📚┃suno-school

[final hazel]

When you hit create.. yes.. but not hitting the "remix" button self

[lyric sonnet]

Btw, is 'chirp' the same as the suno model as on https://app.suno.ai/ ? Then why is it called chirp here and suno there?

[broken lotus]

they rebranded, basically

[final hazel]

Chirp is the same yes. It started as that here in Discord... you can still see som chirp channels where you can create songs too. But that's V1... WebApp is V2 and V3

[finite latch]

Hi, what's bark, how's it different from chirp?

[final hazel]

Suno started with Bark.. which is text to speech. But it is open source now and no longer here on Suno discord (might come back, Devs are not really sure yet).

[finite latch]

Makes sense. Thank you.

Technical question: I made a song with Suno, but now I want the instrumental of that song, so I can re-sing it myself. I tried using an audio splitter, but it's super low quality. Are the original instrumentals saved somewhere?

[final hazel]

No.. you get the best outcome now. It's not that Suno give you a compressed version. This is it for now. They are working on getting better quality in next versions.

[lyric sonnet]

If you can relate to the developers, stem splitting would be a game changer. But I assume they know that.

[flint sparrow]

I was happy to come across this app, it's exactly what I'm looking for, so I signed up for a years subscription. One thing I'm struggling with is the moderation. For the life of me, I can't figure out what it doesn't like about the lyrics. Some sort of feedback about words or phrases it doesn't like would make fixing things easier than just making guesses which still get flagged.

[sonic narwhal]

is there any published paper explaining the basic architecture of Bark or Chirp?

[neat field]

#💡┃suggestions-web message not yet, but there are other methods worth following on open source and research side - VALL-E, musicgen, etc

[neat field]

https://github.com/suno-ai/bark/tree/main see also for raw code

[sonic narwhal]

Thank you very much. I already checked the repository for Bark, but didn't find much information about the architecture other than it was similar to VALL-E. I didn't know musicgen, I will take a look at it, thanks

[spark oriole]

What I do : I click "continue from this song" button for every part of the song, I get numerous "parts 2" and edit them in a Daw. Then, you get something with fewer variations. You can listen to my songs here (https://vimeo.com/showcase/11068089/)

[torpid forge]

¯_(ツ)_/¯

[twin creek]

Hi all!
After reading the discussion above - what I encounter is that when I use "Continue From this song", the model doesn't ignore the original lyrics, but ignores the new ones.
I'm continuing the song from a part in the song that hadn't yet reached lyrics I decided to cut out.
I'm giving it new lyrics and bracket directions in the custom mode section, but it still uses the original ones.
So any suggestions how to solve this would be helpful 🙏

[coral fox]

I joined to say exactly this. I've probably used 100 credits just to get it to say 4 lines and it's refusing to, instead it's saying the old stuff in any old order and giving me huge instrumental breaks

[neat nexus]

Did you by any chance continue from the whole song more than a few times? Doing this too much caused me issues like you're describing in the past which is why I'm asking.

[coral fox]

the first continue from generated a mess, so I've tried again and again and it just gets worse. I would kill for an "edit lyrics" feature that regenerates the same audio with the adjustments

I've tried continuing from a much earlier point in the song (10 seconds) so that it "records over" the old stuff, but that doesn't work

[twin creek]

I just wanna mention that I tried this again now and it did work.
What I did differently this time, was not using the Full Song that was generated to continue from a specific point there,
but to do that with the specific part of the song that was originally generated (Part 2 in my case).

BTW, does anyone know how to instruct for a decoration of an instrument at a certain point?
For example, when I use [bongos break], I don't mean for it to do a bongos solo for a few bars, I just want it do a decoration of like a "bongos drop", I don't really know how to name that specifically..

[verbal pulsar]

Devs are making some kind of update at the moment.
Maybe that generates some issues?

[eager topaz]

Is there a way in Style to turn off the vocal effects? I don't want double tracking or harmony or any of the noisy phasing. I just want a pure, clean vocal.

[final hazel]

It all depends on the genres you are working with. If you do folkmusic you will have less SFX... but EDM is with lot of SFX so you will get with a lot of SFX in general. You can try adding in terms like "minimalistic" but it's hard to avoid sometimes.

[verbal pulsar]

If you mean Chorus and doubled vocals or choir, well, it may be difficult. V3 has this strange tendency...

