen to. I bumped up to the highest paid subscription and made a few songs that I actively listen to on a daily basis. V3 is a step down from alpha and puts it back into the category where it's a cool demo but hard to make a full listenable song.

[cloud python]

My genre is "alternative," "grunge" and "90s" (meant to be akin to Green Day or Third Eye Blind), and I have the same "choir" problem in V3. But my problems with V3 extend much further than that -- for me, V3 just seems totally, completely, and utterly BROKEN. It repeats verses/choruses, ignores [start], [end], or [synthisizer

[final plank]

fr

[cloud python]

Vocals are muffled like the singer(s) have cotton balls in their mouths, the overall sound quality is inferior to V2 & V3A, the version's ability to produce catchy, unique melodies is virtually non-existent, a bad tendency to hallucinate. It doesn't seem to understand what the "alternative" or "grunge" genres are, because nothing it gives me is anything approaching what I got in V2 & V3A, which both were not only consistent in their quality, but in the voices as well. The "singer" for one song I did in V2, "Dollars and Desperation" sounds identical to the one used in "Kindred Spirits," both using the "alternative, grunge, 90s, angsty" genre keywords.

As others also have pointed out, V3 also seems to suffer from a considerable "autotuny" sound to it's vocals, a problem I didn't have in V2 or V3A.

On top of this, V2 also now seems to have become "broken," as ever since V3 was released, I've been unable to get any good results from V2 either. It's as if the whole model just got thrown into chaos.

[cloud python]

Based on your "Jane" example (which sounds like a great song, BTW...minus V3 mucking up the vocals), V3 seems to be adding a vocal layer every bar, so it starts the song with one layer, and with each bar it adds another one to the song, so by the 2 minute mark, you've got like 20 "singers" on the track.

[eager acorn]

I wrote a Cantonese song, but when I continued to write it (Continue From This Song), the second part became a Mandarin song. I didn't change any settings.

[cloud python]

No, I've produced 4 or 5 songs with V2 & V3A using lyrics I wrote myself, and I had fantastic results. V3 seems to be sketchy with both your own lyrics or Suno's lyrics.

[cloud python]

It might be more helpful to users if Suno would release some kind of "instruction manual" on how to best use the new version, since they allegedly know more about it's inner workings than we do. But so far, the vibe I'm getting from this thread is that V3 seems to be a pandora's box even to the devs.

Based on the feedback I'm seeing here, the response always seems to be, 🤷‍♂️ "Try this, try that, try this.."

Well damn - if it was this dodgy, why was it officially released?

[low cypress]

Hey urbandecayunion, I totally get where you're coming from. An instruction manual would be handy, but here’s the thing with AI and Suno – it's groundbreaking and unique, so there's a lot we're still figuring out. The devs understand the model's training but predicting its responses can be challenging. It's why there's no strict manual yet, just tips and guidance. V3’s release, despite its unpredictability, is part of the journey in AI development, where user feedback is crucial for improvement.

[cloud python]

I understand completey, and this is probably a loaded question, but any idea when we might get a V4 Alpha?

The thing is, a lot of us PAID to troubleshoot V3A for Suno in anticipation of an official V3 release, only to be left feeling "ripped off" afterwards with the final version ends up being inferior...

[low cypress]

V3 is not even released a week. So I don't expect next week. But tbh I personally have no idea. I am a volunteer who's answers typically known questions (when you hang around as much as I am 😉 ) But I am user as well. And I pay too. I personally don't feel ripped off because I really feel the Devs doing everything in their power to bring the most advanced music AI there is. I've been playing around with V3 too and yes, it behaves differently, yet it's all in the eye of the beholder, and which genres you pick. But again this is new, they are a small team, and working more than 40 hours per week. We just have to have a little faith and patience.

[half cliff]

I don't think it's correct to say that it's all in the eye of the beholder when you have every song becoming choirs.

That's a categorical reduction in the AI's ability to diversify if it always tries to resolve the same lr similar output.

[low cypress]

Feel free to see the polls.. more people like than dislike it: #🗳┃mini-polls message

[half cliff]

Most people didn't even touch V3A so of course there will be more liking V3 as their point of reference is V2.

[cloud python]

But...surely you can't deny that this song I just produced in V3 (Fleeting Elation) is "better" than what I did weeks ago in V2 (Painted Faces)? The vocals in the V3 song sound distorted and garbled. The V2 song is radio quality.

https://app.suno.ai/song/16653def-8786-4484-9b03-57d6117977e7

https://app.suno.ai/song/fa23de95-4ed4-4c44-98f3-55a9f5182e66

[low cypress]

Here screenshot what the Dev said about the model.

But I am going to bed good people. It's after 4 am and don't have the energy get into debat about this. I am Miss Nobody.. I am a guide.. Only trying to provide the answers that are here on discord. I am in no mood to be anybodies lighting rodd. Have a great day/night/morning

[cloud python]

^^ Ok, but like....I'm having to "dislike" generations in the neighborhood of 30 to 50. As some of us have said, this was not the case with V2 or V3A - V2 would produce a satisfactory result usually with 3 attempts, V3A would do so usually within 10 attempts.

V3 NEVER gets it right. So asking us to burn the thousands of tokens we paid for when the problems have been clearly defined on Discord seems a bit counterproductive.

Not to sound tin-hatted, but V3 almost seems deliberately engineered to be a credit-eater so users will be forced into paying for top-ups or higher-tier subscriptions. I'm not saying that's the case, I'm just saying that NO FREE USER is going to continue using it because they'll burn through 50 credits and get nothing.

[cloud python]

You're not a Miss Nobody. We appreciate your assistance, always.

[low cypress]

If you go NOW to this channel you can see people talk wit the Dev Tony: https://discord.com/channels/1069381916492562582/1154587937611190354 (maybe it helps)

[lavish herald]

I've been getting a weird (possibly display only) bug since yesterday where it seems to take extra credits every time I generate. I have to refresh the page to get them back. Means that I generate once and end up at 35 instead of 40, refresh to get them back, generate again and end up at 25, etc.

[sinful silo]

Not a big deal but the Explore / New page seems stuck on the same songs

[gleaming halo]

When playing a song, just after generating, the playback scrollbar runs up to one minute. When the page ismstagnant for a while it looks like the whole song has loaded and then it does show the real time.

[balmy shadow]

Why it can generate "continue" ? And it deleted my good one who i just wanted to continue to the point. ( Continue from "0:1:44" and nothing complete 🪲🪲🪲🪲🪲) and i lost my work 😭😭😭😭😭😭😭😭😭😱

And credits keeping taking 🤔

[fallow kindle]

The strength of v3 definitely lies in the Doo Wop of the 50s, the upbeat soul of the 60s, and the hard rock of the 80s (my favorite luckily)... but where it seriously falters is in the R&B of the 90s and the pop of the 80s. NOTE! There are some songs that are 80s hard rock and I notice a certain degree of Autotune. Please fix this! This AI has potential! But the Pop and a few 80s hard rock songs are reporting bugs.

[strange fjord]

I also believe that this is the Case. There should have been a Poll With only paid subscriber involved who actually tested V3Alpha. ..

[mortal tide]

@tame latchOh and I forgot to mention this. When I asked you for a refund I tried like 5-6 other music generatin AIs. Well, I cancelled my refund. Even tho I am not a big fan of V3, it is still a lot better than the rest of the AIs I tried, so Im here and sticking to it and supporting Suno and looking forward to the next version. Suno is on the path to greatness.

[spark fiber]

Hello,

Is it just me, or has the entire "New" library stopped updating at some point and isn't adding any new tracks?

[willow pollen]

it actually seems to be avoiding some words. I had some test lyrics for a rock song and I just used random Ramones lines. The line 'Beat on the Brat With a Baseball Bat' wasn't blocked but wasn't sung. Switching to 'Beet' instead of "Beat" made the line audible

[spark fiber]

Few Bugs:

1. The "New" library isn't updating.
2. When I add a freshly generated full track to a playlist, many indirectly generated parts of the track are added to the playlist as well, even those that have been rejected and flagged with a thumbs-down.
   Regarding point 2: To remove these unwanted parts from the playlist, I HAVE to throw them in the trash because the "remove from playlist" option does nothing...
3. With the "continue" option enabled to listen to generated tracks on my phone, I have to temporarily switch to "simple view". When expanding the "custom" options again, the connection with the previous part of the song is lost - I have to go back to the previous part of the song, click "continue", select the time when the new part of the song should start, while the text, name, and genre remain unchanged - I didn't have this problem in V3 Alpha.
4. Generated tracks seem to be in mono, especially if they have Overdriven Rhythm Guitars. Sometimes it's due to artifacts that cancel phases, and other times there really is only one guitar "in the middle".

