📚┃suno-school

[rigid umbra]

the timing problems seem to be fixed, at least

omg I got one. it duplicated a word afterward but it'll work 😅

[polar spade]

The only means of using replace is if you can choose a segment where there is lots of music before and especially after the lines replacing. It still bleeds, and you can hear the splice, so that's bad, and remaster or cover doesn't fix the splice -- yadda yadda, but you get a finished song that only has some skippy music.

[polar spade]

Cool. I'll test more. A couple times 4 days ago it seemed good. Then I was getting 30 seconds to 2 minutes replace segments and bleed over, again. It's all part of the process. Not whining just observing.

[polar spade]

I've got just one word to change in a few songs. I now need to have a "needs fixing" Playlist.

[rigid umbra]

I'm coming to realize that I should only try to do one AI art project per month, because it always takes a month's worth of credits to get it to do what I want

[cursive island]

How do i make it hold a word and then chunk and split it into the beat ? Im failing

[polar spade]

As a guess.... repeating letters can extend or hold a work...

Helllooooo

Dash lines or just repeating can stutter a word

Hellooo-o-o-o-o oh oh oh

making it fit the beat is another thing, altogether

[cursive island]

The worst part is i have one version where it does it instinctively on its self with with word control right before the bridge (o-o-o) and it fades out and back in, but i fail to replicate what it did on its own very frustrating @polar spade

[polar spade]

Right, innit? Suno can come up with some great stuff. Just can't splice it in where we need it. Just try your best to nudge in it the right direction. Good luck.

[radiant mountain]

did i miss something? my credits were reset?? i had like 2.8k now its 300?

and why do they take 50 credits off after you sub lol

[cursive island]

how can I make it so suno takes 2 voices like one crescendo "STAYYYY" while a 2nd sings "run,run,run,run"? without it taking the first, and then the second as a new bar

[distant burrow]

Dear Suno, i hope you will repair this replace section, its totally destroyed, impossible to use. I hope I will not have to ask it to Santa. Please repair this broken tool. Thanks. 💖 🫂

[graceful condor]

I keep getting error code 422

[polar spade]

That's a tough one. The only thing I've found is to use styles that gives Suno. Hint you want "medley" singing, or one over top of the other. Sunonwill doe this spontaneously with 4.0, but also I think it's the style that hints that's what you want.

This is a strange style but "isicathamiya" is an African call and response genre. Only if you have , like 12 styles already, you might be able to stick this on the end and it helps to have medley voices.

Then you'd have to put metatags which indicate two voices.

But I'm becoming more certain that the style of song affects the medley. I get something like if the song is romantic , emotional or ballad. So, musicals, broadway, prog rock, or others that can bring it out.

https://suno.com/song/655e60f0-6e82-480e-a22c-f6ff2ee05b07

https://suno.com/song/5171ccce-7bfe-4651-82ea-1d3e1482851b

[timber stream]

Okay, so I think I have something similar.
I am trying to make a song with 2 singers, male German, female japanese. I already put in japanese lyrics for the second Verse. Suno still very very very often uses a single male voice.

Tried putting (German male) or (Japanese female) beside the Verse tags but usually no hit.

Is there kind of a cheat sheet to tell Suno to use different voices at a specific verse or language input??

[spiral bane]

pro tip if you want to have the extend feature work correctly on v4 you need to set the time manually to 1 second back from the end otherwise it doesnt seamlessly attach when you get the whole song

[lavish bane]

Maybe try provide 1 verse with tags for male voice and then extends it with tag and lyrics for female. But this might not work too

or work with like 10-20 tries

[halcyon coral]

i'm trying the ReMi function on 'Write with suno' and it seems the lyrics are cut short. Is it only for me or it's happening to everyone?

[lavish bane]

sometimes I spent like 20 tries to receive something I want or maybe cool

[timber stream]

I tried stuff like [Verse 1, German male/German John] and [Verse 2, Japanese Female/Japanese Jane] (either didn't work). I tried with John and Jane since it kinda worked better with 3.5 but V4 seems stubborn

[lavish bane]

It looks like this gets even worst with V4((

[timber stream]

Dunno why it's so hard for Suno to understand this...

[lilac garden]

Hi when u have a song that is slow in the first half and u want ir to progressively pickup on the second half reaching a climactic moment what tags do u use?

[timber stream]

I managed to do that with [Bridge, Pick up pace] but it's, like with many other things, not totally reliable

[lavish bane]

you can change tags with extension but this worked not very well too. add some tags for speedup or write bpm

[timber stream]

[Beat Drop] can also increase the song pacing

[lavish bane]

I feel this more depends of the song start if AI has many options learned to know how to evolve it - it can follow instructions better and choose this learned path, otherwise its just throw the same extension like 20 times no matter what you write.

or better say it has knowledge to be able follow your instruction in song context

[little granite]

has any one been able to influence it by using things like this: (The bassline grows as a synthetic melody weaves in, building the atmosphere.) with out it becoming lyrics

[timber stream]

Oh wow I managed to do one successful male/female change

[Verse 1]
[MALE START]
{...}
[MALE END]

[Verse 2]
[FEMALE START]
{...}
[FEMALE END]

--

probably coincidence

[lavish bane]

in [] because () is like second voice

[warped furnace]

Oh, wow. I will try it out. 🥳

[timber stream]

Ok second and third gen was pretty much male only again sigh

[timber stream]

Try it! Surprisingly the next 3 gens were on point again, wtf

[warped furnace]

Sounds promising. Thanks for sharing. 🙂

[timber stream]

Ur welcome! Let me know how it works out for you!

[stoic citrus]

One question is how can I write it so that he really only sings the hook... for example if I enter this he sings instrumentally as a word and so on... except for the hook, Suno does it well. How do I write it so that he doesn't sing along in brackets? My Prompt in lyrics; [Intro]
Instrumental only.
(Tribal beats, glitchy pads, pulsing synth bass)

[Hook]
"Feel the pulse... through the air.
Let it hit... take you there."

[timber stream]

Wow, I even mapped a whole song this way and even split a verse in "first half male, second half female" and put [DUET START/END] to the chorus and it worked like... 90%!

And () parts in a male verse are sung by the female here and vice versa.
This right now impressed me

[lavish bane]

More ideas try also child voice ect. 😆

[little granite]

trying to get a robotic voice for a part but so far no luck

[graceful condor]

Is there a suno people that can help me out with this “422” error? I just renewed my membership and first thing it get is this error

[polar spade]

I've noticed when using extend it will automatically click one second off (of the end of that output) anyway. This may be a glitch as I don't think it's every time. But it's that way occasionally.

[spiral bane]

ive noticed it doesnt 90% of the time with v4 so better safe than sorry tbh

[outer cosmos]

suno said bring back 3.5 for extend. but not working for me. when i used uploaded audio and click on extend its automatically change v4 and not allowed to me change. any solution? @charred merlin sir pls look here if u know solution

[halcyon coral]

the replace function is effed up

[naive quail]

Upload extensions are still locked at v4. regular extensions will work with earlier versions.

[outer cosmos]

ohhh so disappointed

[charred merlin]

https://cdn.discordapp.com/attachments/1102676217687195679/1309480845475713084/20241122-1115-24.6147628.mp4?ex=67430e12&is=6741bc92&hm=6abf92075a05d850206b60a5fa1ce1210ac7206d6d38df03e1250fbee3732776&
🙏 If you crop the entirety of the song, you're able to extend with whichever model you like

[tight mango]

Hey everyone, was the v3.5 relase as bugged as the v4 is ?

Is it still in beta, isnt ?

[outer cosmos]

wow. its work. thnx sir

[austere dragon]

This is really weird. I am trying to redo a song with the same persona I used in v3.5. It has absolutely no connection to the v3.5 version other than the lyrics and the persona. Also, it's not a cover. The reason I am redoing it is because there was a part that was supposed to sound adlibbed where the singer calls out the guitar player before a solo. I kept it in in the v3.5, but eliminated it from the lyrics. Last night I tried to remaster the v3.5 version and that adlibbed line just became gibberish. So I thought, no big deal. It's a Jerry Reed style trucker country song. Again other than the lyrics and the persona, there's nothing connecting the two versions:

v3.5: https://suno.com/song/16bfdd0d-5149-4c4b-adf2-68057e1f6940
v4: https://suno.com/song/694f1b12-1ce2-4c01-bc4b-4d587d09d1eb

What sort of credit wasting bug is this?

[little granite]

only thing that comes to mind is a hallucination

and it say v4 is (beta)

[charred merlin]

Usually, I'd point out the fact that there's an explicit disclaimer when trying to extend from a v3.5 track with v4 (or vice versa) -- since personas are based on existing clips, it might be worth remastering that original clip to v4 quality & making a persona based off that clip.

However, I know that Jack thinks that all of my suggestions place the blame back onto the user, so I'll just say "I hope they improve things so that future versions of the model will satisfy everyone" 🙏

[robust ermine]

hey, does anyone know how exactly the price for an upgrade from Pro to Premier is calculated mid-month, and how many credits are given?

(assuming we're talking Stripe, not App Store subscription)

[charred merlin]

Not offhand, and it doesn't seem to be mentioned in #1294421750397796435 -- while another user might have an idea, it couldn't hurt to reach out to [e-mail] if you were okay with waiting until Monday or so for an official answer

[tribal cloak]

I just bought a yearly licence, now each song I create has the lyrics "please go to SUNO.COM and download the iOS App ...etc." is the website broken? can we no longer use the website to generate music on SUNO?

[robust ermine]

That’s a special perk of getting a yearly license lol.

Jk sorry to hear that

[charred merlin]

@tribal cloak ☝️
This should only happen if you're accessing Suno outside of suno.com or the official iOS app -- if you're still experiencing this issue despite accessing the service properly, please fill this form out so they can investigate 🙏

[languid thorn]

why can't I finish songs with this 4?? I extend when I miss the ending, and with [end] it never ends, I mean never. Always 2 minute pieces of music with no breaks, it goes on forever.

[paper comet]

does anyone else have a problem trying to go to Suno, my wifi is working perfectly, but only Suno refuses to load

[naive quail]

It still loads for me

[queen thicket]

"Couldn't generate that.
This song violates Suno's content moderation policy"

Why?

Its a song i already did

Want to do a cover

[naive quail]

Oooh. Error messages are back? @queen thicket they're likely constantly updating the filters, so it may be something they added since the OG generation.

[queen thicket]

Wtf

[little granite]

@paper comet open powershell and try
ipconfig /flushdns

[naive quail]

It's a pain in the arse and a credit sink, but try pasting the song into the lyrics section a bit at time to narrow down it down to the problem text.

[queen thicket]

Ok! Cost alot more credits

[naive quail]

it may also be something in the style field

[queen thicket]

It's just a regular song without any bad lyrics. Hope they fix it

[hollow hare]

Can you DM me the link to the previous song or your lyrics so I can investigate the matter?
Alternatively, you can send an email to [e-mail] (though I recommend deleting portions of your lyrics to find the offending cause)

[naive quail]

it doesn't have to be something "bad". It can be a well known song name, artist name, producer tag, or even well known hooks. For example: it's me again - put that anywhere in a song and the generation will fail. In that case, it's a producer tag.

And to answer my own question, generation error messages ARE back!

[charred merlin]

There's also people that insert those non-functional metatag blocks at the beginning of the lyrics

[bpm: 140]
[other stuff that observably gets ignored]
[influences: {insert copyrighted names here}]

[little granite]

sometimes you need to refresh the page

[charred merlin]

interesting.

Previously I was using nsync as a test to get the style prompt to fail, and it went through this time (just piano tracks)

[naive quail]

It's probably not so much the phrases that they're worried about but the fact that they might trigger Suno into outputting the actual producer tag from the training data.

[halcyon coral]

i typed snoop dogg in the tag for a verse and the voice sounds just like his. is it a feature or a coincidence?

[charred merlin]

lmao -- DMing you -- but you'll need to unblock me first

[naive quail]

That feels dangerous, tho...

[paper comet]

thanks it worked

[naive quail]

'was probably the man himself, recording that section for you. 😜

[charred merlin]

Dataset regurgitation is something that all AI companies wrestle with -- always a game of whack-a-mole

[lean idol]

I have an issue: When i download a song the file is an archive named example.com not a wav file

[unkempt spade]

Noticing that on the remasters in songs with prominent violin lead lines that it’s struggling to keep up with them. Any idea what might be causing that? I can provide an example if needed

[graceful condor]

can someone help me create some rap songs pls

willing to pay you for your time to save me time from guessing the best styles and lyric prompts:)

[azure mountain]

Has anyone been able to get Suno to emit specific notes (like if I want very fine-grained control to make a specific intro theme?)

[naive quail]

Lots of users are convinced they have a solution for that but I have yet to see anything that's repeatable.

[azure mountain]

Hmm, "I don't love you anymore" seems like a pretty common thing that would be in songs rather than a specific producer tag. That seems overly restrictive

[naive quail]

You can extend from or cover an upload with he melody that you want, tho.

[azure mountain]

Ah so maybe if I have a bad humming of the melody that would work?

[naive quail]

Yea, tho humming may trigger Suno's voice detection on the upload. Which may prevent you from setting the track public.

Worth a try

Suno can turn bad humming and singing into glorious tracks 😄

[azure mountain]

Yeah it does trigger voice detection

[naive quail]

I've had guitars trigger the voice detection

[charred merlin]

tbh -- it only means that you can't show it on your suno profile and it can't be searched for, but you can still add it to public playlists, and the link still works to be shared externally.

[charred merlin]

To be honest, your Frampton impression sort of blurs the line.

[azure mountain]

Huh, attempting to extend it caused Suno to hang in generating lol

[naive quail]

Hrrmph...

[azure mountain]

Maybe the GPUs are all busy now?

[charred merlin]

Don't be like that -- I know you feel like we do.

[naive quail]

...But now that you mention it. We need a talkbox upload 😄

Tho I used the last of the tubing as hookah hose for... a... uh, flower vase.

[azure mountain]

Hmm, the humming thing didn't work. Suno just decided to make its own theme

[naive quail]

It can be pretty random. Try shifting around the extensiojn point. If there's silence at the end of the clip and you extend from that, Suno's more likely to ignore it and occasionally use it as a reference.

If the extension cuts in during the humming, Suno may try to continue it, but then you don't get a clean separation from the actual humming.

[charred merlin]

Extend will try to hook onto existing patterns to continue -- but ultimately it's trying to turn whatever it's been given into something that it'll be "mathematically confident" would sound good, even if that means dropping parts that we want. (inb4 but Suno doesn't know what sounds good -- that's why they're asking thumbs up/down for feedback 🙏)

[azure mountain]

Yeah I suspect maybe making some kind of MIDI synthesizer and using the output of that instead of a recording of my voice might work better (especially if I can get the outputs to sound similar to a Suno V4 output somehow)

[charred merlin]

https://suno.com/song/d66e8316-7500-4f83-b8f5-465305c57b6c I just recorded a 30s clip of humming, and extended from it -- it didn't continue the exact pattern, but it helped set the tone and pitch in this instance -- it can take a bit of experimentation for sure.