[eager topaz]

I'm doing different genres, but mainly Country/Country Rock

[eager topaz]

Yes, very strange and annoying! I don't want harmonies etc. I just want straight voice, no reverb, no harmonies.

[river horizon]

anybody knows how you can add suno to private message as a bot?

[broken lotus]

Current thinking is to possibly not mention the vocals, and try style words like ‘warm’ ‘friendly’… seems weird but ‘cold’ got me chorus voices, while ‘warm’ got me the single clear voice…
Of course there was more to my prompt than just ‘cold/warm’ … ymmv

[broken lotus]

The discord bot is v1, much shorter and not as good quality, but just send the bot a message and you can use all the /chirp words to call it.

[tulip granite]

How can we have a intro at the beginning of a song?

[eager topaz]

This usually works for me [Short Instrumental] If you have instruments in the Style then I use: [Short Instrumental, mandolin, guitar]

[lyric sonnet]

Is there usage statistics for suno somewhere? Would be interesting to see how fast it's going ballistic

[lime mauve]

Guys, can I use chirp to control the music tempo? Or is there anyway to get a tempo I wanted?

[gilded hound]

You can use the most viewed song on trending as a proxy. The first song that hit 200k views was only a few days after the first 100k

[soft beacon]

Hello there. Is there any way to download .wav or .flac from suno.ai?

[fallow gull]

Not currently

[soft beacon]

So just mp3? 😢

[final hazel]

Suno is using some pretty neat, new tech that's always evolving. As Suno keeps developing, things are only going to get better and better. Just wait and see!

[nimble quiver]

Awesome, thanks for the resources. Yeah, I’m really excited to see more progress on the open source side as well.

[paper badge]

How to download WAV or Flac files

[heady hearth]

@final hazel I'm waiting. All I'm seeing - should I say hearing - is a fucking mountain of unusable clips of noisy unmusical shit. Not throwing shade on you, you've just got the unenviable voluntary role of answering questions/guiding user engagement with the app and company. But seriously, what is happening? Did someone spill a drink on a device of critical irreplaceable importance. I'm so sad that this is where we're at.😭
https://app.suno.ai/song/7b5c76db-a20f-4fbc-96c8-7b05eb142c2d

[alpine venture]

OK, so I saw a post on reddit talking about how v3 is SO MUCH WORSE than v3a because of the vocal distortion effect caused by the harmonization/chorus bias.

I realized there was a really easy way to prove that this is a case of what amount to mass hysteria, or rather, it always was this way, you just weren't noticing it before because it hadn't been pointed out yet. A similar effect occurs after using high quality LLMs for a long time upon noticing certain particular limitations.

anyway, I put this album out BEFORE v3, it was made entirely on v3a

https://open.spotify.com/album/4x62keQbmpsdZtsyhPrCOJ?si=EXC-DdAYS_KjFZWzfjuHLw

song 1, by 40 seconds the degradation has begun, and it just continues

the second song is sort of edm-ish so you don't notice it

but other ones it's obvious on are

Nuclear holocaust

I think, from talking to people, it's a bigger problem in metal where this sound is undesirable

bc singing with others isn't very manly, and may be perceived as a sign of weakness within that culture, much like dating

(this last bit is just in jest, I know yall be getting it too lol)

in all sweiousness, there IS a bias in v3 toward chorus vocals, I do agree, my point was to illustrate that vocal distortions were already an issue previously, and the changes imo are more horizontal as opposed to a direct quality change

I've noticed there is less autotune in some types of gens which I like, but there is a bias towards chorus for sure

to illustrate what I'm talking about, here is one I have that is particularly clean-vouced from v3a

https://app.suno.ai/song/3d801afd-e215-4d7b-b361-f5ddecd7f8a2

not my fav song, but it's clean

you can hear though that as it goes on, we do get some distortion in the voice

here is one of my more moderately clean v3, non-alpha, gens

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

imo, the vocals are more consistent throughout, and subjectively the sound is at worst rhe same, maybe slightly better, in the sense that it sounds like it is just regular produced indie music

Anyway, I haven't made a concerted effort thus far to recapture that solo high-quality expressive voice so I'll try now to remix the v3a one in v3 and see what I can achieve with prompt tweaking