[mortal tide]

Did you try to clear cache and cookies? It works fine for me so it might be a local issue

[proven socket]

Suno is leading the way 🚀

[proven socket]

I think it is stuck actually 😛, but im sure they're aware of this
it's been showing the same tracks for an hour or so (maybe longer)
they might be testing something out

[mortal tide]

Ooh ok ok

[half cliff]

AI skips entire verses. Had it just ignore everything I wrote and jump straight to the final line then hallucinate from thereon. Keeps happening on continues.

Have wasted like 600 credits to only get 2-3 continues that actually follow the lyrics

[willow pollen]

Oh @proven socket , weren't you the one who wanted to post an example of a non-choir song just before the server crashed yesterday?

[proven socket]

it was actually a joke track lol, i had writen "ethereal-choir choir choir large-choir gospel-choir choir" etc into the prompt and titled the song "Suno Solo Singer Example" 🤪

[willow pollen]

(also, just completely off-topic : Iain M. Banks ftw;)

[willow pollen]

joke's on you though : you used up your keyboard pointlessly. 'male singer' would have been enough 😛

by your lack of reaction I assume you are not aquainted with the SciFi author Iain M. Banks and his seminal work "Player of Games" 😛

[proven socket]

oooh haha, i somehow thought that comment was related to someone else. I did once look that up and read the Wikipedia article on, but I know nothing about it

[willow pollen]

Iain M. Banks is a must-read if you like SciFi 🙂

[proven socket]

might have a look into it eventually. I like Sci Fi tv shows and movies, though haven't done as much reading 📚

[nova lynx]

This site says it can be used up to 10 times a day, but for some reason it can only be used 5 times.
https://www.techno-edge.net/article/2024/03/22/3020.html

[limber terrace]

please give us back v3-alpha model, just wasting credits & time with v3, all v3 results are just unusable and miles away from v3-alpha which produced such great results. plz ❤️

[slate zodiac]

Hi Guys, i have problems with the "continue of song" function. The 2nd part is always only 3 seconds long and my credits are refound. Could somebody help me with this?

[crimson osprey]

V3 alpha had the same problems that V3 has now. Simply check out the polls regarding alpha improvements to see what issues people had with alpha. I have used thousands of credits on both versions and created songs with both versions. Using AI Generation, especially with very early versions, is extremely random. It is absolutely possible that you had great results with V3 alpha but even if V3 might reduce that chance for distorted voices you could have more distorted voices there. randomness just works that way and I doubt any of us spent enough credits for any valid statistics regarding quality. only the devs can do that.
The devs have made clear that a rollback to alpha is not possible.

If you want to help improving Suno, then dislike the bad results and like the good results.

I know it feels weird to pay for a product and might getting unusuable results. but this is how generative AI works. you buy calculation time on a server and the right to use their model. Music generation is still in a very early stage. and if you feel like you don't get your moneys worth, I advice you to stop spending money.
Though I also want to point out... even if you spend thousands of credits on a single song, that is still incredible cheap for producing a whole song...

[crimson osprey]

if you feel like the AI keeps giving bad results, you might need go a few steps back or change up your prompts. I also experiences this kind of "stuck" situation. it feels sad to abandon a certain stage but in my experience it saves both time and credits to start new

[willow pollen]

the devs have already stated they will not bring V3A back. That being said : there was a dramatic reduction of vocal quality fr omV3A to V3, which is only apparent when you don't want a choir singing or a chorus effect , and apparently is also genre dependant. Enough people have documented it by now to make 'look at the poll' irrelevant.

it's got nothing to do with 'prompt better', and just a degradation of the model

[half cliff]

The actual quality of voices when you don't get a choir or duo is really good with V3, but due to how hard V3 pushes choirs, it can make a lot of solo singing sound low quality because of the low choir effect being played.

[willow pollen]

yes. The first tact is PERFECT ... then the surprise sudden choir starts to show up.

by the 4th bar you end up in glee club musical hell

[astral adder]

Randomness is a valid point but you can't sell that as the THE reason people complain. They said alpha costed 4x more. The question is what optimizations were made

[half cliff]

Yeah. As time goes on in the song the choir gets more and kore erratic until it's just incoherent screaming, or so out of sync and so low bitrate that it's nails on a board.

[crimson osprey]

I am simply stating that nobody here should state a reduction of quality as facts, as nobody here can create valid statistics regarding this.

[limber terrace]

not asking for rollback, just for adding also v3-alpha for optional model selection, so people can decide which works better for them (for me clearly v3 is a big downgrade of the model), at least till a new version is released...

"stop spending money" => Funny when I just subscribed for the premier plan with yearly after good experiences with v3-alpha, and now the v3 results are just miles away, and clearly its not a rnd/prompt issue.

I would even PAY EXTRA for v3-alpha access

[half cliff]

Ah, so my guess was right then. I figured they did something like that.

[crimson osprey]

the degrading of quality has been a problem with V3 alpha too. none of the problems that are mentioned here are new to v3. you are stating that they are happen more often now. and I say we don't have enough data to prove that.
on my side it works well and i get less distorted results and v3 follows my prompts better than alpha.

[half cliff]

Seen enough AI models go down this route enough times to spot when it's "optimisation" that's the cause lol.

[willow pollen]

sigh ... empiric results. I dare you to produce a song like this one, featuring a single isolated voice over the whole song, with V3. https://app.suno.ai/song/82bf5dbe-426b-4222-8168-1e46bc451939

I'll buy you a premium month if you succeed.

[half cliff]

So far the only way I've managed it is by interrupting choir's with instrumental segments. But that makes for a rather choppy song in the long run due to how frequently it happens.

[willow pollen]

yeah ... or generating a segment, then continuing from just before thechoir starts with v2, then continuing again ...

[half cliff]

Pretty much impossible to do it with a duet though. I've given up on song I'm making now not having choirs and just try to pick the ones with the least choir. I started with 7200ish credits. I'm on 3470 now.

[willow pollen]

which reminds me : the premium gift offer is only valid if you have a >1:30 mn initial segment and not chopped 10-20 secs segments put together

I can rest easy in the knowledge that it is impossible

[half cliff]

After this song I'm just gonna wait for them to fix it. No point in starting another.

[limber terrace]

So why not add v3-alpha as another option? Nothing to complain then, you will be happy, and all other wishing v3-alpha back also 🙂

I would not even mind if they add a paid option for this, lets double the premier price if someone want to use this model, I would buy it directly!

[crimson osprey]

sure, i am also a fan of more options.
but since the devs already said they won't bring it back, I doubt this would happen. all i can say it currently works for me. I reduce the degrading of quality by using as little continues as possible (like it was recommended during alpha). and i hope we can soon get better models 🙂

[willow pollen]

sadly it degrades significantly within the initial generation now, so 'avoid continues' is moot.

[crimson osprey]

very strange. I don't recall having that at all. curious how we can have such different outcomes.

[willow pollen]

what type of music are you generating?

[limber terrace]

and i only can say I would pay even 4x the price of premier.
If v3-alpha has 4x ressource-usage, just fair, let me(us) pay it if we want it...

[crimson osprey]

on v3 i am currently doing pop, rock and disco music
on alpha i trid jazz too but i didn't like the results on alpha with jazz

back in alpha i felt like the degrading with jazz was stronger. might be because of the training data having already noise in them since those are often older recordings.

[willow pollen]

hmm ... I used to do a lot of rock ... again with single voice and the occasional chorus effect. Check for example https://app.suno.ai/song/daff5dbd-e06f-4536-acb4-66f72aa087a5 ... you'll notice that the singer's voice stays clear, with a very slight chorus effect in the ... aeh .. chorus 😉 but there is an obvious single singer throughout. It's impossible to generate anything similar with V3 ... I spent around 1500c trying.

[half cliff]

All I can say is that when I'm making slow jazz piano romance music, choirs shouldn't be barging in and taking over.

Like it shouldn't even tangentially related.

So whatever optimisation they've done, they've really screwed up the dispersion of the tokens in the data set. Because nothing about slow jazz piano romance should be correlating with huge 200 people choirs.

[willow pollen]

I just tried it again, just for you ... https://app.suno.ai/song/96dc2724-420e-480d-8d42-f3eadc437105 and https://app.suno.ai/song/2a5ef92b-e207-4442-b72e-9c1c33717cee ... those are NOT outliers. All generations I had with V3 exhibited this behaviour. It's also noticeable that the music is ... 'moar boring' somehow. it lacks the dynamism of the V3a example

also : thanks for the lyrics 😉

[half cliff]

You can fix the dynamism a bit by putting [dynamic] in the lyrics.

[willow pollen]

yeah ... or generating more and choosing, but as the choir mush's the music anyway ...

[half cliff]

Yep.

[charred gust]

Does anyone have a discount coupon for me to sign?