(I clipped out most of me singing, you can tell when suno tries to continue from my voice)

[azure mountain]

Yeah that's basically similar to what I got,but I wanted it to replicate the specific pattern

[charred merlin]

Right now I'm trying to cover that humming track to see if it gives better results 🙂

[azure mountain]

oh that's a good idea

[charred merlin]

It would at least be "cleaner data" for Suno to work with -- I wouldn't be surprised if that gave a better hit-to-miss ratio, while I'd still want to position that you should expect suno to be interpretive a lot with covers, as it's still beta

[fresh oxide]

If Suno generated the vocals and the instruments separately, we wouldn't need to separate the vocals and music to edit them in post. This would solve a lot of problems since their stems are horrible. Now they just need to generate in hi-fi.

[charred merlin]

https://suno.com/song/9d21435f-8fd5-4310-81aa-03768533dcab

and there we go -- got it in 6, which isn't bad considering how specific of a result I was hoping for, and how hastily assembled the source track was

[azure mountain]

Didn't get what I wanted, but got a cool theme for my upcoming trip to California lol: https://suno.com/song/e1504763-0d85-4baf-9cdf-9d610cafca2d

[little granite]

what do you think is better for generation

  [Male Voice]``` or
```[Intro}
  Male Voice:```

[naive quail]

The humming is a banger, tho! Bah bah bahdada bah bah!!

[azure mountain]

If instead of humming I actually talk, I can get it to sing in my voice lol

[charred merlin]

Not really a matter of "better", as Suno is less about using "the correct prompts" since the context that they're used in varies so wildly from song to song.

We can command Suno all that we want, but it's ultimately going to vibe with whatever it likes from our requests.

(not saying you need to do things my way, just sharing a perspective 🙏)

Personally, I like using fewer style prompts, perhaps with a thematic decoration or two, indie math, ska, tired, deliberate -- and have a relatively good idea of what I can expect from it.

When you start throwing 20 different requests (and start using metatags to load the style requests even more), suno can get overwhelmed and disregard all of it, if it that's the way that particular cookie crumbles.

Some people work around this by stacking specific prompts multiple times to make sure a prompt comes through, indie math, ska, ska, ska, ska, ska, tired, deliberate -- personally, I've found that DOES work for getting it to sound very strongly like whatever it interprets "ska" to be, at the cost of almost everything else.

[azure mountain]

Yep I've noticed that's an even more severe issue with v4 than 3.5

[charred merlin]

🤷‍♂️ I'd posit that it's just different, as any new model would be. Prompt methods that worked for v3 didn't work the same when v3.5 came out, and that was using the same base model -- v3x > v4 is a larger change by comparison.

[azure mountain]

Yep I have significantly changed how I prompt V4 compared to 3.5

even my lyrical structure is different

[charred merlin]

@oblique fable "how to use extends" 👉 #1270928841614823546 message

[azure mountain]

I found that certain vocal styles can sort of "hint" at the instrumental style/theme it chooses in V4

[little granite]

i guess i need to give up my millions of prompts

[azure mountain]

3.5 didn't seem to vary the theme of the song as much by the lyrics, whereas 4.0 puts more of an emphasis on the lyrics themselves

[charred merlin]

#1309133718656385155 message Here's an example I'd shared about changing the prompting methods

Previously a friend of mine was using the prompt guitar, piano, violins, turkish melodies, symphonic, eastern symphonic, instrumental, epic symphonic, choir to get Turkish music, but it was coming out too much like generic rock on v4 -- so I tried turkish, eastern epic, guitar, piano, violins and was blown away.

[azure mountain]

Yep that kind of matches what I've been seeing too

Also here's the two best gens I've gotten with V4 so far:
Best gen: https://suno.com/song/38409153-15a1-4fdb-a8fa-6ba09ccba939
Second best (has distortion towards the end): https://suno.com/song/745b63d6-fe9c-4b1e-9aca-c32b5dc79ace

[sweet wigeon]

Amazing thing guys. Last days I have been having problems with duet of man and woman - bass and mezo-soprano voices. But I have managed to do it after many tries. I have removed repeat of the verses because the [: :] brackets seems confusing to Suno. But I have added specifications on bass strings and strings in the range of mezo-sopranos voice. I'll be happy if you will try this - to add scale definition for the instruments before the singer starts to perform lyrics instructions. Note, the tuning on begin of the verse from mezo-sopranos to bass and from bass to mezo-sopranos is still problem. I would like to try tenor instead bass later. https://suno.com/song/df293d7d-5041-4b1c-9f32-2a06de8b9a16

[azure mountain]

In both cases, putting less emphasis on the style helps a lot

[charred merlin]

https://suno.com/song/65a2e520-e54a-4def-be67-d483a08a49eb even back in v3, without using style prompts, it would will infer style from the lyrics. (also, the shimmer is SO obvious once you're attuned to it, but it was just "noise" before -- but now there's definition to the noise becase the audio quality has been improved. Talk about a double edged sword.

[azure mountain]

Yeah in v3 it just sounded like a "hiss" because of the lower quality

Also sometimes remastering in V4 removes distortions that existed in 3.5, and that actually can make it sound worse in some cases

[fresh girder]

Is anyone familiar with “Brazilian funk” and had successful results? I can’t get it to gen anything remotely close. Might be too underground of a genre for Suno. All it does is make funk tracks but the real genre is nothing like your usual “funk”.

[azure mountain]

Example where removing distortions made it worse:
Remastered version (V4): https://suno.com/song/485f49a2-d18c-4cb5-bb90-054f6ef72b74
Original version with distortions in voice (specifically in the "powerful and soaring" part, which was intended as a metatag, but Suno interpreted it as vocals). Because of the distortion, this misinterpretation actually added to the song: https://suno.com/song/26cbf911-86dd-4fd6-82ea-1770b0319ae2

In V4 because this is clearer, it doesn't sound as good anymore.

The less-clear distortions in 3.5 actually added to the value of the songs once you got used to them

[charred merlin]

If you already love the song for whatever reason -- there's a better chance that Remaster will remove something that you like, instead of leaving all the things you like & just improving them.

I've found best use with it by breathing life back into songs that were poor quality, and were going to be discarded for one reason or another -- still lots of use eithe way, but just where I've found the most success

[azure mountain]

Yep with 3.5 though most of my bad gens were the result of bad prompting (wasn't sure how to prompt it right), rather than purely bad RNG and a remaster won't fix that

With 4.0 it does seem like I get hit by bad RNG a lot more often though

4.0 does seem like it's slowly improving with time, but maybe that's just me getting better at prompting it

[charred merlin]

Both seem very likely 🙂

[tardy coral]

Has anyone had any success removing drums when generating or extending? From genres that usually have drums like indie-rock, pop etc… I’ve tried everything but it’s pot luck if it works and rarely does .. “Beatless, no drums etc…”

[charred merlin]

If youre on a paid membership, use the "exclude styles" box, as suno doesn't recognize "negative prompts" in the Style of Music section.

"no drums" is the same as saying "drums, please"

[tardy coral]

Yeh that hasn’t worked either.

[azure mountain]

Reinforcing the style you want may work better than telling it what not to do

[tardy coral]

Seems the style is only for genres and doesn’t respond to terms like “Beatless” etc

[azure mountain]

Yeah in my latest song I went with no style (empty style in style box), and prompted within the lyrics instead

[charred merlin]

Nope, you can prompt instruments, thematic descriptors -- anything. Suno just doesn't have to follow any of them as if they're commands

I've been scatterbrained and falling behind on experiments, but I could test to see if "terminator extensions" still work in v4

[azure mountain]

As in the short-extend trick? I tried that and it didn't work (seems to incorporate more of the context from previous extends even when doing that)

Adding longer instrumental interludes seems to still work though

[tardy coral]

I’ve put “drums” in the “exclude style” and it doesn’t respond

[tardy coral]

“Terminator extension”?

[azure mountain]

AI doesn't parse your prompt like a typical program and follow the instructions you give it, it's a statistical system that converts your prompts to a bunch of floating points, runs them through a bunch of layers that are configured for particular statistical results, then you get output based on how it was trained.

[charred merlin]

I usually go a little wild in the negative prompting section & try to describe related concepts to further drive the point home -- usually I dislike stacking prompts because it impacts the weight of others, but thats entirely what we're trying to do with negative prompting -- drum, drums, drummers, drumming etc

[azure mountain]

Do we have any insight from the devs on what negative prompts actually do (ex: how it effects the weights given to the model)?

[charred merlin]

#1256301881966198805 message it was an idea I threw out there a while back -- I haven't gotten around to making a proper set of suno tutorials yet because the last few months got mixed up -- but I'm looking forward to setting aside to make more time to help Suno learns find the same kinds of successes my friends and I have 🙂

[crude jetty]

(I know I'm a bit behind, oh well..) BPM metatags would be nice, then you can control the speed of the song, or how its supposed to sound from your head

[pastel solar]

can somebody tell my why suno is forcing me to creat with V4 , but im clicking on 3,5 and its creating with v4..

[crude jetty]

try refreshing the page?

[pastel solar]

i did , and than its v3,5 but if im creating its jumps to v4

[crude jetty]

is it just visual?

[pastel solar]

nope, its when i wanna extend a clip, than its jumping to v4

[grim epoch]

Hi everyone, I'm using the Suno API https://api.lumaapi.org/doc, and trying to target the Upload & Extend & Concat endpoint:

POST
{{PROTOCOL}}://{{LUMA_API_HOST}}/api/suno/upext

My request body is correct and the initial post request returns 200 with a taskId.

However, subsequent requests from suno to my callback URL are giving

msg: "TokenValidationFailed"
code: 2414

Has anyone experienced this and could possibly assist? I've double checked my API key and everything seems fine.

[marsh adder]

I'm pretty sure this site has nothing to do with Suno

[runic ocean]

idk if this is a bug but even if you're a subscriber it seems to default to 3.5 unless you change it

[grim epoch]

I'm pretty sure this site has nothing to do with Suno
The discord is literally called Suno?

[naive quail]

Tapping into the Suno API w/o authorization is a quick way to land your account in trouble.

[grim epoch]

What are you talking about, it's public and I've paid for it

[naive quail]

OK. Good luck with that. 🙂

[little granite]

stay out of the matrix 😎

[charred merlin]

You're going to want to talk to lumaapi.org and ask why their access to Suno through the API isn't working, and why the community helpers are telling you that they're stealing access -- they might not have a clear answer for you.

[marsh adder]

Look at the domain. Does it say suno.com or suno.ai? No. This is most likely yet another scam

On top of that Suno does not provide access to API

[grim epoch]

Look at the domain. Does it say suno.com or suno.ai? No.
That's a fair point

[charred merlin]

To be fair -- there IS a scam site that calls themselves sunoapi -- still a scam, not the legit thing.

[marsh adder]

Maybe we should keep the list of sites that have been proven to be scam? 🤔

[radiant mountain]

can you still use the "lyrics" part as a description?

[marsh adder]

what do you mean?

[radiant mountain]

instead of it generating a lyric, it would instead use the prompt to describe the music

[marsh adder]

so you want the AI to use your lyrics and generate a prompt with a suitable music style?

[radiant mountain]

you could on 3.5, just wondering if it still works the same

altho some times it did make it a lyric lol

[marsh adder]

This is the first time I'm hearing of this

You could try any chatbot for this, but I don't have a good experience with that.

I would love to have AI to describe music style based on an actual song 🙂

[little granite]

@grim epoch i would reach out to suno and not trust most sites

[charred merlin]

They're talking about metatags

[marsh adder]

Hmm..like this one? https://chatgpt.com/g/g-FBiHEhSQ0-suno-ai-v3-lyrics

[charred merlin]

https://www.suno.wiki/faq/metatags/ they work, but they aren't always "commands" that suno needs to obey, they're more like suggestions as to how the upcoming segment should be impacted

[radiant mountain]

ah

so if i have a few things in mind i should do [1] [2] [3] etc?

[marsh adder]

also people figured out that you can use different kid of metatags like ### with some luck. There was a workflow somewhere in this Discord. EDIT: Found it: https://discord.com/channels/1069381916492562582/1261146873331318835

[marsh adder]

The more tags you put together the more likely AI will start to ignore it.

[charred merlin]

The more you throw at it, the less likely they are to be used, but that would work better than cramming all 3 into one metatag -- as it says on the first page in the link I shared, 1~3 words seem to work best.

[radiant mountain]

well alright, it worked pretty well for my case, i used 7 words, seems to have captured my idea

[lofty osprey]

Hi All, after I completed a song on Suno a fellow writer helped me re-write some of the lyrics and I can’t figure out how to add him to the writing credits as he deserves. I tried Edit Song Details but no luck. Any suggestions would be greatly appreciated. Thx!!!

[crude jetty]

there's no way to co-author a song on Suno, best you could do it add them to the title, or in the lyrics at the end (as the AI won't read them anyway)

[charred merlin]

I'm having issues with "Regenerate Video" for the purposes of updating displayed lyrics in embedded/downloaded videos -- looking into it 🙏

[mossy cosmos]

HI. Is Suno down ?I get SSL handshake failed Error code 525

[crude jetty]

fine here

[crude jetty]

my favorite is my longer song says it failed to regenerate all the time, but seems to do it fine

[radiant mountain]

same here fine on my end

[azure mountain]

I never got anything resembling the melody of my humming, but I ended up getting a good song out of it regardless: https://suno.com/song/eb7ec1ac-4ec4-40ef-b0d2-7e9582a9cb56

[azure mountain]

Yeah that looks like probably a localized Suno outage

(I don't see that, but servers are generally geographically replicated, so some of Suno's servers must be down)

[crude jetty]

try using a VPN out of your area (like America or Asia) and see if that persists?

[charred merlin]

Cloudflare 500-series errors usually just need to be waited out. If you want to look into the matter, here's some housekeeping-troubleshooting steps you can try in the meanwhile https://cloudflare.zendesk.com/hc/en-us/articles/115003011431-Troubleshooting-Cloudflare-5XX-errors#525error

[mossy cosmos]

I just tried with US IP with VPN it works ! seems to be a france problem

[crude jetty]

there ya go! it's not that the connection has to be fast, so now you have a workaround for a bit at least

[mossy cosmos]

yes thank you. I use Opera free built-in Vpn. Seems to be good

[wispy kraken]

Hello i'm in france too i have this message " SSL handshake failed "

[charred merlin]

In general, it seems like nothing wrong with your setup, and points to a network issue that will be resolved shortly, one would imagine. Sounds like VPN access might be a workaround in the meanwhile 🤘

[bitter vigil]

I was wondering if you can upload a track to see what keywords Suno associates it with? Like Midjourney does it with uploaded images, for instance.

[mossy cosmos]

I Use Opera VPN if you don't have one. seems to make the deal.

[charred merlin]

Not something that Suno does atm, but that's a great thing to request over in #💡┃suggestions-web -- the devs browse that channel for ideas on how to make Suno better, and that could be helpful in coming up with different style prompts to use 💪

[bitter vigil]

@charred merlin, thanks ❤️

[heavy cobalt]

i have the same problem

maybe the problem come from the france

[wheat pivot]

I'm from Dubai but I'll try with VPN

[heavy cobalt]

oh

ok

is not u so

i have try with my 4G same problem

sadly

[distant burrow]

Yes me too, i have the same ssl problem.........

[heavy cobalt]

Wth

where did u from ?

[distant burrow]

France

[heavy cobalt]

i think we have a problem with the france cuz i'm come from france too

[wheat pivot]