OK, so most clarity in voice thus far (I hate the style but I am working for clarity solely)

https://app.suno.ai/song/35ec7eb1-c564-41d8-8546-684c00196edc

it's actually quite clear

went through about 600 credits to get here btw lol, I had pretty good clarity earlier but I wanted it crystal

here it is with the original prompt totally unchanged switched to v3 (I much prefer this version, and could def get it better with continued gens)

https://app.suno.ai/song/f721aea3-ede8-471b-bbdc-1ee7f91d383a

but it's got the "chorus" hint people are talking about, and the goal was to eliminate that entirely

[verbal pulsar]

[start]

[Instrumental interlude]

[inner wagon]

I have a question. I generated this song which has a really nice melody that is ruined by I think is supposed to be nature sounds, but just sounds like bad noise. Is there a way in Suno to regenerate a song by keeping the same melody but remove the background insturment?

[scarlet mica]

Is it just me, or are the vocals for all songs with "metal" in the style extremely distorted and robotic?

[coarse rampart]

hi

https://app.suno.ai/song/d5cc7bc0-2885-48dd-a7df-c804d3527c6c

[broken lotus]

Agree 100%

People have subjective memories of music...

everytime I hear someone saying "Suno is giving me bad generations today..." I am always thinking it is just as likely to be a person's mood.

I have to wait and hear my songs at least a day later to know if they were actually any good or if it was just the endorphin bubble of completing a 'game'.

And the mass hysteria about the vocal chorus is... over-rated.

[vast ledge]

Try downloading the mpeg then import it to audacity where you can export it to WAV or FLAC

[meager nacelle]

😭

Co-pilot seems to work 😁

[burnt needle]

My song one the website is stuck in limbo lol I can't download it >.>

[agile veldt]

converting mp3 to flac will not bring back the quality lost by the mp3 compression

[waxen hazel]

is it normal to get 100% cpu usage long time when creating? can't do anything else on comp

[untold briar]

No. Shot in the dark, do you have browser 'hardware acceleration' disabled maybe?

[waxen hazel]

nope, probably just my shit vivaldi browsers gonna check safari

any way to edit the image with the song?

[versed valley]

Accepted terms of service.

Welcome! For help, check out Suno Help

[chrome jasper]

Latin song https://app.suno.ai/song/85a4e3e5-9d6a-4c4f-9bac-83898550eeab

[alpine venture]

how's that for vocal clarity

https://app.suno.ai/song/dfb6eb2c-8a8c-4078-8ec8-e06a7e743caf

[graceful condor]

DF premium users only now

That's some BS

[supple hollow]

I just wanted to make some breakcore but I can't

[graceful condor]

Ik I needed background music for a song I was singing

I'ma slap whoever thought making it fully premium was good idea

[cunning musk]

So I'm not the only one who noticed this. But overall it seems to me that metal has become too noisy compared to V3A

[verbal pulsar]

So I decided to test what and how much affects the degradation of the quality of tracks generated by Suno.

I noticed the greatest degradation in vocals and overdriven rhythm guitars, so the test will be divided into 3 main parts, starting with:

1. Base, which is a preliminary 2-minute track that will serve for further testing.

2. Guitars, which can go from a powerful "Wall of Sound" to an unintelligible mess in the middle of the stereo spectrum, moreover pushed far behind the bass and drums...

3. Finally, vocals, which as the foreground element of the song can affect the entire mix.

4. BASE
   The base will serve as the parent of the next generations in both parts 2 and 3 of the test, so it must contain elements necessary for further testing:
   a) No Lyrics, only Metatags
   b) A track where only rhythm guitars will be audible and must fill the entire stereo spectrum, i.e., a classic Wall of Sound
   c) Guitar solo\lead guitar that will sit tightly in the middle of the spectrum with simultaneously well-positioned rhythm guitars
   d) Acoustic interlude or solo
   e) Breakdown
   f) Finally, rhythm guitars alone again to check if degradation occurs already in the first track due to all the stages occurring between a) and e), and I will compare this in the DAW.

For the next generations, i.e., Continuations, I need all the above points and as objectively good sound as possible.

I will generate no less than 10 tracks (50 credits) for the base. Of course, only the best will remain as the parent, and the rest will be used for statistical analysis (yes, I know - the sample size is too small for conclusive results, but I can't afford a larger scale).