[willow pollen]

ask in general or offtopic, but you'll probably be disappointed

[bold marten]

chrome, auto refresh bug: has i just managed to deleted the full song.which some how managed to get in the trash can.although it seems to be here still? on discord

[charred gust]

ok

[crimson osprey]

you don't need to try anything again for me 🙂
like i said, i usually brute force my results. even if i spend the 10k credits of my subscription for just one song it is totally worth it since it saves me so much time.

at all times I believed you that you are getting bad results. never denied that. maybe i am just lucky that my prefered style of music goes better with the latest model...
as I said before i would welcome having the choice between models. I just don't think it won't bring us anywhere demanding this all the time, as it was declined. In my opinion, we get faster to a better model if we give feedback on the tracks for future training.

[half cliff]

All jazz music, all have choirs jump in early:

https://app.suno.ai/song/52d74e70-6a4c-411d-8547-897afe3cd05a/

https://app.suno.ai/song/7629b012-ac1a-4c0c-8b9f-5669950ee762/

https://app.suno.ai/song/ae395eb9-d489-466f-95df-9646bb02a6f1/

[willow pollen]

don't worry ... I have >11K credit, as my sub renewed just before V3 was released and I just don't generate much anymore as I don't get any useable results

[bold marten]

should this be in the show not the bugs sections

[willow pollen]

but yeah ... my initial point was : the devs have spoken, so it's pointless to ask for V3a, no matter how much sense it would make

[willow pollen]

nah .. .it shows a bug ... correct channel

[half cliff]

No, they all have a bug. They shouldn't have any choirs. Choirs shouldn't even be correlating with that music but it is

[crimson osprey]

this whole thing somehow reminds me to that meme where this guy shouts at chat gpt to generate an image of a bowl of ramen without chopsticks and it just wasn't capable of doing that.

[gleaming halo]

WHere was ot said Alpha costs 4 time more? I'm willing to pay more for good quality. But the whole overdubbing voices were added in final V3 and it makes no sense to think this is to reduce cost.

[bold marten]

@half cliff ok, just don't want to let people think that they can paste songs anywhere and i sure the devs and mod have enough problems solving than people dropping in different channels. So Sorry if that was a bit rude.

[willow pollen]

the main problem here is that it USED to be capable. And many have subb'd or even (as I did) upgraded their sub to continue using a model that gave some great results ... just to find out that it was nerfed (for their style of music/preferences) to oblivion

[crimson osprey]

well we don't know that. it seems like the training changed since it lost capabilities. but it could be for all kinds of reasons.

I can imagine getting training data especially in music is a mine field

[half cliff]

No worries. I was just posting proof of the issue. That even with slow jazz romance songs choirs still show up.

[willow pollen]

I wasn't blaming the devs ... I just described what happened. WHY it happened is on a completely different plate

[crimson osprey]

" just to find out that it was nerfed (for their style of music/preferences) to oblivion"
ummm? you said they nerfed it for their style of music?

[willow pollen]

the model was definitely nerfed, yes. Doesn't imply it was on purpose ... but the model definitely degraded significantly for a portion of the users

[half cliff]

In my experience with AI companies, they never drop back to their superior but more expensive options except in the case of where they let you have API access and pay per token rather than per plan.

V3 will likely retain these issues in some form or another as time goes on. I fully expect for the current issue of it ignoring lyrics to become a bigger problem when they fix the choir and sound quality issues with more optimisation.

We won't see V3A quality unless they open up API access, or when release V4 in the future.

[limber terrace]

we got it that for you v3 works well, maybe even better.

stop arguing that for me/us/other it does CLEARLY not!

[crimson osprey]

where did I do that? that isn't even the topic here?

[proven socket]

I'd hope that's not the reason, and if it is... as more competition evolves that should help (since not everyone will just work on optimisation alone). The main thing I noticed in v3 is sometimes a light distortion effect that was more subtle in v3 alpha.

I have faith that this will improve as they go forward, even if v3 might have some quirks it probably improved on about 200 other things 😆
Though I will admit getting into a good rhythm with using v3 alpha lol. That said... now I'm starting to feel a bit more in line with v3 as I use it more and more.

[willow pollen]

anyway, the only reaction possible on the affected users part is to unsubscribe and hope the problem gets fixed at some point (even though the devs seem to find the current state acceptable 'because the poll says it's better'). It's sad, but it's the way it goes.

[crimson osprey]

absolutely agree with you there

[half cliff]

The devs will further optimise the model and smooth out these issues but these kinds of optimisations will cause other issues to crop up. Like a lazy AI that tries to cut corners by saving output by ignoring instructions/lyrics, like GPT4 Turbo.

It'll become more useful than it is now but I doubt it'll be as good as V3A.

[crimson osprey]

anyways, before this discussion gets more emotional, I am out. hope you guys find a way to get your results! best of luck!

[half cliff]

This is true. And unlike GPT, these music models are nowhere near as prohibitive to run energy wise. so there will absolutely be a lot of competition going forward.

[proven socket]

do you know that for sure, this part: "these music models are nowhere near as prohibitive to run energy wise"
that's good to know if that's the case

[willow pollen]

training is the prohibitive part

generating is "easy" ... I can easily run riffusion locally ... but the available model is extremely low quality

[half cliff]

Yes. Music models use about 12gb of vram compared to gpt4's 3520gb of vram.

[proven socket]

wow! absolutely did not know that

before if you asked me I would've probably guessed they used more. That's a very good thing to know at least

[willow pollen]

Image Generators can use even less (SD-XL DOES use ~12GB though)

[half cliff]

LLMs have way more parameters to consider. Nothing is more complex than language.

[proven socket]

i see i see. Though technically aren't theses LLMs also, or????

[half cliff]

Except for maybe physics lol

[willow pollen]

let's move this discussion to https://discord.com/channels/1069381916492562582/1159929934270181437

[half cliff]

Yeah but they're small llms.

[split moss]

Guys, I've created some nice waltz tracks, and I've been enjoying listening to them. However, when I tried to share them with others, it seems they're not working anymore, and it looks like my credits have been returned. That's so sad :(.

[wicked dune]

Could someone share if they've noticed any slowdown in generation speed today?

[tacit geyser]

I've also noticed a slowness in generations at least compared to V3 alpha.

[proven socket]

I hope this is something can be improved (if it's cost viable) because it really helps with workflow when this is fast. I mean Suno is a miraculous as is, but if it's possible 🙏

Edit: During the alpha you were able to play the generations a few seconds after pressing "Create" which was excellent for workflow. I can understand though if the longer generation time is due to having more users now. However if it can be improved it would be amazing.

[dusty sage]

Here is the list of issues and bugs with the Suno V3 application based on the user messages:

1. Vocal quality problems:

• Vocals sound distorted, with a "muffled" effect
• Excessive layering of vocal parts (choir), even in songs where it is not appropriate
• Overall degradation of vocal quality compared to the previous V3 Alpha version

2. Lyric/text following issues:

• The model often skips parts of the text or poetic verses
• The model sometimes deviates from the given lyrics and starts "hallucinating" its own lines

3. "Continue" function issues:

• When using the "Continue" function, the new part of the song is very short (around 3 seconds)
• Credits are deducted, but the desired result is not achieved

4. General issues:

• Freezing of the Explore/New page and lack of updates with new songs
• Inconsistent credit usage (more credits deducted than generated)
• Slower generation speed compared to the previous version

Users note that many of these issues were absent or less pronounced in the V3 Alpha version, and express a desire to have the ability to use the previous version again.

[humble cobalt]

The article on that website is wrong. Free users can generate 5 times per day (and it will create 2 songs each time.)

[half cliff]

It is significantly slower.

[wicked dune]

I share the same sentiments.

[haughty bolt]

Are Shamisen not included? if I put those in I don't get it out or rather it sounds more like a regular guitar

[proven socket]

are there any types of music that typically use that instrument? you'll have a higher chance of getting it that way. It's like a Japanese guitar or something isn't it? or am i thinking of something else

[haughty bolt]

not that I know of, I mean putting in electric guitar or drums etc. is working also

[proven socket]

i'm trying a few other prompts to see if i can get it

[proven socket]

Are any of these it?

• https://app.suno.ai/song/e6a59ec4-d2c5-44f3-8d66-467d6e902c1f
• https://app.suno.ai/song/8466850b-6d19-4f10-aac1-b869a41050ce
• https://app.suno.ai/song/8d47580f-1882-4225-816b-0f5ee1349a4b

[willow pollen]

okay ... I might have an actual breakthrough about the choir. NOT particularly great, but still ... If I don't expressely state what the voice should be I get a singular voice (sometimes ... other times there is NO voice), https://app.suno.ai/song/92cd24f3-2e87-4f46-936c-04468c15d60d ... as soon as I add a "voice" in the prompt it gets completely stupid https://app.suno.ai/song/32ee0fb9-14d5-4089-b7f3-8d2266c4eac4

[haughty bolt]

at least not from me, but when I listened to it they don't sound like the ones I know, the second one sounds like a turkish lute or something. The Shamisen has only three strings and has a distinctive sound

[haughty bolt]

for an example like this https://www.youtube.com/watch?v=RcqO0zkyQRo

[sweet spruce]

Do I always get two versions with the subscription account also?