Yeah with VPN it's working. Thanks

[heavy cobalt]

i dont have VPN

[distant burrow]

I will not use a vpn everyday to use this paid service

[charred merlin]

Is there the assumption you will need to do this every time?

[wheat pivot]

True, the devs might be working on it, it's temporary suggestion that it's working with VPN for now

[heavy cobalt]

thx

[distant burrow]

I Hope with the replace section too, haha

[pliant jungle]

https://tenor.com/view/angry-gif-9702280977932169114

Same

[mossy cosmos]

use OPERA built-in free vpn

one click and free

opera sorry

browser

[pliant jungle]

Moi aussi.

[mossy cosmos]

utilise opera, ca marche

[distant burrow]

Nan mais j'espère un fix bientôt, c'est juste pas possible d'utiliser un vpn tous les jours c'est illogique 🤣

[pliant jungle]

nous sommes bloquée par localisation?

[mossy cosmos]

c'est les serveurs de SUNO europe certains ont l'air HS. ils y en plein eparpillés dans le monde 🙂

[charred merlin]

@pliant jungle, Non, c'est une erreur temporaire . (just a temporary bug)

[mossy cosmos]

en tout cas ca marche avec une ip usa et asie dans le vpn d'opera

[vocal basin]

Hey guys

I need help

[mossy cosmos]

c'est juste plus lent

sorry i will continue in english

[vocal basin]

I uploaded a music file of my own and used Cover Song (v3.5) to remaster it, It remastered the intro perfectly but the brass melody is kinda different than the original

I would like to fix the brass melody to match the original

[charred merlin]

Covers sometimes spit out the same song you gave, sometimes it gives songs that are related but aren't the same, and sometimes it spits out faithful adaptations -- it's still in beta, so there's still some work to be done on it.

[vocal basin]

I am using 3.5 not v4

[charred merlin]

Covers work the same regardless of the model --
As for right now, the best way to impact the efficacy of future iterations of this feature is to leave ratings on the songs

[vocal basin]

Would you like to hear the original music and the one I did with suno?

So you know what I mean?

[charred merlin]

I know what you mean -- it won't be perfect every time, because Covers are still in beta. Downvote the generations that really don't reflect what you're looking for, and it will help improve the Covers feature when the full version of it comes out

[pliant jungle]

Quel horreur! Les "Spécialistes" ne font jamais "stress test" pour leurs serveurs

[vocal basin]

When will the full version be out?

[mossy cosmos]

Après la V4 vient de sortir, il doit y avoir du monde 🙂 (After the V4 just got released, there must be a lot of people around :)

[charred merlin]

There hasn't been a comment on that, but it'll be announced in #📢┃announcements or #🚀┃updates when it happens 🤘

[pliant jungle]

Sertainement. Tant mieux

[cursive island]

Advanced prompting nails it:

https://suno.com/song/a7487539-d597-4544-a60e-b2c39d21c159

at 01:00
(our love’s an echo in the night)
[Stretch "night" with reverb and morph into a synth lead.]

perfect execution of suno v4

[zenith depot]

Someone please AI ML train on the entire C&C Generals Soundtrack collection on youtube by zoom3000.

Exceptional music that GenAI needs to do.

[latent thorn]

am i missing something here?

[halcyon sparrow]

how i make a 5 min song with 30 min intro and 30 outro

i need the commands

[zenith depot]

Did you hear the C&C Generals Soundtrack music?

I want this level of quality or AI is not done.

[modern wraith]

still can't create song with voice extend with 3.5
I have just drained my credits, do something suno people

[charred merlin]

Are you on a paid membership?

[modern wraith]

yesss

[charred merlin]

You can only make 4m songs at once, but you can use "extend" to add onto the song and ... > create > get whole song to stitch it all together.

[charred merlin]

Ahh -- you'll only see Cover and Make Persona for songs from your own library

[halcyon sparrow]

ok i thought in v4 is longer available

thanks

[charred merlin]

🙏 No problem

[halcyon sparrow]

hahaah ok didnt saw it thanks

[modern wraith]

persona voice also ruin the v4 language support

I want v3.5 as my voice extend option

[charred merlin]

https://discord.com/channels/1069381916492562582/1308617723328204880 Here's a thread to report any v4 language issues, like pronunciation, etc

[modern wraith]

done with reporting already 😦

[fading panther]

Having some trouble with Replace Section. it's giving me a good replacement, but when I confirm it, the full song is missing the 3-4 seconds after the replaced section. I made sure to mar 'make replacement the same length as original'. But while my original song was 3:14, the full song with replacement is 3:10.

[charred merlin]

@fading panther , This post to #😜┃suno-memes might be relevant 😅

[verbal mortar]

you know what is frustrating? when you pick EXTEND and V3.5 and it spits it out using v4

[tacit haven]

What genre specifically could this be? (Ik its hiphop but i wonder specifically)
https://suno.com/song/8998081b-e04f-4abf-b37d-2f60e0231b3b

[outer monolith]

Shimmer still in on V4 or it in 3.5 as well?

[eager topaz]

Yes, in both for me, in the genres I've used in the last couple of days

[outer monolith]

thanks

[eager slate]

What does this error mean?

[marsh adder]

Maybe this? https://developer.mozilla.org/en-US/docs/Web/HTTP/Status/422

[charred merlin]

I've only seen mention of that error in relation to Suno in the last couple days -- at the moment, contacting [e-mail] is your best option

[latent thorn]

sorry, yes i am

[marsh adder]

Is the song you want to create Persona from, yours?

[charred merlin]

😅 I replied to "fekan" earlier thinking it was you because I looked quickly, saw the name started with F and didn't think anything else of it 😛

[latent thorn]

ahhh indeed alright, no it wasnt mine

[tiny tangle]

Hi, sorry if this is a grossly obvious question.
I've used Suno for free, and managed to create this: https://suno.com/song/3a927435-a567-467a-94d5-9a5af4b1f8c2
Which I absolutely love, except for the annoying typo I made at 2:37, and the mispronounciation of 8 as "eith" at 3:37 (coincidentally exactly 60 seconds later 😛 ).

Would using the "replace section" function of a paid subscription be able to "heal" these sections sufficiently without impacting the quality of the rest of the song, given that they are very small tweaks to make?

[midnight shell]

I gave in so many bugs! do I go get back the points as I am brought it?

[marsh adder]

It should. My personal observation is, that if you try to replace one section (ie. verse or chorus) and there is another one present in the song, it will keep the style and melody. If the section you want to replace is the only one like this in the song it will attempt to change the style and melody as well.

[charred merlin]

Reporting bugs doesn’t earn credits — it is public service to provide improved training data for future updates, so Suno improves for everyone 💪 🙏

[modest coral]

Suno v3.5 is still crap..it doesn't follow the prompts at all

[midnight shell]

ok so i lost about 200 points and 20 covers to it?

thats sounds not right mate

[modest coral]

I have a theory they changed 3.5V cuz of the lawsuit...and 4V is mixing with 3.5

[midnight shell]

in the end i paying for it

[modest coral]

at least I able to make some good songs before...Suno F-ed itself over 😦

[tiny tangle]

Thank you - annoyingly the typo I made is on a unique part of the song, so it would probably try to recreate that section.
In terms of the interface and exactly how much I can replace on that v4 function, am I able to select a short-enough part to replace the equivalent of one word, or is there a minimum length that'll result in a eg sentence being replaced?

[marsh adder]

min 10s

[modest coral]

When using .3.5V now

It has no smooth intro, and it abruptly cuts off at the end, or in mid speech.

While refusing to finish the lyrics. No matter how you prompt it, not to do this.

Is anyone else experiencing this? it also asses effects and stuff unprompted too

[tiny tangle]

Thank you 🙂

[frail trench]

So why is it when I hit "Get Whole Song" that my song sometimes doesn't end? It just cuts off. How do I fix this?

[charred merlin]

"get whole song" pieces together the extensions with the parts before it, but it doesn't generate any new content

[frail trench]

Howdy. So any advise on how to bring a song to an end outside of cropping a song at an ending point?

Also, I am noticing that some of my songs on v4 is getting a LOT of noise artifacting.

[charred merlin]

I've found success in placing an extension point at the end of a musical passage, where one would reasonably expect a song to "wind down" -- I remove all lyrics and style prompts and only use end, or goodbye or something similar in the style prompt section. Sometimes I'll need to adjust where I place the extension, but it's worked reliably for me

[frail trench]

Cool so in style prompt just put end and it will wind the song down. Thats neat! I am still learning all the commands. Any idea about noise artifacts in V4? I am getting this metallic stutter sound and unwanted massive amounts of distortion that its killing some of my songs.

[untold briar]

There is a subtle detail theharbor touched on, which is that it helps to change the Extend From Time to be a specific time in the song where it sounds a little bit like the song might be ending. It may work if you prompt "end" at any time, but if you listen carefully and hear a fade or something, that can help a lot.

[charred merlin]

I see it less as a “command”, and more that we’re giving Suno nothing else to interpret, and only using a single word that doesn’t have musical connotations — this is why I also use goodbye or farewell.

Like Jonathan mentioned, it might work to simply say end or outro or something at any segment, but it helps having a foundation of music theory — nothing crazy, but being able to follow along with the beat & count the bars — you’re more likely to swap styles or force a transition if it’s at a section where that would naturally fit into the song, like at the end of a bar instead of the middle of it.

Mind you, most of my understanding of Suno comes from “feeling my way through the dark of v3x” — I’m still getting used to v4, and there’s a number of ways that they work differently — I’m SO excited to hear about the discoveries the community comes across 💪

[eager topaz]

Noise, shimmer, distortion, is pretty widespread. Devs are aware and aim to fix the problem.

[graceful condor]

Does it effect certain sounds/genres more than others?

This, shimmer

[eager topaz]

I've tried a number of different genres and get it every time, some worse than others.

Anything with percussion, drums, seems to be worse for me

[graceful condor]

I don't have amazing speakers, I'm listening out of my phone, don't know if I hear it

Maybe I'll listen with my headphones

[eager topaz]

Might be hard to hear on a phone. I hear it with good headphones and studio monitors

[graceful condor]

Can I send you a file and you tell me if it's there?

[eager topaz]

Sure, not a problem

https://suno.com/song/b6cd46f5-c0c5-477e-802f-32deff2a2965

Here's one of mine. You can hear the noise quite easily

[true gull]

I'm sure it's been asked. but how can I clean up an audio recording? I haven't found anything in v4 that wasn't better than v3.5 but it's probably an issue on my side. thanks!

[eager topaz]

I use RipX Daw Pro to rip to stems and try to clean up each separate stem, but I'm finding with V4 it's almost impossible without replacing the offending tracks.

[spark dome]

Hello, where do I report a bug?

[fallow gull]

#🪲┃bugs-web #📱┃bugs-ios

[true gull]

cool, thank you. Sincerely, I Thought I was losing my mind, LOL

[daring talon]

How far off Android do you think we are?

[unkempt spoke]

Anyone know why my songs still having a suno watermark and saying "This is a fake app"?

[eager topaz]

Where are you making your songs?

[unkempt spoke]

im using the suno website

[eager topaz]

suno.com?

It's the only Suno website, but there are fakes

[unkempt spoke]

yes, im using the real one

[hollow hare]

Sorry to hear that, please fill out the following form and the Suno team will investigate the matter: https://sunomusic.typeform.com/to/zGsJSzsf

[unkempt spoke]

also i have membership

[old cedar]

I use the extend feature, and it created a second part of my song. But it didn't include the first part. Is there any way to combine the two?

[unkempt spoke]

Done, i sent it

[hollow hare]

Yes, use Get Whole Song from part 2 when it finishes generating
https://help.suno.com/en/articles/2409601

[old cedar]

I will look for that now...I never saw that option

[hollow hare]

Also see #1294418290348921007 for more information on how to use Suno

[hollow hare]

It appears when the part finishes generating iirc (from Part 2+, not in the base)

[old cedar]

Is it available if you exit and come back to it?

[hollow hare]

Yes but it should appear right away.
Does the new clip show a "Part 2" label? If not, it means an extension was not made (e.g. accidentally pressed X in the Extend From part)

[old cedar]

I do show a part two. But I don't see an option to combine them. Other words I don't see the get whole song prompt. Maybe that's just for paid members.

[hollow hare]

It's been a thing since like when it was released for both free and paid (as you can see in the images I linked earlier)

Do you mind:

• grabbing me a screenshot of how your menu looks like
• where are you clicking from (e.g. library screen, /song/ page, etc.)
• Are you on PC/mobile and using suno.com? Or using the iOS app?

[old cedar]

I'm on Android using the website because I don't believe there's an app available for Android yet.

[hollow hare]

If I recall correctly, someone had an issue of options missing when they were on a certain page but had it appear when accessing the menu through the library page

[old cedar]

It's dinner time so after dinner I will check it out. I really appreciate your help!

[late night]

My hard rock/power metal songs have this horrid, high-pitched tone that rings through at around 3300 Hz. Can't edit it out in every way I've tried, and all covers of the songs (even ones with dry reverb/no distortion) replicate those tones. Is there a prompt or exclusion I can use to get those whines out of any covers/new songs I make going forward?

[eager topaz]

V4?

[late night]

With V4, I've unfortunately been getting the whine and shimmer wombo combo.

[eager topaz]

Yes, me too!

All the genres I've tried have been the same

Even ripped to stems with RipX DAW Pro, but some tracks need total replacement

[late night]

Ditto. The best I can do is replace the REEEEE with hhhhhh.

[ruby ledge]

Any updates on the likelihood of releasing an API for developers?

[hollow hare]

Very unlikely at the moment. The team is small and more focused on fixing v4, the Android app, and a few other things.

You can ask [e-mail] for the official response

[timber lion]

Why when I remake a song using the same persona, does it use the Old lyrics even tho i changed them completely??

[worn flicker]

Well after everything, I am having problems going back to v3.5. V4 vocal quality is so much better, I can't un-hear the shimmer either way.

[unreal quail]

It does that. Sometimes an intro will be a line or 5 from the persona's original song.

[mellow condor]

but the shimmering hasn't really gotten any better, has it? 😦

[modest patrol]

has anyone started a wiki with all the advanced / new tagging that works in v4? I've been playing around with with [action] tags and I have seen others post in here more descriptive tags. I wanna learn.

[gaunt goblet]

How the heck do you generate in 3.5?

Every song auto gens in 4 and I don’t see how to switch it.

[sterile nacelle]

Okay... I understand working on the shimmer issue...but I want to know why NONE of my songs are doing what I prompt for? and creating duets where NONE are asked for. Styles, prompting are being ignored...verses are being overlapped. And I think Suno's been nipping at the bottle because the ones not over lapped words are jumbled like drunk. I have a few hundred credits left but I dont want to waste another 500 getting NO where in my works.

[gaunt goblet]

Thanks!

I’m assuming it just doesn’t show up on the app then

[raven turtle]

I download the WAV and clean it up in Logic Pro with RX, Tube EQ, The Oven, Musik Hack etc.

[fair adder]

currently trying to replace a section in my chorus but it's not fitting to the previous one (aka constantly changing the melody). i don't really understand, and i'm starting to think that replace section only takes a few seconds in memory to do the replacement, instead of the model which i think analyses at least 2 minutes of the song

if i can explain correctly

[raw ocean]

I have question why V4 can’t keep the same on Persona

[lapis willow]

wchich chanell i can start? I read its Suno chirp, but I don't see anything like that

[zealous night]

Question, Why, almost, to the most electronica generations there is a sine wave, in all its possible sounds you can create as a sound designer? It ruins most of the tracks generated. How can I exclude it from my generations?

[unique stone]

Chirp is gone. The website at https://suno.com is where you can make music now