[verbal pulsar]

2. RHYTHM GUITARS.
   a) The second generation, i.e., riffs, riffs, riffs - created from the Parent where rhythm guitars play alone and have the best sound.
   Possibly without the involvement of lead guitars and other "additional" instruments.
   In this generation, I only examine whether there is degradation due to the continuation alone without the involvement of other variables.
   Again, I will create a minimum of 10 continuations (50 credits) and again choose only one for further parts, with the rest being for statistical purposes.

b) The third generation, i.e., solo.
In this generation, there must be:

• main riff of rhythm guitars
• after that, a solo, which will be in the foreground, possibly in the middle of the stereo spectrum (from this point, another generation will be created)
• after the solo, the main riff of rhythm guitars must return.
  Here, I examine degradation after the solo occurring in the continuation.

c) The fourth generation, i.e., acoustic solo/break.

• main riff
• acoustic solo/break (from this point, another generation will be created)
• main riff

d) The fifth generation, riff, riff, riff

As you can see, subsequent generations are created in places that are not natural for rhythm guitars and their big sound.

That's the action plan, but I don't exclude the possibility that there may be changes during the test due to unexpected results. However, I would still like to stick to the above guidelines.

I'll try to keep you updated and inform you about the results and my conclusions, attaching them to the previous posts to avoid getting lost in the Discord mess.

[verbal pulsar]

https://app.suno.ai/playlist/2a4ba29b-7cfb-43c3-8a37-b18ac268c1de

Damn, I didn't expect that. After generating the first generation, I can already say that the degradation is audible already at the stage of the first track... All it takes is the lead guitar to appear and the rhythm guitars come to the middle of the spectrum and retreat behind the drums... Only two songs are suitable for further testing. Marked on the playlist with thumbs up. In both cases, only in the intro there are rhythm guitars, which are not covered by lead guitars, and it is in them that you can best listen to how the quality of the guitars degrades...

[verbal pulsar]

The second generation confirms the initial thesis. Again, only two pieces are suitable as parents for the next generation.

[verbal pulsar]

The entire third generation turned out to be a failure because I Badly chose the place for the continuation - An additional instrument, probably a string one, came on, but I didn't hear it, and an entire generation was born with constant artifacts. So I have to generate again.

[summer dagger]

@inner wagon I was looking on Suno for you to see if there is a way to add a file and have it generate that melody and delete that nature sound the song. Right now, I have not come across that feature yet.

[inner wagon]

Thanks for the help!

[verbal pulsar]

The third generation "two" brought an interesting conclusion - it is not the lead guitar itself that spoils the mix, but its specific type - a lot of effects such as reverb, chorus, etc.

[verbal pulsar]

Because Suno didn't fully understand the prompt [acoustic break], we again have the opportunity to listen to guitar solos and a few prematurely ended songs. Only one of the 10 contained a piano And served as parent to the 5th generation. When it comes to degradation - the more effects, the worse.

[verbal pulsar]

The fifth generation is a disaster again. After the gentle piano, out of 30 takes, Suno never returned to Progressive Metal, he always finished the song with the sounds of the piano...

Conclusion 1:
The noticeable degradation in sound quality is not related to the continuations themselves, but to what happens during the song. A large number of effects negatively affects the mix. You can hear it already in the same part of the song (Especially from the first generation). Of course, if you continue from a place that already shows signs of degradation, the quality will continue to decline as the song progresses.

Conclusion 2: less than 20% of the generated instrumental parts are suitable for further processing... I assume that the vocals will be even worse, but unfortunately I don't have any credits left to verify it 😦

Finally, the entire piece consists of 5 connected parts:
https://app.suno.ai/song/13afa2d9-9398-491f-874c-28d1ac6d8de2

[untold briar]

A few general notes. In any continue clip, the style prompt has a weaker effect if the previous clip is very long, and a stronger effect on very short clips. In V3A this meant that 10 to 20 clips could change music genre entirely. In V3 this is more difficult, if you want to completely change genre, often have to go as short as 4 or 5 seconds sometimes (not always, 8 or 10 can work). for instrumental you do not need to totally change the genre but if you want your style prompt to have a stronger guidance effect, you can try to use arrange things so you continue from a clip closer to the starting time. This way the clip is shorter and the style prompt is stronger. Conversly if you 'get whole prompt' on that clip, it becomes longer and the style prompt change things less.