[willow pollen]

I retract my claim ... V3 is still as stupid as before.

[half cliff]

Hnm. Edit: I was completely wrong

[willow pollen]

sigh ... it was a fluke apparently. No chance to get a single voice after that

[half cliff]

And here you had my hopes up

[willow pollen]

ask me... I was THAT close to resubscribe

300c gone to waste ... doesn't REALLY matter of course

[half cliff]

Try this?

• https://app.suno.ai/song/66fa601f-4918-4e59-b766-78f5dd7baeac

[haughty bolt]

ah thanks

[verbal glade]

I think a lot of the discussion here about v3-alpha vs v3 would be curbed if people felt like suno was at least aware of the user feedback and/or an indication on whether they are working on it. A lot of the feedback is because most of us feel like they aren't aware its a problem or don't think its a problem.

Is Suno aware of the feedback on discord?

[solar grove]

I'd say the answer is yes....
#🗳┃mini-polls message

[craggy charm]

It has to be slower to give it time to assemble the massed choir😀

[willow pollen]

they are aware, they don't think it's a problem because (meaningless) poll

[keen holly]

the poll doesn't offer the choice that i wanted to click: "I like v3 better than v3alpha except for the choir problem"

[willow pollen]

so which did you chose?

[half cliff]

They said they'll fix V3 so they do think it's a problem. They just don't want to use the older more expensive model.

[keen holly]

that i like v3 better. because i generally think it's gotten better at following the lyrics and insctructions in the lyrics section. but it's not really a precise answer.

i'm sort of torn

when it comes to vocals, i prefer v3alpha

i liked the cleaner sounding and more un-effected single-person vocal that v3alpha often did.

[willow pollen]

then it's NOT better ...that's the point. So many people have chosen 'it's better' despite hearing the problems that it LOOKS like it's better overall. Vocals are absolutely screwed right now, so V3a was better.

[keen holly]

it was incredibly hard for me to get songs to end properly with v3alpha, so v3 is a huge improvement there.

[willow pollen]

just in case you missed it on general : I didn't even have to do multiple takes for this song, the problem is THAT predictable ...

[keen holly]

I think the best would be v3 but with the vocals from v3alpha.

[willow pollen]

ah ... wrong track .. .this one was better : https://app.suno.ai/song/32ee0fb9-14d5-4089-b7f3-8d2266c4eac4

[verbal glade]

Ideally they would have turned v3-alpha into v3 and made whatever model is currently v3 as a new v3.1-alpha. Leaving both.

Or, I like what chatGPT does where it'll sometime generate 2 options with 2 different models and ask which one you like more

[willow pollen]

note how the style prompt was "lonely Gypsy jazz song, male voice, string guitar, violin, bongos, whimsical tune,'

not "all the neighborhood joining in"

[verbal glade]

speaking of "male voice" I found that v3 will sometimes still generate female singing, even if I specify male

[keen holly]

he he yeah, and v3alpha did that too

even [Deep male voice] could give a bright female voice sometimes

[willow pollen]

yeah ... but it didn't turn into an unwanted [insert group activity that is censored on Discord, starting with o]

[dire burrow]

@solar grove
hi I have a problem I had 2 active subscriptions the most expensive and the least expensive I wanted the most expensive so I wasn't even charged then which must be charged ajd and the least expensive that I even wanted more and what's more I didn't even receive the my comem credit is the one I don't want, it doesn't matter as long as I'm reimbursed

[solar grove]

Have you emailed [e-mail] regarding this?

[dire burrow]

no

lest go mail to suno

[solar grove]

Please start with that.

[zinc talon]

Can Suno please not have a choir singing everything? You can never get a solo singer anymore.

[craggy charm]

seems like this is a pretty universal complaint, I have pretty much stopped generating songs for this very reason. Problem now is I have 2500 credits that will get cancelled on 1st April if I don't use them.

[zinc talon]

Yeah, I stopped too. Most tracks get downright ruined.

[dawn radish]

Was just about to say this. Sometimes the songs start of as a clean vocal the all of a sudden the vocals get harmonised for the rest of the song. Very annoying. No matter what I put in I can't get just a clean solo singer. I even tried a rap song just to see if it would give a clean rap but it didn't.

[willow pollen]

just scroll up for the last 2 or 3 days ... it's a constant complaint. Devs are aware "some users seem to experience degradation" but the poll clearly shows that most people are happy with V3 compared to V3a, so ...

I have > 10K credits I'm slowly burning with stupid songs, because I just can't get anything useable anymore anyway.

the only thing you (and I) can do at this point is to accept that that is the status quo, unsubscribe (unless you feel the extra money should go to the devs and further development of course, and can spare it) and hope that vocals and prompting in the next version will at least be more alike V3a

the choir problem is actually so predictable that I even made a song that uses it #🪲┃bugs-web message

[thorn obsidian]

i think most people would just vote yes on any question they'd ask ^^ They should limit the polls to people who have X generations with the version/s they ask questions about. Discord is prob. not a good place for such polls anyway

[elder wigeon]

I had two issues last night...one I shrugged off a little till I found out I lost over 200 tokens into the new month. New subscription turn over is for today...there is No time set anywhere to tell exactly when you roll over..its a guesstimate- which is not cool for anxiety. Lol..anway I was tying to get through 500 tokens so I didnt lose them...im typing lyrics and the 'Generate' button kept clicking over..No way to stop THAT!.. But since It still showed I hadnt used them up yet I was like just a bug to mention... when I got to ZERO tokens left I went to bed...when I got up, I checked to see if My new month had started, yes AND I had all ready used 200 of my monthly plan! Wtf? NOT cool Suno.. technically thats 3 web issues in one instance!

[half cliff]

Yeah. I ended up spending 4720 credits on making that song I was talking about. Turned out good, but it still had some choir, just managed. But it shouldn't have taken 4720 credits, most of which was fighting the AI to not give shrill choirs (I relented to the idea that it was gonna have some choir)

[willow pollen]

not completely. People say that v3 has improved over V3a because it HAS improved in some areas (voice clarity ... you know.. .for those 3 seconds no choir appears) and instrumentation sounds better (composition is worse imo, but that's subjective I guess). Most people are aware of the sudden surprise Glee Social Club Musical bug, but depending onthe type of music they generate they don't mind. Choir and chorus in pop or hiphop is usual ...

so as a few things got better they say yes, but would prefer a "in some points yes, in others no' option. Basically it's a glas half-full/half empty issue. Depending on your need you'll say yes or no

[half cliff]

You shouldn't have to burn through half of your monthly credits to get 1 good completed song that doesn't hurt your ears

[willow pollen]

especially as a 500c song was considered expensive with V3A and then it was generally stellar in quality (albeit a bit scratchy on thevoice)

[half cliff]

Granted I was making a duet so I wasn't helping myself. But that's still like, 3000 credits too much.

[half cliff]

Yes. Typically I would spend about 500-700 credits on a good song. I have no interest in making more for now.

[willow pollen]

yeah ... I generate ambient game instrumental at the moment, although honestly I have now enough for 50 games and still too many credits. Though again I have a feeling that the compositional skills of V3 are not up to par with V3a. The sound is better, but it's very repetitive, even in instrumentals.

[half cliff]

Honestly it'd be great if we had access to the models temperature setting. Would probably solve a lot of issues

[willow pollen]

hmm ... maybe

[zinc talon]

I even generated a spoken word track but guess what

also, of course a piano came in at the end 🙂 because Suno!

[vernal harbor]

does anyone know why continuing song with V3 sometime finishes with 2mins long, but most of times just 1 minute even with some lyrics left 😄

[merry sorrel]

I'm not happy with either of them, it all depends on the bitrate and quality, how muddy a track sounds, etc. It varies for any version I've ever tried, no idea what it's caused by though

[winged spruce]

Is this me or is "explore" broken at the moment? Especially "new" section?

[mortal tide]

Yeah seems broken

[dawn radish]

I get that all the time. Part 1 is 2mins part 2 is never more than 1 minute.

[modest iron]

UI is starting to bug out a bit, mainly song images not loading in the list, the menu button hiding itself, and the custom mode container is overextending into the space to its right.

(since today)

[quasi rose]

[Download Bug] If you renamed a song and download it after, the name shown in the Video download is the old one instead of the new one. Happened even like 15 minutes after renaming so it's not an instant thing 😄

[sterile brook]

It feels like v3 doesn't know enough about synthwave and its subgenres. Spending 1k tokens on just one decent track seems to be the norm. I don't think that's fair 😦

[solar grove]

Never?