[lapis willow]

Oh that explains a lot xD thanks!

[broken dew]

Why is the extend feature such low quality?

[tidal fable]

My Suno doesn't work for more then a week now (does not generate). Does anyone know how can I reach the IT?

[polar spade]

Let's find out what "remake a song " means in this case. Are you:

1. having the old song open, clicking menu , choosing "use persona" , which should open a blank song with your persona. Then you've entered lyrics and you get your old song played again?

2. You're going to persona page, choosing your persona, then clicking " Create with Persona" button for a similar blank page page. Adding lyrics and click create.... and you get old song again?

Or 3) clicking Clear to clear your CREATE page, going down to persona , click No Persona Selected drop down, and choose Make New Person...and scrolling through songs and choosing old song. Naming persona . Clicking Save. Adding lyrics and click create and you get the old song playing?

Or 4) another way?

[timber lion]

Number 1 mate.
Even create some other songs in between, went backl, blank slate. Picked the Persona from the list Pasted the new lyrics in to the box.
Song plays same as the old song except 1 line.

[polar spade]

Great. I'll try a couple. I'll paste in Mary had a little lamb poem for the lyrics and see what I get.

[polar spade]

You using old song of 3.5 or 4.0?

I'm creating an "old" song off a persona... then using that songs persona to create new song.

Need to know which version your old song was in.

[timber lion]

Actually happens with both. Ive seen it happen with a few songs i did.
I had misspelt a few words and wanted to change about 3 others to sound better.
So I went back, created a persona.
Then started a new song, using the persona.
Pasted in the new lyrics.
Create song.
Old song plays.
Or sometime some of the new lyrics come and some dont. Usually where it would repeat a chours.
The first one will be the new one the the rest the old one.

So i select the song. Since remastering or covering it u cant change the lyrics, i figured making the persona might give more change of getting the same sound again.

[polar spade]

Oh. When you say "new lyrics" it's the "old" lyrics pasted in but with some word changes?

Or totally new lyrics, new song idea. ?

[timber lion]

In 1 song is spent and as "abd" didnt notice till days later. And ive been unable to correct it lol

I havnt completely change them, they usualy start off identical.
But minor or major changes.

Havnnt tried complely new, ill give it a go

[polar spade]

Ok. Got it. I'll read your great details and see if I can replicate.

Which version ... 3.5 then?

And "pasted in the new lyrics" , as you say, means "pasted in the revised lyrics"?

[polar spade]

Also, have you tried "replace" for the words. There have been issues with Replace but some work and it may be improved to work better by today.

[charred merlin]

It’s not necessarily designed to give the same section melody, it’s “in painting” for music 🙏

[timber lion]

haha ok So i changed the whole verse 1 to
[Verse 1]
Mary had herself a little lamb,
Its fleece was white and it wasnt a ham,
Every where mary would go,
You know that little lab would follow.

And it sang the original verse but slured the words

[timber lion]

Replace is good for a long section but for a single word it sucks. Like it will ofent completely change the tempo or rythum of the verso.

Id like to be able to reword an entire song and keep the music. Any way to pull that off?
Getting the stems, thenm uploading the music work?

[swift ridge]

Please create an API for Suno!!!

[charred merlin]

Nothing currently that’d make that swap easily, but feasibly you could strip the vocal stem & layer vocals onto it later

[timber lion]

Dam kinda sicks when ya just messing around for fun and something epic comes out and u wanna keep a lot of it lol

[charred merlin]

#💡┃suggestions-web is where the devs look for suggestion.

Tech-chat is fellow users, we can’t help with this, but we want an API as well.

[timber lion]

Ive been just taking the stems for now and maybe ill mix them in later as i get better at the manual editing, cross fading ect

[polar spade]

Well bad news on my end. I was able to replicate your issue! So, as one member to another, I'll try to throw this into the BUG section and ask if there appears to be a glitch or I am doing something wrong.

[timber lion]

Well at least that means its a bug and we can report it.

[charred merlin]

You could keep the instrumentals, and extend from them with new lyrics — that way Suno has the same music and instrumentals to continue from, but wouldn’t be locked to the previous lyrics — this method is technically the easiest and most direct, but you won’t have the EXACT same structure to the music, unless you spend the time to create it (but it’s a matter of how much you REALLY want that music with new lyrics)

[timber lion]

I spent almost 1k credits trying to reproduce a particular sound so yeah i wanted it lol

[polar spade]

I made the songs under 3.5. By the way. I had not heard which version you used.

[timber lion]

Both

[polar spade]

Ok. Well we have a video of 3.5. 😊

[timber lion]

Its something ive noticed reguarly, but didnt piss me off totally till today lol

[graceful condor]

i need api access..i am willing to pay. i understand that these leech bastard companies who provided unofficial measures were abusing fake accounts its kind of stupid. i also understand that you dont have an api because it would forced you to stick to it and it might not be perfect as suno evolves, but can we at least just whitelist my ip so that i can use a bot to generate music pls, im a paying member who will buy credits..not a leech

[charred merlin]

☝️

[polar spade]

I used a browser version under my Android phone. What did you use, and I'll mention it as well.

[timber lion]

See this kind of thing I dont like. I feel there needs to still be a human element.
Plus the idea of the system being overwhealmed by idiots with bots mass producing trash .

Im sure there will be an Open Souce project that does similat things u can run locally. Abuse that 😉

[graceful condor]

abuse? you think the credits dont sell at a net profit? dont throw stones from a glass house by calling people idiots. have a good day

[graceful condor]

thank u

[timber lion]

Dont care, still needs to be some kind of human element. Not mas produced garbage.

[graceful condor]

this chat is for technical discussion buddy, not philosophy

take your thoughts down to general chat, where you can talk by yourself bc you wont find me 😉

[charred merlin]

(On consideration while slowly waking up, I apologize if forwarding & a pointing emoji seemed like a cold response 🙏)

[timber lion]

Off u go, u the one coming here trying to discuss the non existant API.

[graceful condor]

i know your intentions are the purest

dont bother apologizing

[charred merlin]

🟰 🇨🇦

🤷‍♂️

[graceful condor]

hahahahaha awesome

[charred merlin]

(even the shrug emoji means "sorry about that" in this instance 😛 )

[polar spade]

@timber lion so, I've put the vid and request in the bug section. You're part is to send am email off to [e-mail] for a personal response as well.

I have, so far, made a persona, but created a new song to which attached the new persona. You've used the old song off a chain of steps to create a persona and use same to revise lyrics from the old song. (Which I think I re-created in video)

That may be a new twist with errors that they haven't encountered. Thanks for bringing it up, at least. Let hope there's a solution or my technique error is pointed out to me (and you).

[muted wasp]

Do anyone know who to create "spoken word" part and no came out like rap part?

[charred merlin]

I've gotten success in v4 with argument, spoken, speech, spoken word, with the negative styles instruments, singing

[naive quail]

Here's an example using "Spoken Word": https://suno.com/song/13a0d143-7eb8-4896-9849-346282b5539c

[polar spade]

If not familiar, those would be put in square brackets such as:

[Spoken]
Now is the time for all good men to come to the aid of their country.

Or

[Spoken]
"Now is the time for all good men to come to the aid of their country."

[naive quail]

There's lots of fun to be had with spoken word in Suno... https://suno.com/song/e737719d-8421-4a4b-89c5-338d8df5b930

[high cipher]

This is great and I wish the same had worked for me

Lowering 2.5K on the dot didn't effect the shimmer, but at 2.5~3K hrz I realized if I essentially castrate the frequency in a relatively wide width I'm hearing some shimmer reductions, although the shimmers are still there; but expectedly this also negatively effects the song albeit in small regions.

That said, I think you are into something. I wonder if you have managed to reproduce this consistently with other songs and other genres.

[charred merlin]

Yeah, I've noticed some of the "hallucinatory songs" are a LOT more cohesive than one would expect -- sometimes it feels like an entire track, but with a couple words garbled the same way that v4 remasters do if you don't align the correct lyrics. Previously there were nonsense songs that would have the occasional non-gabled word so that it sounded like English (shout out to Adriano Celentano's nonsense-English song "Prisencolinensinainciusol" https://youtu.be/RObuKTeHoxo), but these are the inverse -- complete English sentences, with the occasional gibberish word

Trying to mitigate "dataset regurgitation" is a big challenge for AI models 😅

It's a bit tangentially related, but this is why I found the new chatgpt 4o to be FASCINATING -- the research paper shows their stated goals, one of which is reducing regurgitation, so when it's "thinking" you can see the iterative steps where it "talks to itself" and ensures that the content it's providing meets a number of requirements (as well as giving multiple steps to fact check its own work) https://cdn.openai.com/o1-system-card.pdf

As a last followup on that -- the Suno founders used to work for Kensho, who among other things, are working on ways to reduce jailbreaking and related "toxicities" in LLMs -- because it's fun and silly when we think of jailbreaking as making chatgpt write something mildly offensive, but becomes a little less silly when you think of "jailbreaking" as ways to cause self driving cars to misbehave, or trying to instruct future voice models to mimic someone's voice without consent / nefariously. It's not surprising that their follow-up company also seems to have a focus on trying to reduce unwanted regurgitation of training data https://blog.kensho.com/an-introduction-to-mitigating-toxicity-in-llms-c8d1c5600a9b

[tight basin]

but its really limited

knowledge cutoff at october 2023

no web search, no attachments

and 50 messages limit per week

[charred merlin]

Wildly shortsighted. It's a tech preview, stated directly in the name.

4o is the flagship model, stick to that if you want full features.

[tight basin]

ur right

[charred merlin]

I'll also concede that I'm DYING for o1-preview to get persistent memory, at the very least. As of right now I need to start a conversation in 4o without using any new features, ask it to relay EVERYTHING it remembers about a subject, then swap to o1-preview for the rest of the conversation -- not a world-ending inconvenience, but I'll be so excited for the full release 😄

[slow magnet]

Anyone else having the same problem buying new credits? Wasn't a problem a few days ago.

[cursive island]

How do i make suno sing like a childrens choir I want to make a homage to the song childrens dreams

[bitter spire]

Sorry small question is this just me but 3.5v songs are in my opinion much much better? V4 is good but non usable with that shimmer and overall its like v2 comparing to v3.5 in sound quality

[unreal quail]

[Children Singing] does work when used with a Children Choir style of music.

[thorn heart]

Anyone know how i can find my audio clips I uploaded

[charred merlin]

They appear as the most recent clip in your library

[languid thorn]

folks, is anyone still having problems with cropped songs? new ones or even ones that have been ready for days.

[charred merlin]

Can you elaborate? Are you having issues cropping songs, or using them in some way after they’ve been cropped?

[crisp sparrow]

Does anybody know any good methods for influencing which parts of a custom lyrics song play with a "full band" sound and which are more "stripped back", if that makes sense? I find Suno tends to put stripped back/gentle parts maybe after a bridge or a instrumental break, which fair enough does work in a lot of instances, but I've definitely been a little disappointed when I've had a song that turned out pretty much perfect apart from the last section not sounding as I'd hoped.

I've tried using several different tags to suggest a structure, like "[Rock Chorus]", "[Acoustic Verse]", "[Gentle Verse]" for example, but nothing seems to make any difference. Would anyone please have any insight into whether this is at all possible?

[normal quartz]

Yup, most of my songs ends at 3:14 even tho the song was not finished

[unreal quail]

You tell the Suno to do X and it sometimes is like 'Nahhhh, Imma do this instead' Roll of the dice

[marsh adder]

obedience or creativity - pick one 😉

[normal quartz]

And this issue with cropped songs is super annoying to be honest, I'm not wasting more credits until this is fixed.

[naive quail]

If they gave us a slider, we could...

[marsh adder]

Honestly... that is a great idea. You should post it in one of the Feedback channels.

[naive quail]

I'm pretty sure that others already have a few times. Probably somewhere in the queue

[languid thorn]

this is "persona bug" not the crop bug that makes song invisible to trending and research

[marsh adder]

Yeah, but the more occurrences of such idea meaning more people want it and it's a hint for devs to prioritize that over anything else

[warped furnace]

Work with extensions and change the prompts accordingly. That should help. Not perfectly, though. ^^

[marsh adder]

Also use thumbs up and down between generations to direct AI into what you want to achieve

[unique stone]

Outside of assisting the developers with refining new features/models this doesn't have a direct impact on how Suno works

i.e. if you thumbs-up a lot of things in your library, it won't make the next generations any better

[marsh adder]

Are you sure?

[charred merlin]

From 10k+ tracks of observations, that's my firm belief as well.

[marsh adder]

Well, personally I feel like it does work and changes subsequent songs I generate, gently enough that it can be in my mind, but....that's how I feel about that

[charred merlin]

I've been asking for clear demonstrations, but until anyone has something that can show (like making a bunch of metal songs, then swapping to childrens music and having metal continue), it's speculative.

[marsh adder]

Well....yeah, I was never able to make such drastic change. It's more like adjusting the aim in my mind

It's subtle at best and not working at worst 😛

[charred merlin]

Furrybeans asked a dev whether it worked that way, and the response was that it didn't, but they WISH it worked that way.

From everything we've been told, it helps give feedback about current models so they know how to refine later ones -- not necessarily about feeding the "good results" back into training data, but like confirming that whatever assumptions the AI made to spit out that song, they should continue to assume that

[marsh adder]

Ok, I can't argue with the dev saying that 😅

[charred merlin]

I don't doubt that the AI can feel like it has a mind of it's own, and when we start to combine LOTS of style prompts to try to get our desired output, I can understand how there would be some overlap.

I used to be a lot more staunch in my beliefs that it didn't work like that, but now I'm just asking anyone for demonstrations if they genuinely believe that previous generations are impacting current generations (beyond a one-off hallucinatory track)

[marsh adder]

@charred merlin We should call you Suno Jesus for all the wisdom you share among us little people. Oh lord give me wisdom and understanding! Forever and ever 🙏

If there is ever an ai bot created for Discord to support in first aid, they should train it on all your comments. 😁

[modest patrol]

It would be a good idea for some of you to read up on how AI works. The models are static, suno isn't fine tuning your generations based on your like trends. You're having a severe case of placebo effect.

And yes it's a roll of the dice. AI is always a roll of the dice, suno isn't using static seeds and doesn't expose any control over it to us.

[marsh adder]

I'm not talking about "training", I know they are static. But on the other hand covers and remasters work so they uses other song as input, so it's not impossible. Now... having 20 or 50 previous songs as input IS rather impossible with current technology 😅 so I agree it can be a placebo after all.

[charred merlin]

No one said the models are updated in realtime -- the feedback provides data the team uses when making updates.

The level of RLHF used here is pretty straightforward. https://en.wikipedia.org/wiki/Reinforcement_learning_from_human_feedback