[rare delta]

this sounds amazing, what genre/prompt did you use to get this kind of instrumental? I was trying to get something like this but it was very soft guitar

[verbal pulsar]

Non stoping "Progressive Metal"

[rare delta]

Thank you!!

[untold briar]

"Degradation Generation" sounds like a song title. Not sure it the Degradation Generation would be the older or younger generation...

[verbal pulsar]

I have a problem with renaming songs... Some bug causes one of the Parts to be renamed, not the Full Song

EDIT:
And in response to your post - unintentional second meaning 😂

[alpine venture]

fun experimentation with "hyper" prompts

https://app.suno.ai/song/72e9046a-a244-433a-8ec0-886759917bcc

[wise terrace]

Hey guys! Just started experimenting and really love it. One question: Is there a way to edit the lyrics? I stitched some clips together and the lyrics are now not completely accurate for some reason. Not super important but yeah.

[verbal pulsar]

I'm afraid not. Many of us would like such a feature, but unfortunately changing text requires a completely new generation.

[wise terrace]

Ah thanks, good to know!

[rare delta]

if I subscribe right now, when exactly do my credits expire? on april 30?

[alpine venture]

wtffffff

https://app.suno.ai/song/ea97a4cc-3d27-4a91-b2df-48cde9ec6faa

[fallow gull]

Definitely understood the "hyper" part.

[verbal pulsar]

Will you also make Sagem song? 😊

[alpine venture]

Sagem deez nuts?

yea, of course

just a sec

[alpine venture]

https://app.suno.ai/song/91b76b70-7e0f-4ccf-94d6-e0a1814b544c

[bright root]

About not returning to the initial material - that's expected (not to be confused with "OK"), since it's AI with its context. Like LLM model, it will forget what it started from, pretty soon. I guess a smart context, like in some LLM solutions, will be added in the future, so that initial fragment won't be forgotten. At least I hope so.

[alpine venture]

https://app.suno.ai/song/16f02364-1e28-4a06-8379-4440fb3df419

yall have a good pne

[verbal pulsar]

I see. Only the AI ​​got instructions in the form of style and meta tags, stating that after the piano there should be another rhythm guitar riff, then a breakdown (not necessarily related to the previous ones - they could be completely new) and end of song. Despite this, all 30 parts were a dozen or so seconds long piano endings.

Of course, the last part referred to the previous one with a meta tag.
I'll have to try inserting a [break] between the piano and the next riff - it helped me once, but I forgot about it 😅

[exotic fiber]

@verbal pulsar interesting, could you please try with this in ur testing?

in the style prompt to check if we get a cleaner result:

{unprocessed}

Also try with {FLAC}

[verbal pulsar]

Unfortunately, I have used up all the credits... But feel free to experiment. Below is a link to the created playlist with all generations.
https://app.suno.ai/playlist/2a4ba29b-7cfb-43c3-8a37-b18ac268c1de

[exotic fiber]

i think i get less effects with that... but a little bit distorted, however, this type of tiny distornion can be hanfled on any DAW very easy...

[exotic fiber]

mayve tomorrow...

Due to heavy usage, generations are temporarily only enabled for Pro and Premier subscribers.

https://tenor.com/view/broke-gif-4800807

[verbal pulsar]

However, I don't think that forcing uncompressed audio will help (and I doubt it can be done), because Suno mix has a problem with limiters. Uncompressed audio won't make peaks magically disappear...

[exotic fiber]

but makes it easier to work after u split the track into stems... better to have unprocessed audio for that...

because u are only going to loose information on ur process... not on sunos and then urs..

however, i still dnt know if the unprocessed tag works, i dnt know if is just my imagination or i get a little louder, but less effects

[verbal pulsar]

It makes sense. Someone once wrote that downloading MPEG and isolating audio from it gives better quality, but I don't know, I don't confirm.

[verbal pulsar]

Can you give a link to a project with this tag?