::cough::

https://app.suno.ai/song/2c745f13-2001-4981-8f4a-d912504558c1

[delicate warren]

Thanks for helping get “freestyle” whitelisted!! 🥳✨💖🤩

[viscid remnant]

Hi, is that possible that when I click on the create button it starts to generate the same song twice and consuming credits for 2 songs while both of them will be the exact same?

[visual skiff]

Hi, I wanted to create a continuation of an existing song, so I created it from 'Continue From This Song' with new lyrics and the same Style and Title, with v3, but I got an empty song with no sound in it I tried it twice, all empty, 3 to 10 seconds songs. Is this a bug? By the way, the lyrics are in Japanese. Thanks.

[solar grove]

Did the part you are trying to continue from have a quiet moment toward the ending of that part? Try setting an earlier continue part to get some music at the end of the part, even some trailing vocals can work. Because if the AI thinks the song is over it won't continue it.

[dense mountain]

Same

[visual skiff]

I see. That could be the reason. I will try your suggestion. Thanks for your help!

[verbal glade]

Toward the end of that, the distortion in the voice starts to move away from single singer and toward multi. around 3:48

I think thats the issue. Its overfitting to the voice so bad that its distorting

[blissful harbor]

since today every prompt i try gives me an error

Couldn't generate that.
Song Description flagged for moderation

[low cypress]

Try different lyrics.. I think you get that because the lyrics are too graphic

[solar grove]

I think I hate you right now for pointing out that little bit of distortion toward the end....

I can't now hear it now.

[low cypress]

a little bit distortion is still far away from hearing whole choirs 😂

[solar grove]

If I can't fix it I may have to live with it.

(╯°□°)╯︵ ┻━┻

[low cypress]

https://tenor.com/view/sussy-cat-pain-smiling-cat-agony-gif-24664701

[solar grove]

lulz

I'm wondering if that distortion is just baked in. The few times I got a good sound like what I had it seems to fuzz right about the same spots again...

[gleaming halo]

baked in to what version 🙂

[solar grove]

Whatever the song segment is.

[low cypress]

I actually did a different approach by trying things that have choirs.. I would have expect that it would be distorted as hell but it pleasantly surprised me. https://app.suno.ai/song/edd0eb2d-8951-464d-aeb2-64e437d8cfda

[gleaming halo]

je prompt koor in het Nederlands dat telt niet 🙂

And then combo with latin?

[low cypress]

heb ik nederlands gebruik? huh waar dan? lol

[gleaming halo]

hah ok my google translate making a mess of the latin …

[low cypress]

try chatgpt lol

[solar grove]

I give up. I've decided to live with it. I like the original too much. Warts and all.

[zinc talon]

Point taken (although it kind of sort of glitches out sometimes 😉 )

[solar grove]

Yes, but it always did that to a certain extent. V1-v3...

If you expect perfection get a studio and be prepared to do a million retakes.

And overdubs

And all sorts of franken cuts.

I tend to treat a song as a sort of "live event" when I'm making it.

I have to for my sanity.

[zinc talon]

My only comparison is v2-v3a, but the amount of choir/chorus you get now is very VERY much more.

Every gen I've made for the past few days have ALL had this problem.

[gleaming halo]

ok gave my choir one more shot…. Naaah.. https://app.suno.ai/song/52e9614a-a36e-4f3f-ba73-de6345b65453/

[silent karma]

Hi, is it normal that 225 words exceeds the lyrics limit?

[quasi rose]

character limit is 1200, you can extend it by continuing the song and adding the rest of the lyrics tho

[silent karma]

I try to keep writing but it doesn't work

[quasi rose]

character means individual letters / numbers / spaces, not words 😛

so depending on average word length it can be less than 225 words

[silent karma]

I hadn't counted all that, but even counting them I arrive at 724

[dense mountain]

Note that letters in some languages will each count as multiple letters if there are lots of letters in the language or if it's a rare language

[silent karma]

Ok but I speak French so it's the same character as in English 😔

[loud lotus]

Apart from the chorus: did you also notice that V3's response to (even dramatic) changes in the style for follow up parts is close to zero. With alpha and V2 it's possible to change the mood, tone, style between parts which allowed to compose quite interesting songs with contrast. Now it seems to just go on, which is especially annoying if the first half of part 1 is great, but the second became boring and dull. Continuing and cutting the boring second half more often than not doesn't work anymore. I just get more of the boring stuff I tried to cut.

[wicked dune]

You too into blues:) Love that!

[neat sail]

Yes V3 is more resistant to style shifts. Ballpark on an 8s clip it might shift only 10% time versus 60+ percent. But if you use a very short bridge clip (as short as 3s if necessary) you can still force a shift. Or a reset, like in #1221135437230379048 message

[broken cloak]

I'm getting sick of backing vocals on every song too.

[hearty garden]

can we report false positives on the censorship here

[broken cloak]

https://app.suno.ai/song/c3dee9b5-f049-490d-b13e-c522b683c026 https://app.suno.ai/song/10f68e96-c0f0-4a3f-9cd1-4b05d9e4adac it's impossible to get it to sing the first line.

[elder oyster]

I would like to share two observations about Suno version 3. Often when trying to reproduce a chorus of a song, I notice that version 3 seems to show less accuracy than version 2; she doesn't reproduce the chorus in the same way as the original, sometimes she seems to try to create a new version of the chorus. The second situation is that by changing at least one lyric from the original chorus, she tries to create a new chorus. This is quite inconvenient as I often find myself having to open my song editor just to change a word in the chorus. I would like to know what I can do to minimize this situation. Are there any tips you can offer?

[broken cloak]

https://app.suno.ai/song/9492ec38-748e-4897-ac09-e745a3396010 The ai hallucinated 3 times before saying the line

[ornate falcon]

Why does Chirp sing things written between [] ? like one of 2 clips i have all in sing
I give information on the mood of the sound, the instrument, except that he sings it, he does not take into account the []

[honest shoal]

would really love this, particularly with the band name detection. there's a million artist names that are just common words and styles, i was getting mine flagged for the word 'latino' apparently being an artist

[sinful silo]

Suno's created two lovely pieces but is refusing to sing?
https://app.suno.ai/song/b39e29af-ee7c-4fb3-b117-4e94b4886e72/
https://app.suno.ai/song/74dfdbae-501e-46ac-baed-22e1bc5d9e8a/

Tried again, it's singing now but still came up with some curious results, this is not late 1950s skiffle... was the model not trained on skiffle? Entertaining but inaccurate...
https://app.suno.ai/song/da454ff8-50e4-4ea4-96c6-a5af18e89eb9/
https://app.suno.ai/song/6fa86ad2-378b-43d8-b111-e62af9459c5a/

[eager tundra]

There is something seriously wrong with how Suno generates the continuity of a song. I wanted to stich a full song together. The style of music is exactly as the first 2 minutes and I simply added verse 3 with bridge, pre-chorus, and a chorus. Part 2 was totally different. Suno did not follow the lyrics and produced a different sound. No continuation at all.

[sinful silo]

There's a curious hiccup at 3:40 in this song. Not sure if I'll be able to correct it with an edit in an external program...
https://app.suno.ai/song/74e3d379-c9d1-41ed-9df8-6292d044960e/

[mortal tide]

Iif you reload the page you have to add the styles again, did that happen maybe?

[fallow kindle]

seems like v3 works better until certain time of day, i've been trying to create a 1980s hard rock song but it continuing creating that 2000s rock i hate. I was doing great looking 80s style songs until past midnight honduras time, now it t creates for me themes that are listened from 2010 onwards

[proven socket]

@neat sail: just a UI / accessibility thing, can the Suno tabs in the browser show what page they're on:

• "Explore | Suno"
• "Create | Suno"
• "Explore | Suno"
• "My Playlist | Suno"

It does do this already for songs, though not the other pages. I often have like 3-4 tabs up the top and I have to remember which is which. I'm sure this is something that would be added eventually, but just mentioning it in case!

[neat sail]

Probably more for a suggestions list (which I or somebody will probably pin eventually, at last for common/easy ones)

[hearty garden]

https://app.suno.ai/song/5f613fd7-c5b0-4b1e-b8ca-06561ba7fa75 only music came out, no sound

no voice*

[sinful silo]

It's almost like it's getting tired of our sh*t and needs a break once in a while 😜

[sinful silo]

I had the same issue yesterday (see my post here #🪲┃bugs-web message )

[willow pollen]

hmm ... I have a weird song that will NOT continue, despite stopping in the middle of a verse... all continuations are just silence

(I 'reported' it ... maybe the devs can find the reason and fix it in the future)

[fallow kindle]

yeah. I'll do it if i was this AI XD

still, this AI is awsome!! can't belive i can be able to relive the 80s

[low cypress]

@solar grove @carmine basin ^^

[eager quest]

cool

[fluid wigeon]

Seems to be completely bugged again today. It constantly ignores the prompt when it comes to genre. The vocals and lyrics work, but the style never fits what you want. It really only works at certain times of the day. Unfortunately, it's a bit disappointing and don't come with "unpredictable" or "not yet fully developed" for me. These are clearly bugs and not an "unpredictable" AI.