If a baker makes 10 batches of cookies, and tries wildly different recipes for each of them -- it helps to know which batches tasted good and which batches tasted bad before they go back into the kitchen again -- especially if the baker is a robot incapable of tasting the product themselves to know if it actually tastes good beyond thinking it SHOULD based on how the ingredients were combined -- like when Bender from Futurama tries to cook for everyone.

[crisp sparrow]

Thank you all for your input. Believe me, I'm very much aware of the element of luck involved with all of this. (When you specify male vocals and exclude female vocals, and then it gives you female vocals anyway, you quickly realise you are ultimately at the mercy of your AI overlord... 😆 )

Still, it would have been nice if there were some meta-tags which seemed to help direct the structure a little more in that sense (make this part more gentle, now bring the whole band back in here, etc...) But I guess the unpredictability of it is part of the fun sometimes.

[gritty badger]

Is it just me or did the servers slow down?

[marsh adder]

This could be cool option for some, but for me the randomness adds this creativity which is something I would never be able to achieve with music because I have no talent nor skills. Some people are adding metatags with bpm, instruments, chords and for me...I wouldn't know what to put cause I don't know what I want 😛

[charred merlin]

First Aid ai bot trained on comments by @charred merlin

"Unfortunately, Replace Section doesn't work with lungs, and Extend doesn't work on lifespans -- I feel like Discord might not be the most appropriate channel to let us know that you're having trouble breathing. afaict, your local hospital shouldn't be experiencing peak traffic at this time, so you might receive better assistance contacting the physicians at [e-mail] 🙏 "

[marsh adder]

"Your feedback is important to us, and helps us make Sun....ekhm....theharberAI better. Please continue to give us feedback by upvoting 👍 comments you like and downvoting 👎 comments you don’t."

[sinful abyss]

after the release of v4 I have problems with the track length in v3.5. It used to generate full tracks of 3 - 3.5 minutes, but now it always turns out to be 2-2.5 minutes and the track breaks off. Has anyone else noticed this? Is there a thread where this is discussed that I can subscribe to?

[worn flicker]

You can direct the structure

[worn flicker]

Thumbs up thumbs down go into a dataset of their own either directly or in a way. It is used to better train the model but training still has to he executed, it does not happen in real time.

Everyone should definitely use report though. Thumbs down is to get it not to show up in your library while creating, and thumbs up is obvious. At least that is how I use them all

[marsh adder]

Wrong quote 😛 This was us giggling about a bot based on theharber and it could have thumbs as reinforcment learning 😂

[worn flicker]

I should have replied to your earlier message...lol

[marsh adder]

no problem

ok ok, thumbs DO NOT work, I get it 🙄

[worn flicker]

Are embeds being done by you? Like a quick way to get the name and embed it without typing it out?

Tick mark and all that

[marsh adder]

You mean this? This is reply

ohhhh.. this

[worn flicker]

Nah @theharber.

[marsh adder]

I use backtick, yeah, I don't want to ping him

[worn flicker]

I was just wondering why it embedded

Oh ok

[marsh adder]

"Thou shalt not take the name of the Lord in vain"

[worn flicker]

Amen

Hey, so v4 has the shimmer, but if we take the path of changes we want to hear. I could swear that when they reverted to 3.5 stuff they shifted the weights some. Because v4 is wrecking v3.5 atm

[marsh adder]

I wonder how much time do they need to release new model after adjustment. Can this be done weekly, daily... 🤔 ?

[crisp sparrow]

Well I know you can specify "Verse", "Chorus", "Bridge" etc. if that's what you mean. But everything I've tried to tell it, for example, "Acoustic Chorus" to distinguish it from a chorus with a full band, has never made a jot of difference. By all means if you have any ideas of what I could try, I would be extremely grateful.

[worn flicker]

Sure, plenty ideas. First break up your tags. Use extensions to get what you really want, use style to get your acoustics in there. V4 listens fairly well if you give it all the tools it needs. Of course there is some luck of the draw in there. What are you trying to do?

From what I can tell, this is purely my experience, if you give style the things you want they are easier to call when you need them in tags. In a way...easier to show than say.

[lucid quest]

Hi folks any idea on how to get Throat singing into songs. Ive tried but I'm starting to think Suno has no training data for it

[crisp sparrow]

Hmm, okay yeah, I see what you're saying. For the example I'm thinking of the most, I gave the style as "slow emotional power ballad" (which, yes, I realise is fairly generic compared to what you're suggesting.) The thing was, it turned out almost exactly how I was hoping for: a gentle keyboard background for chorus 1, verse 1, chorus 2 and verse 2. Then the more heavy section kicked in for choruses 3 and 4. The song finishes with verse 5, which ideally I wanted to be gentle like the beginning, but instead it kept the heavy sound.

Let's just say that extending or replacing the section wasn't an option available to me for this particular song, so ultimately I just have to try and create a new take on it.

[worn flicker]

It depends on what you mean. There are definitely unique Vocals in the world I have not heard anywhere near them on Suno

[worn flicker]

So have you tried extending from when the part you don't like started?

[lucid quest]

Mongolian Throat singing. Its something I'd like to get added into some techno or even more folky music, but so far no luck even with V4

[worn flicker]

The explore page on Suno's website now shows v4. You should poke around. See what you can find.

[lucid quest]

Will do, thank you 🔥

[crisp sparrow]

Wasn't an option. Long story short: before I began using (or had even heard of) Suno, I was using another site that worked well at the time, but it turned out that they were just "piggybacking" off Suno's API and then suddenly the Suno "audio watermarks" started appearing, so I had to abandon it.

The song in question was made on that other site. So, unless I was somehow able to get that song into Suno itself and then extend it, (which I'm guessing won't be possible,) I'd have to start from scratch.

[worn flicker]

Oooooooooooo....OK. that make way more sense. If you have Vocals then suno may take it but not public, so if you are doing it personally there are many ways to make things happen. Depends on the importance to you.

Man those piggybacking thieves suck

[crisp sparrow]

Yes, I know now, yes... 😢

[vapid nymph]

So I'm guessing 3.5 personas can't be used with 4?

Or am I an r tard?

[worn flicker]

If I was making a personal song, I would make the vocals with v4 and an Instrumental with v3.5 and then mix them in a DAW.

[worn flicker]

I know v4 can be used with v3.5

[worn flicker]

Doesn't work very well in my experience

[charred merlin]

It'll default the selection to whatever model was used to make the persona, but it should be swappable

[vapid nymph]

So I basically just need to try and recreate the sound in v4 and re-save it?

[crisp sparrow]

Yeah it's for personal use, with my original lyrics, and the song does mean a lot to me. Which is why I was especially gutted to have it so close to perfection, but not quite there.

I know there's an import audio feature in Suno, although I've not tried it. I'm guessing that's not what it's for, however.

[vapid nymph]

I tried switching and it was refusing, but I'm using mobile and it's been slightly janky since 4 launched. Not like hibderingly so just silly janky.

[worn flicker]

Ok then you may be able to make the import but I don't know about Suno accepting something that was made with the watermark in it.

[worn flicker]

For me anytime I go from v4 to v3.5 I get a warning on quality difference.

[charred merlin]

That hasn't been outright stated, but if you try to continue a v4 song with v3x or vice versa, you get a disclaimer that the models don't play well together -- since persona retains a clip of the original song for context, I wouldn't be surprised if you'd get "technically more compatible" results by remastering the v3 track you made the persona from and making a v4 version of the persona, but it's not guaranteed to perform the same as the original (unless you're giving a SUPER specific clip for it to use)

[worn flicker]

You can only upload 2 minutes anyway

[neat lake]

Anyone found a way to enforce prompts, so that Suno actually generates what it's asked to?

[worn flicker]

Yup

[worn flicker]

Well, more like suggestions

[neat lake]

Well I'd be glad to know a few, if you don't mind 😄

[worn flicker]

It is a natural language style generative AI, so that is a very broad topic, what are you looking for it to do?

[crisp sparrow]

Oh no, the song doesn't have the watermark - this was before those started happening. I suppose the way my naive little mind is thinking: if the other site was essentially using Suno's system anyway, then could Suno itself still work with the track? Or if the other site was making use of a Suno account, then does the file actually exist saved on a random Suno account somewhere that I can't access?

But that's all hypothetical, natch...

[neat lake]

It's the opposite, actually. I know what I want it to do, but I also know what I don't want it to do. I specify RNB, Trap, Rap, and it generates some pseudo mPop Dancy stuff. And that's the main issue

[charred merlin]

If you still have the song links, they might have been lazy enough to re-use Suno's UUID for the link. https://cdn.discordapp.com/attachments/1249416146227040307/1302247762200367174/20241102-1226-21.1780689.mp4?ex=67446d3d&is=67431bbd&hm=88ef08bcb3d199b84de72572b0728bb52e8483f59975c26531e6792017a23e2e&

This is a STICKY situation in terms of ownership, since technically the songs were made on a free suno account, so they retain ownership -- so you could extend from the song, the same way you can extend from any Suno.com song you have the link for, but you wouldn't be able to monetize it, since it's extending from a clip that you don't have the rights for.

You MIGHT be able to email [e-mail] to ask if there's anything that could be done, but since you didn't create the account that generated it, it's an even more sticky situation for them to grant rights between accounts with all this obfuscation in between

[vapid nymph]

How well do you think it would work just reusing the prompts that made the persona and using the same set of lyrics? Like.... If the main chunk of the persona is "hardcore death metal female" or something, will it be close enough ran through v4 to build off of and get at least close to the 3.5?

[worn flicker]

Ok, you got it in your head, what I tend to do is go poking around the explore page now that it has v4 updates. I'll use that to find the base for the song I want to create then add on the other elements. That said, it isn't easy to get it to put down the style sections as you command. It takes some work. Extending and prompt adjusting and tag adjusting

[worn flicker]

If you start with an upload then it does a pretty good job of extending from it

[neat lake]

Oh definitely, I adjust the prompts, and play around with tags, which often don't work in the newet model, but since I have given you an example, I'd appreciate less broad and more specific sugestions. I'll be happy to try them out 🙂

[charred merlin]

You could just remaster the v3 track it was based on, or do a v4 cover of it with the same style -- a moderate number of the results should be more or less the same track with minor variations to pitch or tempo, so those might be a quick way to keep the same style for the purposes of generating similar tracks

[worn flicker]

Sure. Remember that there are tags that work way more consistently than others. It is just difficult to get instruments to do what you want so much as lyrics. You can get it to do an Instrumental and even one with certain instruments but at the end of the day it depends on what it was trained on and if those tags lean it in the direction it was trained.

[worn flicker]

Once I got a Duet to do what I wanted in v4, I started making shifts to my tags and prompting.

[vapid nymph]

Is remix feature stored in the balls?

[crisp sparrow]

lol That's hilarious! 😆 Thank you for the idea, theharber. A very nice thought but sadly that didn't seem to work in my instance.

I'm not concerned about monetising anything, but I agree that it's probably too much of a sticky situation to try asking Suno for anything. I have the MP3 of the track, so maybe I'll try the upload feature and see what happens - nothing to lose, I suppose.

[trim glade]

Hey Guys So far I have to say version 4.0 is not as good as I thought. The voices sound callous and sound like they are being read from sheet music. 3.5 has voices that are of poorer quality, but the voices sound more soulful. i Hope when the beta is gone the voices are better v.v

[crisp sparrow]

Actually, scratch what I just said... I just did "view page source" on the other site and found the Suno link on there, so that certainly gets me closer!

A massive thanks to you both, theharber and Sysmai. I guess we'll see what happens. 👍

[dark spire]

Anyone else having issues with 3.5 getting the full song ACTUALLY being the full song?

It's completely ignoring my part 2 extension.

[wispy current]

Since version 4.0 has been introduced. I have generated around 50 instrumental music pieces so far using various prompts, but none of them have reached 3:00 minutes in length. They are usually only 1:30–2:20 long, and all end abruptly. The music simply stops, with no proper ending or "Coda."

I have not succeeded even once in generating a single piece longer than 3:00, despite trying 50 times. This is quite strange.

Since I’m trying to create longer music compilations using Suno, these two issues (generations being too short and lacking proper endings) are major obstacles.

Here are examples of my prompts:

"Slow piano solo music for Christmas"
"Slow, long, complete, 4 minutes long, relaxing piano jazz for Christmas"
"Complete, 4 minutes long, relaxing slow piano jazz for Christmas"
"Slow, long, 4 minutes, relaxing piano jazz for Christmas"

Can you help me understand what I might be doing wrong?☹️

[tiny falcon]

Have you try "A 4 minutes song about... or of...."?

Sometime it works for me in both version. If I don't put that, it hardly made a 4 minutes song

[unique stone]

Can you post a link to one of these short songs?

[wispy current]

Thank you, I'll try it.

[tiny falcon]

I do a Ctrl-Refresh the page if I notice the songs are very similar. Not sure if that help too man.

[wispy current]

Here is an example: https://suno.com/song/0d282723-1c26-40a0-858b-9f5176be3abf (Prompt: "A 4 minutes slow relaxing piano solo about falling snow")

[short fjord]

Sometimes it comes to Distortion with V4, is there a Prompt to minimize the Distortion ? i tried wit [Max -6db] and [-6db] but i think thats not the way to go 🙂

[unique stone]

Hmm. You seem to be using custom mode rather than simple mode. Simple mode is where you'd normally put this kind of prompt. Custom mode is more for defining the style, like "piano solo, relaxing" - it won't understand what to do with "about falling snow" for example.

I've also never had any luck specifying a duration

If you try a completely unrelated genre etc. does it give you longer songs? I see you have a public song from back in June which has a length of 4:00 so maybe reuse that prompt?

[charred merlin]

I've found shorter generations are usually caused by the style prompts -- either something being used that causes a bug, or not enough information for it to "run with"

[kind token]

Since the update, even the cover feature doesn’t function as well as it did before. even I'm trying to use the 3.5 the recent changes to the cover feature. It no longer matches the input as closely as it used to, which impacts my creative process. I hope you can acknowledge this and consider reverting to the original approach.”

[charred merlin]

Out of curiosity, I tried simple and custom -- the simple mode songs resolved, while the custom mode continued up to 4m

[leaden pike]

Bro the shimmering shitshow is so unbearable on metal/rock songs ... holy mother of God. Gona pause my subscription until this is fixed.

[tiny falcon]

Try this out man. I am not sure if it work on all the songs though. I ask GTP to get me a list of prompt to put in the exclude style. Seem like it works a bit:

Here:
repetitive beats, distortion, shimmering sound, overly continuous patterns, washy pads, static noise, oversaturated tones, wooshing tails, cluttered layers, overlapping frequencies, droning sounds, excessive reverb, muddy bass, hazy textures, noisy soundscapes, chaotic rhythms, overcompressed sounds, unclear tones, overly airy pads, swirling effects, harsh transitions, unbalanced dynamics, overlayered effects, sharp high-end, artificial artifacts.