[exotic fiber]

https://app.suno.ai/song/92ebc672-69d6-4abd-9eb0-630013343529

it makes the bass and kicks louder. but for me thats better to process...

ahh.. and its not rock.. its Reggaeton.. xD

https://tenor.com/view/homer-simpson-dancing-grinding-lol-gif-20312555

https://app.suno.ai/song/fadff9a0-8f0b-4f3f-a54c-ad6c47bed486

another urban style

[verbal pulsar]

I'll check on the monitors tomorrow - in the earbuds it sounds like it has a lot of subbass and distorted kickdrum, but my headphones are lying😅

[exotic fiber]

high bass and kicks levels using the uncomprssed, i usually dnt get the bass and kick that louder without the tag

[exotic fiber]

hahaha yep yep.. sounds ajust a little bit distorted, but, to process the stems is better like that.. more room n range to work with

[eager topaz]

I've tried that and there doesn't seem to be any difference at all for me. I don't see why there would be any difference.

[autumn perch]

https://app.suno.ai/song/8939a179-ab44-4113-af58-f307b3137bf5

This could be much more cleaner.... too many distortion on intense parts.

[bright root]

Noise builds up after first minute. And becomes unbearable by the end. If my headphones are not lying.

[final hazel]

In a perfect world yes.. in the current state, it's hard.. They will get there eventually 😉

[autumn perch]

It's already incredible what they're doing.... something that even a few years ago would be considered impossible to do

I created pieces that could be instant hits on radio 🙂

https://app.suno.ai/song/6c2ea9b6-a30e-4783-a8a4-d675b1dc2ccb

[exotic fiber]

yes, i know, but r easy fixes processing the stems...

[autumn perch]

I like this one, in my office has become the main theme song:

https://app.suno.ai/song/5f1e8260-6f88-402d-94d8-61af07585693

[eager topaz]

Around 00.50 there is distortion on the kick , listening on my monitors. I get this same distortion on the kick quite often. Also checked on my h/phones and it's the same.

[exotic fiber]

perfect, thanks...

[eager topaz]

Good to post these things bec I'm sure that Suno guys check these posts

[bright root]

ugh... Solo becomes chorus in all 6 takes.
https://app.suno.ai/song/b8f765f3-19b4-4625-baf0-12c08335d678

[autumn perch]

This happens always, I still have to listen to a complete song where there's absolutely no chorused voice. Even if it starts as solo, it will become chorused for sure after some time.

[bright root]

Oh, I had a few masterpieces, actually. No chorus at all. And sometimes, depending on style, it only added to perfection. But what happens in a bit more than a day now, is nearly complete noise pollution.

This reminds me of gold digging in the Wild West. You sit by the water and wash the gold. Some compositions are superior than what some pros were able to achieve in their careers. The majority, however, is just sand.

[autumn perch]

How does the "Public" tag is used for? You can't search for songs on suno site.... you just have "Trending" and "New" tabs.... and which songs goes into "New"? I never saw one created by me even just after putting Public on it....

[untold briar]

In order for a song to appear in new, it has to be made up of at least two clips, so you click "Get Whole Song" and the song will say "Full Song" on it. Then if you like the song and make Public, it has some chance to appear in Explore page.

[autumn perch]

Going forward in a near future there will be way more ai generated songs than real ones 🙂

[bright root]

And that's a good thing. You promote yourself from musician (whistling in shower counts) to producer. You produce what you actually like. I think this will increase the average quality of music.
Besides, as long as your music causes the emotional response, it doesn't matter who or how created it.
Honestly, the best composition in my list TODAY is the one I created myself.

[dark inlet]

What is chirp exactly? Is it by Suno? Is what Suno uses to create their music or is it different?

[bright root]

AFAIK, it's an AI engine, a model, akin to GPT, by Suno, used to create music.

[eager topaz]

It's crazy, but some of the songs I've created in Suno are so good that it's hard to believe that I created them. I've always written songs and I still write all my own lyrics, but so easy to produce a great song in Suno.

[rose ore]

My thanks to the dev team at Suno for all the time they're putting into making all this possible. The v3 model sounds a lot better today

[pseudo steppe]

/chrip

[hoary kraken]

Is there any way to control the vocals when transitioning from Part 1 to part 2 of the song?

Using suno that is

[broken lotus]

https://www.suno.wiki/faq/making-music/my-words-are-wrong-after-continue/

[broken lotus]

Suno is the company.
Chirp has been the name of their text-to-music bot here on Discord, but ever since the webapp launched they've transitioned into just calling the whole service 'Suno' and naming the later AI models as 'v2' and 'v3'.