[craggy steeple]

The V3 test was good, but since they made it official the quality is bad, now most of the songs come out with double vocals. I feel like the quality has gone down a lot, it only works when it wants to.

[unique pendant]

What is going on with the Background Noise? It's killing so many otherwise good Songs Anyone have a Solution?

I'm wasting lots of credits in the hope to have a gen without the background noise

When vocals and guitars clash in later generations, it seems to be occuring more

[hexed onyx]

Yea V3 Alpha was incredible some of the stuff it put out. I've been really underwhelmed by V3 so far. Generation takes much longer, the tracks don't sound as good, and it's almost impossible to get a solo singer- some kind of weird choir effect is going on nearly all the time now.

V3 dropping and V3 Alpha vanishing feels like being rugpulled.

[unique pendant]

V2 is better in this 😦

[zinc talon]

I've stopped generating altogether. Can't handle if I get awesome music but the singing ruins it. Not going to buy more credits until the vocal problems are addressed. This should be their #1 priority, but who knows what they're working on...

[hexed onyx]

I don't understand a lot of what I've read about no chance of Suno returning V3 Alpha for those who want it. The model is clearly versioned, that's why we can select to use V2. When ChatGPT updates their model, you can still select to use older versions of the model for quite some time.

I've searched through Discord for user feedback about this and what I've found are:

• V3 Alpha seems to have cost more to operate.
• Users are willing to pay (or pay more) to use V3 Alpha over V3.

[craggy steeple]

I already think they are feeling the pressure from the media and since they already joined Copilot I doubt they will keep such a good version open to the general public.

We all agree that V3 is bad and does strange things like double voices, the singing is bad...which didn't happen before.

[hexed onyx]

I doubt it comes down to the V3 Alpha model being "too good", it had its fair share of problems too. My observation is for the stuff I'm generating (1000+ generations so far), V3 Alpha was superior. I could get a banger of a track in about 4 tries. With V3, it takes maybe 40.

If cost is the reason that V3 has degraded, and users are willing to pay for V3 Alpha to be restored, why not just paywall it like what OpenAI does with GPT-4

[mortal tide]

Hmm I was just logged out of Suno and I cant manage to login again ^.- Can some staff help me?

[stray urchin]

I cannot sign up either

[low cypress]

For me it is working fine. I log in with gmail.. What login/signup options are you guys using?

[mortal tide]

I have too many discord and gmail accounts, I cant seem to login to the correct one with my sub on it. I have the receipt for my sub but that email is not the correct account

I really need staff assistance, please DM me and I can explain the situation

[low cypress]

Try sending an email to [e-mail]

[mortal tide]

But they never reply :/

[low cypress]

They should.. they said they would have got all emails answered before the end of the month.. they ramped up

[mortal tide]

I sent an email, hoping for the best

[fossil pivot]

How many days did you wait?

I would realistically expect a reply maybe around 2 weeks.

[mortal tide]

Cant wait that long if credits expire >.<

[fossil pivot]

I see.

[mortal tide]

I sent them an email on the 25th that I havent gotten a reply too but maybe this one is faster 🙂

[low cypress]

on my invoice I see my correct email

[fossil pivot]

Yeah I don't think they'll reply that fast lol. ¯_(ツ)_/¯

[mortal tide]

I mightve signed up with discord account then used a diff email to pay with if I remember correctly

Anyway, email sent, fingers crossed

I have the receipt so shouldnt be impossible to guide me to how I logged in I hope 😅

[gleaming halo]

Hi, I have several subscriptions with over 10k of credits rolling over. I have (pre) paid for an annual subscription but the platform is currently not useable after the V3 update. (How) Can I pause these subscriptions and credits until the issue is fixed?

[kindred heart]

Hey new user of Suno it's so amazing might be needing some tips, i did my first tune https://app.suno.ai/song/a516dd97-3648-43d5-9210-8fb067ed0141 and i am quite happy with it. however when generating some parts i notice that the lyrics are not used they whey whey are written in the song, sometimes there is some overlap, sometimes a part of the sentence will be used twice, is there any tips for that or it's a work in progress ? Also i notice that my song is missing the last word, are those bugs or some techniques i don't know ?

[mortal tide]

Solved it, nvm!

I also realized I sent the other email from not the subscription email so makes sense they wouldnt reply to that anyway 😂 Too many accounts and email forwarding can be confusing

[stiff shuttle]

I get an error all the time when creating an account

[low cypress]

How are you trying to create an account? With your discord account, gmail or microsoft?

[stiff shuttle]

I have tried all three

[low cypress]

try re-verifying your discord please: https://support.discord.com/hc/en-us/articles/213219267-Resending-Verification-Email

[stiff shuttle]

I get this all the time

[low cypress]

Please read my message before yours

[stiff shuttle]

I'm going to try, but it also happens to me with Gmail and MS

[low cypress]

I understand your frustration. I am just trying to understand what is going wrong here.

[narrow radish]

I'm here to check on the same issue

[stiff shuttle]

I don't have any option to re-verify Discord email. I imagine because I already have it verified.

[narrow radish]

Can't create an account with discord, MS, or google

[low cypress]

I have notified the Devs.. @narrow radish @stiff shuttle @stray urchin
Hopefully they will fix it soon

[narrow radish]

Thank you @low cypress !

[stiff shuttle]

Ok, thanks a lot

[cyan lynx]

Could have something to do with a big streamer using the website, and an avalanche of users trying to get their new accounts

[narrow radish]

yes I was afraid of that as well - still, good to report it

[low cypress]

Where 👀 ?

[cyan lynx]

Dm'd you

[lunar perch]

I'm trying to make '80s freestyle music with the style prompt "'80s Latin Freestyle Dance" and I got one really good result but it refuses to let me do instrumental... and when I put in lyrics it mostly ignores them. It just starts singing gibberish that half sounds like random freestyle lyrics. I keep trying to continue the song to replace the unsolicited gibberish lyrics but it won't let me... most of the time it fails and refunds my credits. I haven't seen any other genere cause this behaviour yet. lol

https://app.suno.ai/song/22b78c20-44a1-4ba4-88fa-5aa41d7bd853/ this was set to instrumental. lol

Sucks because otherwise this sounds great. but it seems I can't do anything with it

Continuing the song before the lyrics start with more instrumental just makes more gibberish lyrics. The AI really goes bonkers from this genre it seems

I hope the next version gets better trained on '80s music styles. that's what I want to create most and Suno has a very hard time with any of it. But this particular behaviour is really bizarre

[sinful silo]

Usually Suno will start hallucinating if it doesn't have enough info to go by. I'd suggest using Custom mode and adding some more descriptors in the "Style of Music" prompt

[neon ivy]

I would try continuing before the hallucination and adding “lyrics in English, female vocalist” to your existing style

[lunar perch]

I am using custom mode, with instrumental turned on. But I'll try some more in the style prompt. but in general it seems I get the best results with as concise a description of style as possible. it seems to fixate on one keyword rather than combining them

[neon ivy]

My guess is “Latin” is priming it to expect lyrics in Spanish

[lunar perch]

Latin was the only thing I found that actually makes it sound like the correct genre. the lyrics don't sound like spanish to me, but rather half intelligible english. with some of it making sense

[unique pendant]

I tend to prefer V2 over V3 in continuations with vocals atm

[neon ivy]

Unfortunately it does hallucinate sometimes! It’s really frustrating when the “simlish” lyrics sound like a really good song you can’t quite understand 😆

[lunar perch]

Yeah it is frustrating but I understand. Anyway just wanted to report this. I assume more training in the future will fix this kind of thing. but it really does need more training on '80s music. it has a very strong tendency to make everything sound modern if you try to make anything dance or pop, so I was quite excited at first that it gave a good result musically from my prompt.

[craggy charm]

Me = "intro" "guitar riff intro" "start with guitar" etc. Suno = Nah, I'll start with the vocal, heck I might even do a verse of accapella just to piss you off

[rigid frost]

Interesting that Matt Wolfe is not getting that voice doubling using V3 https://youtu.be/e9WuM7h8_bw?t=426

[proud radish]

Same error here FYI

[low cypress]

Devs are looking into the problem @proud radish

[proud radish]

Yes, no worry, I just wanted to report. Thank you for your management 🙂

[sinful silo]

Here's a new one? In this one Suno is reading lyrics that are between square brackets..? https://app.suno.ai/song/9e2c8233-ffb2-4a76-a8db-27bbacdc1ddd/

[old sleet]

Hi guys) I'm running into a funny bug (or a feature?):) My google account username (the one I use to log in) has the word "woodlark" in it, and somehow Suno seems to be picking it up, so instead of getting thematic cover images relevant to the music I'm prompting, all my cover images today are the images of a Woodlark bird😅 is Suno supposed to pick up usernames like that?