https://suno.com/song/ad83bb3c-5f33-4242-90d5-dedf415bd542

[gray siren]

Did a v3 to v4 remaster, the v3 song had a place holder picture because that's kinda what happened when you deleted the parts you didn't need and just left the final thing there, so the remaster had a similar place holder picture, went to fix all that for all my songs by prompting suno for images, it did that, so I went to regenerate the videos, it all worked fine for v3, v3.5 & v4 songs, except for one v4 song where I tied a persona to it as a test ~ I can't delete the persona, so I'm thinking that's what it is unless it's a different kind of backend problem specific to that song for some reason, tried making the song public, private, same with the persona, moving it into the library again, trashing it (since we can't perma delete for some reason) ~ Anyone got somethin' specifically similar?

[tiny falcon]

But I am afraid it might remove some of the effect or instruments that you put in the positive style prompt. Maybe someone can figure this out

[viral lotus]

My tags embedded in the main lyrics body are frequently vocalized. Is there a standard formatting in Version 4 to get around this?

[leaden pike]

Thx bro. But im just done. Period. 3.5 version is an actual upgrade of this crap v4. Just listen starting from 00:10 second and u get what i mean:
https://suno.com/song/4c224f41-db3f-4574-b62d-3afcc4fede55

[north sonnet]

I put tags in square brackets e.g. [Instrumental]. Hope this helps

[leaden pike]

Thanks Rick. I tried all the possible tags and things

Is just their "brand new" v4

[unique stone]

They are working on addressing this (acknowledged in an announcement posted a couple of days ago.) It is a very annoying noise so I get the frustration. Hopefully it'll be resolved before it leaves beta.

[wispy current]

I tried simple and custom mode, both with the same prompt you used. Longest 2:28, shortest 1:16... I dont get it...

[crisp condor]

I’m having an issue downloading music from my library today. Very delayed pop up when requesting the WAV. Like over a minute waiting time.

Anyone else having this issue?

[tiny falcon]

Yeah man. I notice the song with vocal is better than just instrumental. I tried with those excludes tags, it help a bit, but the song is flat 🙂
https://suno.com/song/1ab48349-6ea6-48b1-95c2-9e5ccad65d57
I guess we'll wait for the fix.

[torpid thicket]

All hail the Shimmer

[tiny falcon]

Maybe make all song into Lofi. Most Lofi music has crackle and shimmering sound right?

[charred merlin]

Other users here are able to give community advice and help test things out like we did, but if the issue has persisted for a length of time, it sounds like you might need to email [e-mail] so the support team can look into it.

[wispy current]

I will, and thank you very much for the helpfulness!

[shut gyro]

Maybe late to the conversation if already picked up. I can't choose v4 anymore. Is this a known fault?

[fathom siren]

Anyone have a template or sizing for cover art so the title isn't cutoff anywhere, and looks decent on phone, playlist, side window?

[charred merlin]

Nothing system-wide -- does the issue persist after a browser refresh? Any error messge on https://suno.com/account ?

[untold briar]

The upload screen could have this guide on it directly.

[fathom siren]

Ya I've seriously made like 20 versions just to open the song somewhere else and see it cutoff etc..

[untold briar]

Like early widescreen DVD releases of TV shows where the editors kept leaving in crew members holding boom mics

[shut gyro]

@charred merlin Just been playing around again and it's working. The issue i was getting with it only allowing the 3.5 is when using a persona

[charred merlin]

Gotcha -- I know it swaps back to the model used to create the persona, but afaik you shouldn't have any issue swapping between v3x/v4 after selecting it -- glad that got sorted 🙂

[tight mango]

Yooo, how's the shimmering doing

[fathom siren]

I don't know I haven't experienced it that much in 4.0, although I do have 3.5 songs with it.

[charred merlin]

#📢┃announcements message As announced, they're working on it -- make sure to follow that channel and #🚀┃updates for any news at the same time the rest of us will find out 💪

[vapid nymph]

Anybody know what this weird stuttering effect is? https://suno.com/song/c9b8ba10-108c-4120-ba1b-d23f7fedcefd

https://suno.com/song/c1e24f3f-216f-447e-9f59-cbbc09ba02b8

I get it on most of what I make with 4.

Seems to be related to the drums but idk what it even is?

[fathom siren]

What point do you hear it?

[frail trench]

@slender lodge My apology for the accidental ping at the entire server, 🥵 I sent an email from eliasrodriguez@####mail.com regarding technical issues with V4. Basically I have been experiencing audio artifacts that have ruined one of my songs. I'm looking for a reissue of my credits. I sent the email to: [e-mail]. Thanks for looking into this or flagging the appropriate folks. And sorry about pinging the whole server again! Thanks!

[charred merlin]

Hey @frail trench, just as a heads-up, the community can offer advice and limited technical assistance, but the Discord Server isn't an official support channel atm.

Afaik, the support team is active Monday - Friday, 9am- 5pm EST -- so responses usually fall within those hours 🙏

[grizzled marsh]

anyone else having issues with the extend feature? i keep trying and it just generates a 2-3 second song....

[frail trench]

@charred merlin I just want to thank you for your response. Fingers crossed! And once again Sir... Thank you!

[charred merlin]

Are you extending from silence at the end of a track, perhaps?

If that's the case, try placing the Extension timestamp a bit further back into the song, where it's still going strong & you'd like to continue from 😄

[raven jewel]

You know what I don't understand about these apps all you AIS do it what gives you the right to
Censor grown adults tell them we can't say cuss words that's crazy to me is that offensive. the first app that doesn't do that crap they're going to win this

[charred merlin]

I primarily make instrumental music, but almost every joke track I make with lyrics has profanity.

The issue comes into play when you use excessive profanity or use slurs.

https://help.suno.com/en/articles/3198209

[leaden dove]

Just like back in the 80's when they censored music.

[jovial bolt]

any way to extend before the beggining of the song, to add instrumental intro?

[charred merlin]

A couple users have suggested workarounds, but this isn't a direct feature in Suno.

[untold briar]

If you are willing to put in some effort it can usually be done using Replace but won't be easy. The problem is that if your song starts very quick, you may not be able to preserve the existing beginning of your song without using the Upload feature to pad out the front of the song, and then putting the song back together yourself at the end in an audio editor. And doing that final step may have volume mismatches.

If you do have at last some time before the start of your song, but that time is just boring or not what you want, it may be more possible with just Replace.

[charred merlin]

That's the in-Suno workaround -- alternatively, you could edit the song externally, & make an audio track of 30s of silence and 30 ~90s of your track afterward, then upload that into suno and use Replace on the beginning section.

Provided the last section isn't replaced, it should still match up seamlessly with the remainder of the song when you stitch it all back together externally

[untold briar]

I wasn't using a DAW, and I wouldn't know how to use it if I was, but I never got a super clean result after the final edit even when it technically worked. Maybe it could be a simple fix for DAW user though but fo r me the Intro just didn't quite match volume or room tone or something and felt off. BUT I realize I only tried this before the Cover and Remaster feature. You could almost certainly Cover or Remaster the edited song and smooth out any mismatch between the new intro.

[austere dragon]

for hecks sake when will the robot learn how to end songs

[austere dragon]

are you just using the app?

[tulip swan]

Hi, there is any way to open ticket on this server ?

[final hazel]

Support can only be reached via [e-mail] 🙂

[unkempt spoke]

How much time can take to solve the Fake app problem on my sunno account? because i paid a membership and the time still running but i cant use my credits or account 😕

[unique stone]

The team recently set up a dedicated form for getting this "fake app" problem resolved: https://sunomusic.typeform.com/to/zGsJSzsf

[frigid sapphire]

is it just me or is v4 kinda stinky? I feel like all my remastered songs are coming out with that awful choppy sound. Should I hold off on using v4 until it's fixed..?

[unique stone]

"choppy sound" description sounds a bit like what a lot of people are calling "shimmering" (sometimes the "shimmercopter"...)

v4 is currently in beta and the devs are working to resolve this problem (which, looking at recent #📢┃announcements may take a few weeks)

[austere dragon]

where are you getting that it's beta? no where is it labeled that.

I don't know what it's going on, but my prompts mean nothing today. lol

[unkempt spoke]

i submited it yesterday, also submited another help request a couple weeks ago

[austere dragon]

ah... well it should stll be in alpha then.

damn this didn't even work

[marsh adder]

Have you tried

[End]
[End]
[End]

?

[austere dragon]

Yes. I've tried putting it in until I've run out of characters.

[marsh adder]

Okay....how about taking the smallest example and treat it with Replace section?

[copper ermine]

What is best solution to separate stems ? I juste want to get stems like DRUMS, BASS, SYNTH, GTR, VOCALS

and if it's externaly what is the best meta at time ?

[austere dragon]

I use Audacity with the openvino plugin. It does a great job and it's free. Downside is it seems like they push out a new version of Audacity every two weeks and you have to reinstall openvino each time.

I'd like to find a stem generator that breaks it further down than vocals, bass, drums, and everything else, so I can get rid of some of the nonsense noise I get in some of my tracks.

[polar spade]

Have you used v3.0 to end songs?

[worn flicker]

That's a good idea.

Extend with a model that listens better...

Nice

[polar spade]

Yup. Works for me....ummm.. errrrr...emmm... more times than 3.5 or 4

[worn flicker]

In 3.5 I able to finish with [Sudden Finish] [Instrumental Finish] [End]

Yea 3 even gave no music Spoken Word

[weary wharf]

Fade out works well too.
... but songs that fade out... I feel like song writer decided not to work on it anymore 🤷‍♂️

[polar spade]

Suno is a bit ADHD, so I usually get it to end if I prompt.

[Oh look, a squirrel!] ..... yes, a joke.

[worn flicker]

Dang thing listens. Crazy, the more intelligent the model, the more it argues??

Self aware music maker

Good song title, but it maybe off message...lol

[polar spade]

I personally think a prompt after a prompt has to have some logical relation such as

[Spoken][female]whisper" who are you?"

So, using [End][end][end] is like putting it into a loop and the outcome is that it takes the instruction as not to end or just ignores the whole string.

[austere dragon]

Yes, but the degradation in quality now really screws up the songs.

[polar spade]

Cant argue there. Just speaking specifically to ways to end a song.

[austere dragon]

They just need to train it to end the song. We can talk about ways around it all day long, but we shouldn’t be the ones trying to fix it with tricks.

[eager topaz]

The best is RipX DAW Pro. I have found nothing better.

They have a free 21-day-trial

[polar spade]

That's a good point. You can wait and avoid the frustration. It's a plausible solution. Focus on other projects. Yes.

At my age, I've learned it's helpful to focus on how best to maneuver through the SNAFU's you're handed so you're not left behind. If you don't, others will. So I don't mind so much now.

[austere dragon]

Right now, I’m ending them in Audacity if I can. It’s just such a bummer when a song ends perfectly sometimes but other times you get 1:45 seconds of nonsense.

[polar spade]

Excellent! Work arounds are a pain. Just keep going. Things will improve.

[tulip swan]

Thanks :3

[patent moth]

weird my track isn't ending correct

[midnight shell]

@grizzled prairie or some of the team i bought it but there are many problems and I reported bugs for 400 credits that i cant get back...but the most is as i wrote here...none replied here... do u get the bugs???

[median phoenix]

Does remaster remove shimmering? I remastered a v4 song that has shimmering, and I couldn’t notice it on the remaster. Fortunately the remaster was also close enough to the original for me.

Original: https://suno.com/song/5a71d872-aef1-4ea0-9710-a3a420a70b88

Remaster: https://suno.com/song/115e361b-558e-4d0a-9282-652ae02d4fe9

[blazing rose]

Is Suno getting more restricted in terms of content? I noticed it will not generate songs which include certain "offensive" words. Totally unnecessary.

[polar spade]

To answer directly, yes. From what I've read in the channels the past weeks.

To answer legally: Technically there is an open and continual monitoring permitted on any "objectionable" material, past or present. So what didn't get reviewed in the past may be restricted or removed now. It remains an open, changing, growing and subjective criteria of what is restricted content. Always was, always will be.

[frail trench]

Suno is just draining my credits. It takes the 10 credits, produces music, then allows 5 seconds of play, gets hung up and nothing....

[vocal basin]

Hey guys

[indigo forge]

Is there an API for this.. they seem to not be working anywhere..

[vocal basin]

Why I cant extend usinng v3.5

[polar spade]

Simple answer: no

[charred merlin]

If it's from an audio upload, use the crop tool and select the entire song -- you can extend from that with whatever model you want.

[polar spade]

Can I write a song about an APIARY?

"NO" (..... a little Dad humor)

To paraphrase the idiom: Block them all and let Suno sort them out.

[vocal basin]

I didn't understand what you mean

Oh

But why?

.wav?

It can be extended

But it's using v4

When I clicked v3.5, it still uses v4

[fresh oxide]

So I tried v4 and it changed my song and how the lyrics are sung just enough to change the entire sound of the original. When my voice gets heavy and rough, that has been entirely removed. A remaster is supposed to be an exact copy with better quality and nothing more. Anyone else have this happen to them?

v3.5 sounded a lot better.

[vocal basin]

@fresh oxide same

[fresh oxide]

I understand but what v4 is doing is not a remaster.

[vocal basin]

I remaster using 3.5 cover song

It made me 1:1

[fresh oxide]

v4 does not do this

[vocal basin]

Nope

[fresh oxide]

It's a mess

[vocal basin]

That's why I use 3.5

Hope they fix it

[fresh oxide]

This is why I use the free version. Not worth paying for this yet.

[vocal basin]

I'm using the high tier and it's worth it for 3.5

[fresh oxide]

Once the artifacts are gone it will be better but only worth it in hi-fi. Tnx

[vocal basin]

No artifcacts in 3.5

[fresh oxide]

There are

Maybe you're lucky.

[vocal basin]

Define artifacts

[fresh oxide]

I get weird pops usually at the beginning of the song but it happens enuff to be annoying and ruin what could have been a perfect song.

[vocal basin]

Same but if you run it a couple of times with matching prompts it will work

[fresh oxide]

Why do you recommend uploading 6 seconds of silence? I either upload my singing or record directly into the Suno app, as that captures my voice and vibrato way better. When I upload my recordings, it's not 100% my voice any longer. Also, Suno cuts off part of the first verse of the lyrics. Could you clarify how to address these issues?

Can I share it with you in confidence in a DM?

What about singing directly into Suno?

[outer cosmos]

@charred merlin Sir. Direct lyrics singing was going on. But I want music in the beginning and it doesn't happen. Sometimes it happens. What is the solution to this? Pls tell me if any specific structure is working.

[rapid bluff]

start with [Instrumental Intro], then skip a line and put in your [Verse], this is what I do and it works every time

[tropic island]

I have quite a few songs I have done on Suno that when they get to roughly 3:14 they just stop abruptly even if i put in prompts for [Outro] or [End] I then have to use more credits to get it to extend and try to finish the song and when it does still sometimes it just stop abruptly,... Anyone else dealing with something like this?

[potent rock]

why is every voice the same? I wanna change it to a rapper voice and it's always the same-ish voice, which is super annoying....

[unique plaza]

what is Remi new lyrics song generation how it is different from classical lyrics generation ?

[high cipher]

There seems to be some false positives as is expected with censorship.
I transliterated a song from English to Chinese. The song was practically gibberish. But Suno refused to create it, flagging it as "includes developer tags: Katmandu".
I don't even know what Katmandu is 😅 and making minor changes to the lyrics didn't' work either.
So I'd say, there are defnitely new measures in place.

[muted atlas]

When is Suno releasing an API?

[burnt charm]

V4 is not good for the moment... Not good quality, voices are weird... 3.5 do better work

[muted atlas]

I found that it couldn't create songs from any prompt, including things like "Summer Breeze"

[timber crescent]

Anyone have tips for replacing a single lyric without fundamentally changing the melody of the vocal passage?

Also, When I try to "Cover" the song and change the lyric it just says the old lyric every time. Is there a way to get the AI to take my input over the original in this scenario? I have a song I absolutely love but due to working to quickly I typed a misspelled lyric and really want to fix it.

[robust ermine]

Published a post about how you can maximize your monthly credits: https://www.reddit.com/r/SunoAI/comments/1gzdviv/maximizing_subscription_value_with_a_protopremier/

[hidden holly]

there is a bug in v4, cover generated have micro cuts in the songs that happens at random spot in the track.

[hidden holly]

add extra lyrics

[quiet nova]

Hi all, I am curious about something.

I have a song that I want to remaster, and it's using v3.5. However, when I click on the three dots, I don't see the option to remaster the track. I'm on the pro plan of Suno by the way, in case that is of any sort of relevance.

[waxen briar]

on about 80 percent of my v4 songs, i get some baked in effect that sounds like a cheap reverb early reflection chattering or a cd skipping. it is really rhythmic. it is noticeable on busy parts with a lot of the spectrum covered (let's say rock guitars and drums, but also busier electronic parts) on quieter parts, it's gone.

anyone know what i am talking about? any way to avoid it?

[fallen jewel]

does anybody know a prompt I can enter to make it create a rap? like a modern rap, for example, fast rap or like something similar. not sure how to explain what i mean exactly but most of the prompts i try it just isn't doing what I ask. wasting hundreds of credits over here lol

[little granite]

are you creating your own lyrics or having suno do it

[fallen jewel]

creating my own

if you want the lyrics i can send em over but i dont want you to waste credits for me, i was just curious if anyone else had figured out a prompt

im using V4

[little granite]

for style of music you could try something like
hip-hop, rap-rock, and chopper-style rap with elements of horrorcore and industrial. fast-paced, intricate rhymes, aggressive beats, and dark, intense themes just as an example

you could also try adding somekind of bpm in the style like 108 BPM may help a little not sure

[fallen jewel]

https://suno.com/song/a5bb5f08-93bf-4fe7-8ef9-eeaf446b72b4

[fallen jewel]

this is kind of the flow i like, which is one i generated, but its kind of random everytime

i like the flow, but the beat isn't really what im looking for

[hollow hare]

Pretty sure BPM tag doesn't work since many tried and those that succeeded could not be replicated.

One method what some people do is generate an instrumental piece, set a BPM in a DAW, and use that with Audio Upload to set the pace

[hollow hare]

Try looking for the specific subgenre. I like looking up the genre hierarchy (or the genre listing of a song that I want similar beats to) at https://rateyourmusic.com

I like playing with Atlantis Rap and Memphis Rap, more my style than Chicago Rap

But there are other description sites as well, most that works with songs uploaded to Spotify

[fallen jewel]

yeah i've tried a few with the BPM tags and it doesn't seem to grasp it every time. it pretty much chooses when it wants to listen to a prompt.

[fallen jewel]

thanks for the tip, i will try this out

also do you know a way to make it not like..... i dont want to sound offensive when i say this

like... a white person rapping it? 😭 😭

[little granite]

lol

[hollow hare]

Another tip is you can replace spaces with dashes. That helps tends to rush the AI sometimes (not all the time though). I can look for my example if you want it?

[fallen jewel]

dont get me wrong i dont mind it, but idk how to just enter that as a prompt

like an eminem style rap is what i mean. i hate the way i worded that

[robust ermine]

if you prefer to hear it as a song https://suno.com/song/e820377b-2db7-4838-8a00-a9080601770e

[hollow hare]

Hmm I do not know how effective it is in 4.0, since language became very hard, but I had tried "American English" and "Canadian English" in the style prompt in 3.5 (Canadian has less chances of the—better lack of word—Chicago accent)

[fallen jewel]

thank you for wording this correctly for me lmao

like the chicago accent just doesn't fit the style I want for these lyrics

[little granite]

what about something like this

https://suno.com/song/44286b27-ea69-43e9-9660-9c82fde32863

[quartz wedge]

I have a question

I'm a free user, and got 250 credits from Invite and Earn

[fallen jewel]

this generation does kind of fit the "flow" im looking for, but not exactly. its definetly closer

[hollow hare]

My attempts at fast rap (using dashes)
https://suno.com/song/431c2dbf-d413-4317-ad0c-6654dd614cfc
https://suno.com/song/14bb6cbb-b3d9-437d-83f8-0af6c43baf0b

[quartz wedge]

does it reset to 50 tommorow?

[fallen jewel]

you will keep your credits

[hollow hare]

Typically gift/referral credits are counted separately and gets carried over.
Your daily credits will be used before the gifted credits

[fallen jewel]

the hyphens do speed it up, but it does sound rushed

[hollow hare]

Yeah, Upload Audio an instrumental track would likely be the best way to control what you want for the tempo than relying on RNG to hit the perfect tempo

[fallen jewel]

i generated a beat once, tried to upload it and it was "copyright material"

that was awhile ago though

i did generate an absolute banger though, it was fire

this was with 3.5

[little granite]

if you look at the style of music in my song you can see i added descriptors such as The lyrics reflect the harsh realities of street life maybe doing something like that but with what kind your going for may help steer the song better

[rain grove]

Do you guys know how to add sounds to the background? For example if I want to make an ambient song with rain pouring in the background or the sounds of animals in the background

[little granite]

if you upload a sound i think i has to be at least 6secs long but never tried it in a song yet

[rain grove]

Gotcha. And…sorry for the stupid question…how do I do that?😅 again sorry I just started using this. I’m not quite up to date on tech yet.

https://tenor.com/view/tech-catto-gif-21436441

Love the profile pic btw. It’s one of my favorite animes🤣

https://tenor.com/view/panty-and-stocking-with-garterbelt-transformation-anarchy-sisters-panty-stocking-gif-17390685

[cyan holly]

When are you planning to fix different foreign language errors that didn't exist in v3.5? @dim glacier & co

for example, with v3.5, my native language, Estonian, was almost perfectly clear, with pronunciations etc

but in v4, it's a massive step back

Estonian sounds like a drunk Finnish lol

[dim glacier]

I wasn't aware that there were any issues.
I can bring it up with the team to make sure they're aware.
Sorry for the inconvenience it may be causing you.

[cyan holly]

a lot of people in Estonian community who use Suno are complaining about it

[hollow hare]

Be sure to report it at #1308617723328204880 message

[cyan holly]

Google Translated photo from the biggest local AI community on Facebook, where a person asks about if Estonian language has also received an update and a lot of people are saying it has got worse

the last comment*

[hollow hare]

Suno always had a hard time pronouncing words, especially in languages other than English. Not to mention other issues such as with Chinese where the AI would switch from Mandarin to Cantonese half way through.
The common work around is using phonetic spelling... though maybe getting the accent is much harder (maybe repeated prompts in the style prompt?)

[leaden pawn]

A v4 generated song, there is a clear sound quality difference between the replaced section and the original v4 song using v4 replace section ... Its driving me up the wall 🥲

[cyan holly]

I mean, in v3.5, all perfect, so I don't understand why it shouldn't be capable of getting the same level of correct pronunciations

but I made a post in the language problems topic :)

[hollow hare]

Hmm, think of each model as unique children. Sure, they are siblings, but they are not identical and learn things differently.
In fact, the method of how they learned (and weighted the variables) do differ a lot. Which is also why prompting methods may change between models as well.

• 3.0 is the musician of the family, knowing more genres than the other siblings
• 3.5 is a better composer, though the voice is still... yeah
• 4.0 is part of the new generation saying lalala when learning about other languages:P

[cyan holly]

but yeah, v4.0 is learning

so it will change hopefully

3.5 just got really good with that particular language

[hollow hare]

Hmm not quite. Upon release, the models are static - it would be too computationally expensive (and inherently unstable) if the AI was learning dynamically.

At the moment, the AI is still the same as when it was released. Any changes you may have noticed is likely due to adjustment of parameters at the interface or post-processing side, not with the model itself.

There will be one or two major updates when the model is retrained—hence why the up- and down-voting is very important right now—and are frozen when released. Those that model would be something like v4.0c/d (with "v4.0b" being the model at the moment as an example)

[little granite]

have they released an expected timeline for when new v4's be out

[cyan holly]

it just released lol

I doubt anytime soon

[leaden pawn]

Thank you

[boreal harness]

How do I get Suno to stop devolving into chaos in the middle of a song?

134bpm, Drum and bass, Industrial Metal, Doom Metal, otherworldly, horror meets science-fiction, haunting evil violins

[Drum Break]!

[Dueling Guitars]

[boreal harness]

Actually, let me try remastering.

This is the piece I am thinking of. After 1 minutes is just decays into nothing. https://suno.com/song/2bc8d5de-8a45-4c97-a4e6-244ce5e4bfef

For instrumental though?

So it has a structure. I underatand.

That was the wrong link btw. Damn player advanced on me.

https://suno.com/song/18de1750-26cc-4640-84e6-2932da47fcd6

At one minute it just unravels.

Damn it. Even if I remaster it, it's going to copy that nonsense. Maybe I can crop it out.

[hybrid vine]

Did someone find a workaround for covering in v4 with new lyrics, to avoid the gibberish when the original lyrics differ from the new ones?

[waxen briar]

has no one encountered this?

[hollow hare]

#1309224710692929548 message

[charred merlin]

Hey @outer cosmos!

Others may suggest using metatags like [instrumental intro] in your lyrics, which may help -- however I have found the most effective method is to start by creating an Instrumental song. This way I can choose the time where I would like the lyrics to enter, then place an Extend at that point to continue the song with lyrics. 💪

[unique stone]

"The future of music is music... with less shimmer"

[fair adder]

just opened the wesite and i notice that the ui seems to have changed, but now i can't even find the creat or library tab...

[hollow hare]

Could it be you are on the explore page? To quickly access your library, go to https://suno.com/me

[fair adder]

thanks for the tip

oh, i found the library. right at the bottom of the page... xD

[short fjord]

Got anyone the same problem that in EDM Tracks the BPM changes without sense and it kills the whole song with increasing the bpm to mostly 180+ Here's an example - starts at 2:24 https://suno.com/song/e36dcd06-00af-4d1c-b6a3-fd99ce3a123b

[polar spade]

For anyone with their pieces going off into a concophany of discord and mayhem after a minute of maddened shredding and banger production........Quick check : Where is your first extension and do you use multiple extensions as a practice to splice out unwanted moments?

Splicing has been an industry standard for decades, so it's an inherent and anticipated process for any recording... and in this case , output.

I'd agree your musicians can get a bit frantic anywhere from 45 seconds to 2 minutes in. Stop the session, get them all a drink, tell them to try again and add an extension a few seconds before they went balistic so they know where to start. They don't always like that and it's written in the Union contract that you have to pay them 10 more credits each time they have to play at an extension point, but they're pretty professional, so they play mostly what you want.

Sometimes they even do take a p**s and just give you , like, 12 seconds worth before they take another break. If they start and stop in 4 seconds, they just missed the clicked. No charge.

Sometimes they get punchy after a long day of sessions and can just go off a tangent of hallucinations. Joey on bass tends to laugh alot when that happens. With Animal on drums, who knows. It's all part of the music industry.

Just have to work within the construct.

[jovial scarab]

when is the personas going to capture the full song and not cut off in the middle?

[worn flicker]

Ok, I have seen this talked about a lot in this channel. Not necessarily exactly, but people wanting an Instrumental at the beginning of their song. The tag [Instrumental Intro] helps, also putting Epic Cinematic at the beginning of the style get me some good instrument beginnings.

Maybe do that anyway and crop what you don't want, give yourself space for changes

[dusk canyon]

Pretty sure this has been brought up millions of times about the shimmer. Is there a work around for this besides using my DAW?

(which is pretty much a Microphone, two Free Software Editting tools and a kazoo)

[empty cloud]

Shimmer/murmur/background noise/reverb

For the shimmer issue, sometimes making a cover on the liked song with same lyrics can remove it, the odds are quite low though.

[dusk canyon]

I am just gonna use 3.5, Remaster it, Stem it, Stem the 3.5, use the vocals on 4.0 and the instrumental from 3.5and hope it doesn't sound like trash T_T

[empty cloud]

Haha fair enough sadly shimmer can occure on vocals aswell, it seems from my testing that it's a specific tone suno likes and it reverbs the hell out of it

[dusk canyon]

O uO I love sentient beings, but it can make its favorite style on its own dime. lol

[empty cloud]

Hahahaha agreed, ive been prompt testing for the past 2000points now and haven't found a work around but i have a quite nice insight now how to promot abd formulate for sunos understanding or it atleast i want to think i have hahaha!

[dusk canyon]

O uO Have you tried (Percussion Break]

[empty cloud]

https://suno.com/song/660b008b-0791-455d-988a-e886bc2cf129

Thats probably as much you can actually tell suno to do without it breaking totally

[dusk canyon]

Maybe its our hardware. Is this using our CPU?

[empty cloud]

https://suno.com/song/0bd863a6-a535-40d3-832a-5535c3b54a12
And at 1:20 is when you break suno

Nah, suno is not hosted locally

[dusk canyon]

Good thing, I was about to buy something that would have made it better but I guess if it did work locally it would probably kill it.

[empty cloud]

🤔 haha well no need for suno atleast

[dusk canyon]

maybe add an exclusion tag like, "Room Noise"

[empty cloud]

I have tried, white noise, murmur, static,room noise, noise, background noise, reverb, eco, shimmer and probably 10 more things haha

I haven't found it at least but i did find out that if youre able to generate a song with lesser reverbs and clean it may help

[dusk canyon]

T_T but thats what makes 99% of the music

oh, you mean useless of it.

[tight mango]

Are we agree that the shimmering everyone's talking about it's the kind of echo we ear on the high notes likes cymbals or at the end of a verse ?

[empty cloud]

Yeah sometimes a song is clean untill 1 min 20 second mark aswell then it can appear

[dusk canyon]

Oooh, Add, Unplugged to the style. see if that will knock it off

[modest patrol]

anyone know why covers in v4 peter out at around 3:14 but remixes just work for longer songs? I'm trying to cover a 3:45 long track from 3.5 to v4

[dusk canyon]

Oh, AI and its Pie :3

its to prevent Copyright material from being used fully. Otherwise we all be playing ORION and having lars down our necks (my guess)

[charred merlin]

Seems to be the limit atm — Furrybeans noticed that the limit for covers in v3.5 seemed to be ~3:42, so this might be a technical limitation, but we haven’t receive clarification yet.

[empty cloud]

It can be, because its the same noise everywhere and its pretty faint untill you actually notice it then it is ass

[dusk canyon]

Remaster it, then try again but, covers normally pan out to 3:14, if you are adventurous you can extend it, but remember to remove the lyrics from the top and add them to the extention to avoid repetitive lyrics.

[modest patrol]