[narrow radish]

seems to be fixed now!

[runic burrow]

Yep, it works

[sinful silo]

I don't think that's related to your username. Everyone gets that bird once in a while

[old sleet]

Oh ok then... still it's quite a coincidence, for it to portray the specific species that I have in my username lol Made me suspicious

[spark fiber]

Yesterday, I wrote that the distorted rhythm guitars often sound like they're in mono. I also noticed a noticeable degradation in the quality of these guitars with each subsequent continuation. The biggest drop in quality (and stereo effect) occurs after the guitar solo or instrumental interlude.

[sinful silo]

First time I had that I was wondering if I had a stroke 😜

[proud radish]

Yes! Thx!

[neon ivy]

It’s birds because the engine is called Chirp, the species is a coincidence

[analog ibex]

I've noticed the mutliple vocies in unison as well. I had a song where almost the entire think was three or four voices singing verses and choruses. For me, this started yesterday.

[silent karma]

Hello, at the moment almost at every start of music the AI ​​invents lyrics

[eager shadow]

@craggy steeple if you guys havent yet, make sure you put your votes in on the mini polls, the way the polls look, it looks like this version is far better than Alpha which most of us know is not the case in terms of quality. Structurally and coherency to lyrics it probably is better but I prefer the richer sound and more quality/dynamic range of vocals it had #🗳┃mini-polls message

[hexed onyx]

I voted in the poll but I seriously hope those aren't being used for user feedback. The sample is deeply flawed - there is no way to vet anyone who votes ever even used V3 Alpha.

A better way to get user feedback would be to poll users onsite and check the DB to see if the polled user generated content with both V3 and V3 Alpha.

[spark fiber]

How did you start the "Lyrics prompt" ?

i'm always adding [start] and [Instrumental interlude] before lyrics.
The only bug i've noticed, is if the lyrics are in my native language (Polish), then first word is read in English 🙂 It sounds quite amusing...

[eager shadow]

#💡┃suggestions-web message

it is being used plus another system they have apparently

[silent karma]

I use chatgpt for lyrics, and he often starts with [chorus]

[hexed onyx]

I don't think there's a dispute that some users like V3 more. I bet some users like V2 more. What's in dispute is that some users clearly liked V3 Alpha more, registered/paid based on that, and now are left with a product that they perceive is of lesser quality than what they originally paid for.

[spark fiber]

Try to add [start] first, before [chorus]

[silent karma]

Okay thanks you

[spark fiber]

And such a silly question just to make sure - are you using Custom Mode?

[silent karma]

Yes, I use custom mode

[eager shadow]

your preaching to the choir here, ive tried posting my comparisons and all during the first few days of V3 release. Im very frustrated with this whole thing, can barely get a song finished, the one song I did manage to finish I am not very excited about or happy with. but it took 3/4 of my sub to make, and im not spending anymore until it changes unfortunately

[hexed onyx]

I understand. I hope we get a resolution soon 😞

[spark fiber]

Great! Let us know if that solved your issue 🙂

[silent karma]

I just tried it and yes it seems to be good thank you

[red ridge]

Sign up unsuccessful due to failed bot validation. Please refresh the page to try again or reach out to support for more assistance.

I have been trying to sign up for a while but this error seems unavoidable for me both on mobile and desktop

[ashen venture]

Not really a bug, but I just got back a song (in one go) that was something like 3 minutes and 30 seconds long. At least the first time I played it. Then when I tried to replay it it was cropped back to 2 minutes. What a pity since it was actually doing pretty well. I wonder if someone accidently pushed the development version for a few moments and if we can hope for longer songs in the future or if that was a rare glitch.

[warm tundra]

Are you sure that Isn't a bug because I am having the same issue

[lavish vapor]

i've got the exact same thing. it doesnt let me do anything

[neat sail]

There is a bug that does that. There might be two different bugs. At one point sometimes you could get a double length clip (but not truly double length in the sense of context, it was two independent clips one after the other, basically an error that tried to restart but returned both clips I think.) and maybe some issue with streaming clips in particular.

[red ridge]

i used a vpn and put it on usa

that solved it for me

[ashen venture]

In that case the two clips were fitting perfectly together (not like the two alternatives you always get)

[lavish vapor]

thanks that worked

[neat sail]

Interesting. Could be continuing the first clip to make the second when this happens.

[fierce forge]

Same problem with me

[elder oyster]

I would like to share two observations about Suno version 3. Often when trying to reproduce a chorus of a song, I notice that version 3 seems to show less accuracy than version 2; she doesn't reproduce the chorus in the same way as the original, sometimes she seems to try to create a new version of the chorus. The second situation is that by changing at least one lyric from the original chorus, she tries to create a new chorus. This is quite inconvenient as I often find myself having to open my song editor just to change a word in the chorus. I would like to know what I can do to minimize this situation. Are there any tips you can offer?

[crystal needle]

I've noticed that if a continuation begins with lyrics that are present in the previous clip -- for instance, the same chorus -- v3 seems insistent on skipping it entirely. It just either hallucinates gibberish or skips right ahead to the next stanza. And it's very difficult -- sometimes impossible -- to make it stop, even if you add (for instance) some oohs and yeahs at the top to separate the repeated text from the beginning of the clip.

I wonder if some training to get it not to accidentally repeat phrases worked a little too well 😅

[craggy charm]

site seems to be down, standard continuations coming back credits refunded, all the time

[sinful silo]

I've experienced an issue with cut-outs a few times - where the audio seems to randomly dip inexplicably.
Example 1 at 3:40: https://app.suno.ai/song/74e3d379-c9d1-41ed-9df8-6292d044960e/
Example 2 at 0:07: https://app.suno.ai/song/103448be-1e61-4004-8dac-16a8705dabcb/

[chilly cedar]

I have a theory, if you continue half way through a chorus and then put the full chorus in the lyrics, it'll complete the chorus and then move on. So if you continue from a chorus for example and then want to repeat the chorus, it reads that you're trying to finish the chorus and thinks it's done the job and moves on. I haven't tested it but I reckon if you put the chorus twice in the lyrics on the continue, it will do it once and move on.

[eager tundra]

Hallucinations are all over the place. Found a few really nice first 2 minutes of the beats i was expecting. But doing part 2 or part 3 was extremely frustrating because the algorithm is totally different and the sound was really off the first 2 minutes. Suno even added it's own lyrics even if I typed in my own lyrics.

[eager tundra]

It doesn't help make me go pro or premium if even the 50 free credits a day just gets wasted for clips that do not continue from the previous track

[cursive notch]

Fuzzy white noise creeping into V3 gens is ruining a lot of my songs. I've given up on songs with vocals for now with that and the choir singing. First minute is V3 instrumental, you can hear the 'fuzziness' increasing. From 1 minute on is a V2 gen done from the same prompt today. No editing.

This fuzzy noise was not in V3 alpha.

[spark mesa]

I’m not getting the chorus to repeat the same melody even when I download entire song and continue

[willow pollen]

not really a solution, but generating pop or hip hop songs seems to work quite well in V3 and to generate less choir and noise than when you try other genre. Rock,metal or even Jazz and their respective subgenre (unless we're talking the High School Musical type) are pretty much pointless in V3, but I had some great result with Trip Hop

[pallid pumice]

I feel like V3 has started to degrade.

[gleaming mango]

I was genuinely emotional listening to generations from alpha, nostalgic for an ai app since zombified. I compounded my anguish using those prompts with V3 - results are a pathetic travesty. I may be compelled to seek a refund on my annual subscription, return to messing about with granular synths and DAWs.

[lavish herald]

Having not been part of the alpha, I'm left wondering how V3A compared to the VDay generator. I mostly just want to be able to use the VDay model, even if it did a lot of improv that others might not want.

[placid notch]

V3 is still alpha ? Because the begenning of my lyrics are always skip when I use the Continue button.... I waste my credit because of this bug. Why they permit to use the continue button if there is so many bugs... We pay for credits and waste them.

I don't understand, there is no mention of "alpha beta" or somthing. Just "V3" so I guess it's OK version, but I'm wondering... Maybe I should write something in the begenning of my lyrics to force the IA to take care of them ? [Begin] or something ? Someone have a tips ?

[mortal tide]

V3 and alpha is not the same :((

[placid notch]

So V3 is a definitive praticable version ?

[mortal tide]

Supposedly

It works well for some but not for others

[placid notch]

So why I still have the same bug like during the v3 alpha version

with continue button especially

[mortal tide]

Maybe your meta tags or style promts intefere somehow?

[placid notch]

Maybe. But my continue are so random. Sometime it starts from the begenning of my lyrics, and it's OK, but often it cut the 2-3 first lines i don't know why. It's a waste of credit

V2 started lyrics from the begenning :/

I tried "[Reset, begin lyrics]" but the it doens't work :/

[mortal tide]

Yeah same here but I had to change my meta tags and it worked so I think it was me breaking it

[placid notch]

meta tags ?