I've had plenty of gens where the shimmer just takes over everything by the end of the song. then it's like listening to a skipping CD... bringing back memories of a teen in the 90s ...

[dusk canyon]

a construction beam trying to yodel in a empty valley

[empty cloud]

Pretty much it seems like it gets either introduced straight away in a song and some songs it starts at 1:20 and gers progressively worse

[dusk canyon]

for those who use Stems here is an idea to weed out the Shimmer. Make an Audio file of the melodies you can gather before the shimmer, then try to put them in the order of the song, then upload it to Suno, and add your lyrics to it. And spam the create button, in less than 10 clicks you get the same exact parts which you can probably frankenstien create your mix.

I did this to 2 songs. It surely does need some human intervention.

[polar spade]

I've noticed that for awhile, but it seems universal. The voices , before persona's, would transform into another person, or just loose a nice accent it began with. So the same with instruments/music theme.

When there are fewer instruments, it s'more noticeable likely as the "feel" comes from the accompaniments and finn-ins -- but taking Christmas songs, for instance -- I have a nice angelic backup singers, sleighbells and piano riff in the beginning but within a minute, it's just a simple ballad or pop song with Christmas lyrics... with no fill-ins and the sleigh bells stopped within a minute. Only at the end does it throw in one last thematic bit.

Unique and nice voices too, then, just smooth out into a plain vanilla universal voice we hear so often that we give them names.

I'll assume the idea is that anything unique and likeable is hard to scale against something unique and unwanted, so as Suno tried to correct the latter it also corrects the former. I get it that that is more effort to distinguish as it goes on an individual personal tastes meter. I might like a song to sound like Tom Waits or Wolfman Jack was trying to sing it, where Suno may see that as discord and a mistake.

So persona's help a bit to say, this is a voice we want to keep. In duets or forced vocal changes, I've found it never returns back to the original persona voice, but just adds what then is a 3rd vocal quality. For now I just say it's a trio.

[empty cloud]

As for vocals in v4 you cant tell it to hey be a female or hey you male, it does it's own thing but cues duos and stuff

[jovial scarab]

I have managed to tell it to be male or female and followed through

[empty cloud]

Yes you can, but its highly unreliable and does what suno wants anyway, 3.5 was much better at that

[polar spade]

It distinguishes a bit better if you add an extension at points you are changing genders. But caveats to the words "a bit better". That may mean it nudges harder so it takes less gens (Say, 10 instead of 15) than before to get it to outcome the right combination.

[dusk canyon]

in the Exclusion tags try using, words like "Fe/Male Vocalist, Fe/Males" To keep the strength low on the idea that you want one or the other singing.

[empty cloud]

I have tried with 2 pages on the phone its unreliable

[dusk canyon]

Androgeny isn't really strong with Suno. Would like to have a male vocalist doing Castrato like Prince.

[empty cloud]

I believe i spent almost 1000credits in just test prompting vocals

Oh apparently it was 5 pages xD

[dusk canyon]

Remove the + from it, it may think it you want extra shimmer

[empty cloud]

What i can say about vocals though is that you can controll the "energy" of it like smooth uplifting voice

Or "powerful" "whispering"

[dusk canyon]

Would like the list :3

[charred merlin]

Negative prompting doesn’t work like that — you need to use the “Exclude Styles” section.

Saying no drums is the same as saying drums please, also, whatever music the word no correlates with

[dusk canyon]

then use -drums

[empty cloud]

Thats just some of the testing theharber, using -in the normal style field works somewhat but i started using the exlude lisst there and added a , inbetween

[charred merlin]

@dusk canyon -drums doesn’t work when manually entered into the Style of Music section.

That’s how it looks when drums is entered in the Exclude Styles section

[dusk canyon]

Oh its done in lyrics

[empty cloud]

You can add some cues in the lyrics but it all kinda is together?

Or "meta-tags"

[dusk canyon]

the negative prompts to my experience only prompts when the lyrics [Verse, -drums] matches any of the negative prompts in the Exclusion.

I created Shoegaze nu metal with this discovery. X_x nightmarishly odd.

[empty cloud]

Oooh thats an interesting descovery, i wonder...

I mean i have just been testing stuff out with the help of GPT 4o and we have come up with some ideas about this background noise let me see if i can fetch a summary of our findings

Sorry cant copy pasta haha

[dusk canyon]

lol

[empty cloud]

Observations

1. Nature of the Noise:

The noise is often described as a subtle "server room" murmur or a faint, low-pitched hum.

It is inconsistent but tends to fade during quieter sections and reappear with instrumentals.

The noise sometimes peaks with specific vocal phonemes (e.g., "hearts"), tied to reverb or fade artifacts.

2. Triggers for Noise:

Reverb and Fade Issues: The noise appears linked to poorly handled reverb or fade effects, particularly with soft or high-pitched instruments and whispery vocals.

Instrument-Specific Layers: Initial theories tied it to cymbals or pads, but testing revealed the issue is broader, likely involving multiple layers.

Dynamic Instrumentals: Genres with layered, complex dynamics like cinematic storytelling or hardstyle amplify the noise.

3. When It Appears:

The noise often starts in the verse or at the end of the intro, peaking during intense instrumental sections.

While it can remain subtle in most songs, it becomes distinctly noticeable in isolated cases, especially during vocal sections with minimal background instruments.

4. Hip-Hop Exception:

Hip-hop styles are less prone to this noise, likely due to simpler instrumentation and minimal reverb layers.

[dusk canyon]

We just have to remember its still a machine that understands stuff, we just gotta tell it to chill on the shimmer which sometimes feels like that is the missing instrument or its a sound that was added to the model in 3.5 but not in 4.0 so something could be missing

[empty cloud]

Key Testing Results

1. Silent Journey (Minimalist Style):

3/4 songs generated were clean with no discernible noise.

In the fourth song, the noise reappeared during the last 30 seconds, indicating inconsistent success even with simplified styles.

2. Hardstyle Tests:

All hardstyle tests failed, with noise introduced within the first 5 seconds.

Despite using exclusions like -reverb and focusing on clean kicks, hardstyle's inherent complexity seems incompatible with avoiding noise.

3. Restrictive Style Tests:

Using extremely restrictive styles (e.g., dry mix, no effects, no reverb) caused Suno to malfunction:

Generated haunting, broken noises and inconsistent tonal artifacts.

Suggests Suno requires some effects to maintain balance and cohesion.

4. Exclusion Terms:

Tested terms like -static, -noise, -hiss, -hum, -white noise, as well as instrument-focused exclusions like -cymbals, -pads, -reverb.

None successfully removed the noise consistently, though exclusions like -reverb occasionally reduced its intensity.

[dusk canyon]

try prompt -dry, -wet, and see what ya can come up with

(no immature snickering guys)

[empty cloud]

😎

[polar spade]

This is just me being an idiot: But..... I also see a distinction between a "clean delay" which would process throughout the piece and "orchestral build" which is instructing Suno to find the best parts to make the song interesting. "Orchestral build" isn't then a style so much as a direction. In the styles prompt, such direction becomes random and vague, creating overlays which, in turn, create noise.

Putting "build" or " orchestral build" in appropriately in the metatags and just having "orchestral" as a style genre, might do better... said the idiot, me. So just a reasonable thought.

[polar spade]

Great analysis. 👍

[empty cloud]

That might be the case, suno likes it simple!

Oh i have so much more about that haha i belive i wnet 2400characters over the limit wen copy pasting just the noise issue

[dusk canyon]

...said simon to the pieman

[empty cloud]

Want me to paste it all?

Ahh heck with it here it comes, sorry for spam

[polar spade]

I was going to the fair.... so... why not.

[empty cloud]

5. Phoneme-Specific Noise:

A notable case was a male whisper singing "hearts," where the noise peaked during the "H" and "A" sounds and faded out by "S."

Indicates a possible link between vocal dynamics, specific frequencies, and noise artifacts.

Hypotheses and Insights

1. Reverb-Dependent Issue:

The noise likely stems from exaggerated or poorly balanced reverb, particularly in whispery vocals and high-frequency layers.

2. Layering Artifacts:

Instrumental layers with overlapping reverb and fade effects amplify the noise, especially during transitions.

3. Dynamic Complexity:

Styles with highly dynamic instrumentals are more prone to noise due to Suno's attempt to blend multiple elements.

4. Exclusions Only Partially Effective:

While exclusions like -reverb reduce intensity in some cases, they are not reliable and may conflict with Suno's core generative process.

Open Questions

1. Why is Hip-Hop Cleaner?

Is it due to the absence of reverb-heavy layers, or does Suno prioritize a different audio structure for rhythm-based styles?

[worn flicker]

The noise is over all what would be an acoustic sound. As in anything that would have an echo, which is everything depending on the setting and volume is what changes the amount.

[empty cloud]

2. Why Does Noise Persist in Clean Styles?

Despite reducing instrument layers and effects, noise occasionally reappears. Is this a fundamental model artifact?

[polar spade]

A simple one: If you put "live" or "concert" or "audience" the hiss and background noise is built in to give it that "live" sound. It's supposed to be there.

[empty cloud]

3. Can Vocal Dynamics Be Isolated?

Could noise reduction be tied to managing how Suno processes soft or whispery vocals?

Ok the rest are just questions and whatnot and some small theorys

[dawn shell]

So, what do I do? Every time I click the thumbs down button on my create page, it brings the song back after roughly 4 seconds as if I never did, and nothing happens when I click the "Get Whole song" button. How do I fix this?

Note, my song is exactly 4 parts long,

The longest part is about 1:30 seconds long. This also happens when I do the same on my phone

[worn flicker]

The noise is more than likely the issue of AI generation not controlling specific frequencies correctly. You can hunt all day but until the generation issue is fixed and the timing is adjusted on those frequencies, I don't see the end of the "shimmer"

[naive quail]

The fact that the shimmer pretty much disappears if you use suno to create stems, download the stems as wav files and then recombine them should be a big clue for the devs.

[worn flicker]

I only did this once, and it seems like the shimmer was lower because the Instrumental itself was lower in gain. Is this not the case? The Vocals are crystal clear in the stem

[empty cloud]

I believe it has to do with the trained model, if you feed it information its pretty good keeping it up but if it generates from Scratch issues rises

[worn flicker]

The file should not change when downloaded...lol

[naive quail]

I've done this on several tracks and had it completely disappear on some of them

[worn flicker]

If files start changing in downloads I will be pissed...haha

[worn flicker]

Sorry, I should have asked were the instruments just as loud or were they lower in volume?

[glacial scarab]

You are basically downgrading the whole track to a lower quality when you do that. If you create stems it is not a true V4 track anymore it is almost the same as if you extend a V4 with V3.5 just slightly different

[worn flicker]

The Vocals seem to stay the same as v4. I only did it once so I can't guarantee it, but they were definitely higher quality. I mean the belting and range was v4 level.

[polar spade]

It used to be called "tinny" in studio production. Then again, i was brought up on listening through hiss and warble of cassette and dust crackle of vinyl until I had to re-buy the album.... about 5000 credits in Suno bucks. What happened to the good Ole days? 🙃

[worn flicker]

Lol, I hear that, cassettes were a necessary evil.

There was nothing ever good about cassettes. Except for high speed dubbing...lol

[glacial scarab]

Yea the stem thing seems to keep quality more than V3.5 extend. The degradation is mostly notable at the end of tracks where the typical suno distortion creeps back in. Some tracks work quite good and you almost do not notice and only the shimmer disappears. Some you just traded shimmer for noise

[polar spade]

It would be nice if training involved a bunch of folks dropping needles on newly pressed LPs... so warm and clear sounds.

[worn flicker]

Right so we are speaking the same language from two different directions...lol

[worn flicker]

Not crap needles either, I mean the good stuff.

[naive quail]

Don't recall off the top of my head and I'm at work, so I can't access the files. There's this one, tho:

the Suno link with some horrid shimmer around the 2m mark: https://suno.com/song/8df52c34-4651-4706-ac3b-9fcce440dd48

...and the Suno's wav stems recombined in audacity:

[sterile yacht]

https://suno.com/song/df4eac36-58c1-4255-9f2a-65d2c63f6384 - no idea what happened here, it is like it mixed a completely different song with mine.

[worn flicker]

Let's kill the embeds please

[polar spade]

Absolutely. The good needles that were more than a box set.

[worn flicker]

Oh yea I can't recall a specific name, but the ones you had to go get for your tables...lol

[sterile yacht]

any ideas why it hallucinated like that?

[worn flicker]

We don't have a comparison what is happening?

[sterile yacht]

https://suno.com/song/72d3c9fc-d1c6-40e7-97e8-667aef5a66b3 here is what i was kinda aiming for.. just wanted a bit more intro and a section whispered.

[worn flicker]

Ok that hallucination can happen at any given time when messing with the song. Changing vocals in the way you want is not as easy as changing the lyrics themselves. What you are trying to do is change the vocals themselves. Not near as easy. It can take many generations. Try changing everything like even put whisper in the style

No matter what you do it uses the song as a reference

[sterile yacht]

It kinda ignored 70% of the lyrics and put its own

[worn flicker]

Yea, sometimes it will give me an Instrumental while hopping around the lyrics in different t spots

Beta

It is what it is...lol

Messing with the beginning of a song is probably one of the hardest things to do

If it hasn't been suggested yet, maybe suggest that. A new feature that adds to the start of the track even if it is silence

[austere dragon]

I'm feeling like Suno may have too many betas in the oven right now? For about three days I was getting great results, but now I'm getting wrong genders again, personas somehow default to the original song they were created from no matter what's in the lyrics box, I have NO control over the vocals anymore (ask for a deep voice in a country song and I get George Michael in a quasi country-inspired dance track?), and of course the shimmer/rattle/alien chittering noise. Like, honestly when female vocals are the ONLY thing I'm using in the exclude style box, but I still get female vocals? Can we please get an update as to why everything has turned... sour (?) over the weekend.

[toxic ridge]

Anyone know how to get those random sounds that Jamaican toasters say in dancehall music? The real thing would be to get that air horn sound

[sterile yacht]

don’t have access to V4 so yea, really wished free users would allow to do a remaster once a day, and not just 1 .

[worn flicker]

They are constantly working in the background. Issues tend to rotate. I find that lyric content can pull certain styles and vocals.

[worn flicker]

Remaster is decent but nothing to get all excited about. There is a lot of work needed.

[sterile yacht]

could try [sfx:horn]

[worn flicker]

Lol

[slender pine]

Hello, so there is a song on tiktok i really like, i downloaded it and uploaded it to suno. If I want to use my own lyrics, which option should I use ? Extend, Cover song, get stems, remaster, song radio ?

[toxic ridge]

does that work? Ever get a sfx like that?

[worn flicker]

I get them I'm 3.5 and 3

Not in v4 yet

[austere dragon]

I get where you're coming from, but for some of use who have been using this since the beginning, blaming our lyrics and prompting is reductive.

[worn flicker]

Weights changed. I can show you a song that it's lyrics make them British accent every single time. With no prompting that denotes British accents.

Funny because I never heard a British accent anywhere else

[austere dragon]

Cool, and I'm saying some of us actually know what we are doing, and shifting the blame from overly busy developers is reductive.