[mortal tide]

This didnt work at all:

[Verse 1]
Blabla

[Instrumental]

[Verse 2]
Blabla

it just skipped directly to verse 2

When I removed the instrumental tag it started from verse 1

[placid notch]

I don't use instrumental tag, just [Verse] and [Chorus]

[neat sail]

I am planning on investigating potential bugs with continues, beyond just the AI model being quirky and 'hallucinating' there is possibly a bug making problems more likely. It's tricky since it's still quite random. This may be something unrelated to the AI itself and have to do with continue times or something.

[mortal tide]

Do you do [Verse 1], [Verse 2] or just [Verse][Verse]?

[placid notch]

Verse VErse

it's a mistake ?

[mortal tide]

Try specifying them

[placid notch]

mmh okay got it

gonna try this

[mortal tide]

Fingers crossed

[neat sail]

It does help to label the verses as 1, 2, 3, sometimes yeah. So it knows when things are new or not, or at least, there's a clue.

[placid notch]

Doesn't work :/

[mortal tide]

Can you screenshot your prompt window?

[placid notch]

AI skips the 3-4 first lines of my verse

[mortal tide]

SS this:

Or copy paste works too I guess

[placid notch]

I send you in MP

[mortal tide]

Cool

Just know Im neither staff or guide just amateur trying to help

[placid notch]

Maybe my lyrics are too long, but I tested with shorter lyrics in the past and still the same bug :/

but thanks

By the way

Have you noticed that since about 2-3 weeks, Suno often uses Chorale singing ?

[mortal tide]

Yeah it is a known issue in V3

[placid notch]

okay i'm not crazy

😄

[mortal tide]

Well you are French but in this instance no 😂 😉 ❤️

[placid notch]

ahah yeah 😄

you're from ?

[mortal tide]

Sweden 😛

[placid notch]

Clark is sweden 😄

(Clark Olofsson)

[brazen kestrel]

When I sign in to my account, a random song is generated in the "create" section, but it's just wasting my credits. I didn't even write the prompt; I just signed in, and it deducted my credit and generated a song.

[zinc talon]

v3 went from alpha to release, no beta version... but I feel like the alpha was the beta and we've now got the alpha....

[willow pollen]

well .. the "release" version of V3 has definitely clearer vocals .. for the 5 seconds it takes for the surprise glee tactical choir to make an appearance and spoil the fun

[rigid lion]

Is it already known that there are always 2 songs generated when creating?

[willow pollen]

that's on purpose

[rigid lion]

So it's more like 5 songs daily on the free plan and 250 songs monthly for the pro plan or does the pro plan not do that?

[willow pollen]

it's globally the case, no matter the plan.

[rigid lion]

Ah. Good to know, thanks for the info

[willow pollen]

honestly, it's nice to have two different takes, as it gives you the option to chose the best one (or scrap both)

It's really not that much different than MJ's grid

[devout sonnet]

i have been stuck trying to load the create part of the site all day 😦 anyone got the same problem ?

[neat sail]

Everything seems to be working overall. Maybe check your login/account and logout and back in again

[devout sonnet]

thankyou but doesn't change anything 😦

[neat sail]

What country are you from? There were some users earlier who had issues and reported they had to use a VPN.

[devout sonnet]

ireland

[neat sail]

That should be fine. Strange. Can you try logging in with Incognito mode window so it will force a refresh?

[devout sonnet]

i will try vpning it and incognitoing, really appreciate the help

[neat sail]

You can contact [e-mail] and they can check if there is something wrong on the account itself.

[devout sonnet]

thankyou again! i think i might have to. Have a great day

[placid notch]

I think i'm making a pause of suno until the chorale singer song is fixed :/

I waste my credits so much because of that

[tranquil hollow]

Can anyone help me or guide me to the proper discussion? I have created one song, via browser, and now I want to implement diferent lyrics but to maintain the same singer and style of production. How do I do that? Eventually I would like to create 4, 5 songs in the same style and voice, just different lyrics

[willow pollen]

choose your original clip, select 'continue from this song', set the time to 0:01, keep the style tag but change the lyrics ... should work

[low cypress]

Try to do that in the right channel? This is for bugs. I recommend #📚┃suno-school

[willow pollen]

also : what 3daizy said

[tranquil hollow]

Yes, I've tried that, seems like that's the only solution, right?

[low cypress]

Please continue this conversation in #📚┃suno-school @tranquil hollow

[cobalt hamlet]

Hi there! I have this text when i try to create a Remix, anyone kwow what kind of problem coul be?

[gleaming mango]

@placid notch I've cancelled sub and asked for a refund, super disappointed in the drop from V3a to V3, almost impossible to get anything satisfactory from it😢

[placid notch]

I think they can fix all this bug, there is so many potential with suno. Too bad we waste our credit for something not optimal

I keep my eyes open for the next update

[gleaming mango]

@cobalt hamlet the style prompt field allows more characters than it can handle - remove any superfluous spaces, usually fixes it for me.

[cobalt hamlet]

Thanks you, i'll try 🙂

[keen holly]

After reading a bunch of backlog, i'm still not sure: Is work being done to address the issue with v3 always doing choir instead of a single voice?
Also, is there a status page that tells what the devs are focusing on, and what we users might expect next?

[low cypress]

They cannot change the current model if that is what you are expecting, or do a rollback. I think this message should tell the stance of the Devs in this: #💡┃suggestions-web message

[gleaming halo]

Of course they can... They just choose not to do it. and ignore the long list of users complaining, not being able to use Suno or asking for a refund....

[low cypress]

Maybe, maybe not.. But I cannot help it.. Just telling you what Devs say... 🤷‍♀️

[gleaming halo]

Nobody is blaming you @low cypress 🙂

[grand idol]

When I'm in Library view, going through clips, as soon as I move a clip to trash, the "Page 1" next button gets disabled. I have to refresh the browser in order to go to page 2.

[tacit geyser]

My opinion: I think they either retrained V3 or tried to fine tune V3 alpha. I have a hunch the weights got messed up. Is catastrophic forgetting a thing for audio GenAI? I assume so. Not sure if they have a backup of V3 alpha and choose not to rollback or they don't have one. But I assume the only way to make it better is retraining .. which takes a lot of time and resources. It's probably gonna take a while until we have a fix. If it's like text AI then if we can't convince the model not to sing choirs the devs can't do it either without retraining.

[tepid mesa]

does this come up?
https://app.suno.ai/song/a75e13c9-26cf-4ff2-a063-48a4a49c3609

[keen holly]

@low cypress thanks. i cannot really communicate much meaning by simply pressing like/dislike buttons. so I hope the "choir instead of single voice" thing is on the drawing board.

[tepid mesa]

a number of creations have infinite spinner when i try to play

[dawn radish]

just coming to say this.

[tepid mesa]

is it a known thing? does it resolve on its own?

[dawn radish]

dont know i come here to see if it was just me. probably be fixed soon. just had a spare bit of time on my hands.

[dull meadow]

it's not only slower, i have like 4 or 5 songs just disapear while creating them

the credits are taken anyway -.-

[low cypress]

Just shoot a message to [e-mail] if you are on paid plan. They will sort it out for you.

[dull meadow]

thanks i'll do this

[low cypress]

@solar grove ^^

@neat sail

[solar grove]

Yes.

[lean estuary]

written in a hurry, why not "disabled" but "disable"

[solar grove]

Maybe because they were in a hurry and wanted to try to get things worked out quickly?

Just speculating.

No inside information or insight.

[keen holly]

a part of me is hoping that they're fixing the choir problem, but they're probably just upscaling the server capacity. i'm sure this stuff is using a lot of server power! 😅

[limber prawn]

Admin - we believe in you. You will make Suno even better. And tomorrow all the problems will go away

[craggy charm]

Those into choral music must be creaming their jeans.

[craggy charm]

Remix has never worked, no one knows what it does.

[craggy charm]

now that's a new one

[low cypress]

Remix is just reuse prompt. It pastes a copy of your lyrics and style of music so you can try another song with these settings @cobalt hamlet

[iron jasper]

it should be renamed to regenerate

[brazen kestrel]

Yeah

[ripe gale]

Unable to generate text with my daughter's name "Adélka" because it is evaluated as the singer "Adel". You can't ban all personal names 😦

[low cypress]

You could spell it different.. as a work around

[ripe gale]

That's not possible, her official name is Adéla. Diminutive is Adélka. Both are evaluated as Adel 😦

[low cypress]

Adella doesn't sound the same? Or Adeella?

[ripe gale]

In Czech, the difference is audible. But even if it wasn't, banning people's names because almost everyone has a variant like some singer or singer is nonsense 😦