[swift sundial]

🍉 hi im new here

How can I make a complete song with an outro?

[orchid whale]

I'm already aware of the continue feature, but is there a way, let's say, to change the first verse without altering the entire rhythm of the song?

[alpine venture]

continue with instrumental and use [end] as style

[tulip granite]

How do we get a song with a specific length. Let say 90 sec.

[icy fiber]

Hello guys, is there a way to use 2 voices (male and female) in one song?

Like this:
A (male): blablablabla
B (female): lalalalalala back to
A: (male) blablabla

[deep prism]

https://youtu.be/nvBtHISQFzo

[alpine venture]

https://app.suno.ai/song/f9ff9f4b-67c6-484e-9fa2-9342aa483f2d

some interesting prompt experimentation in this one

started on
paranoid experimental goth trap lofi cassette 4-track homemade amateur, post-creep, autism

and ended up using
emotional psychedelic 70s guitar solo

to get it into some more musicality

then reverted back for the end

[vast ledge]

yes. you will also know in your billing when it expires. If you just want to test for a month, i suggest cancelling right after you pay for a month. Suno will cancel your sub when the current month expires

[vast ledge]

Not at the current version. You use either but need to tell Suno which one since the default is a female voice

[vast ledge]

You can't. The first defaults to a max of 2 minutes or less depending on the lyrics you feed it. For instrumentals, Suno generates the first 2 minutes of it. You can download that 2-minute version and edit it yourself cutting it down to 90 seconds.

[vast ledge]

No

[versed valley]

Welcome! For help, check out Suno Help

[orchid whale]

(Still haven't tried this, and I could be completely wrong)
Maybe you could try something like
[Verse 1, Female Vocal]
[Verse 2, Male Vocal]

Maybe do that for each line, not sure. I'll experiment that myself

[orchid whale]

Didn't work

[verbal pulsar]

Maybe [female Singer]?

[verbal pulsar]

Additionally, a conclusion from tests: the [rhythm guitars riff] meta tag creates only rhythm guitars without the participation of lead guitars (in most cases). This is quite important for creating good-sounding riffs for further work.

Edit:
But for some reason all the riffs created with this matatag sound like metalcorish 0-0-00-0-00-0-000 riffs 🤔

I need to experiment with different subgenres.
And maybe start with V2?

[pliant basin]

@meager skiff eyo bro someting wrong

@dim glacier do you have the auth to ban members

[dim glacier]

I can only timeout. and I can't delete messages in other channels

[pliant basin]

oh, do you have any communication with any admin?

[dim glacier]

I could ping the other moderators, yes

[pliant basin]

All right then, good luck 🙋‍♂️

[lone oak]

How can I keep the style of the third verse consistent with the first verse?

[cold zenith]

After 4-5min, the voice become very shaky and noisy.

i think this is something problematic. And i don't understand why it change suddenly.

[outer karma]

I bought a premium membership, but why are there still 2-minute songs? How can I extend the duration?

[cold zenith]

use continue

on the song you want to continue

[outer karma]

It doesn't work, I couldn't do it

[sharp merlin]

It's not really possible directly right now via an input on the web interface. What I do to make the song end at a point I choose is I find a logical place for the music to change or start to wind down, and then set the Continue From time to that time. Then, delete all the lyrics in Custom Song, keep the style the same, but change the lyrics to "[song ends]". This will make the song begin to wind down (the tempo slows, some instruments stop playing, etc.). Generate the new part with that song ends tag, and the music should have gotten slower over the course of the part. Then, when you find a good spot to make the song end where the music has slowed down, fill in the time in the Continue From section to that time. Then, delete all the lyrics and also the style prompt. In the lyrics, put "[end]", and in the style prompt, just add "end". (without any quotes, for any of these) Run the generation and this will make the song end within a few seconds almost all the time (if the volume is low enough). And the last part usually gets the credits refunded as a bonus.

[verbal pulsar]

I don't know if this is even achievable over several parts. I've never been able to make the same chorus in part 3 of a song as in part 1

[fallow gull]

On the duet issue, see this example piece by @final hazel :
https://app.suno.ai/song/5045e5f8-d213-4cac-9d87-99b370e89adb/
Duets are achievable, they are just a bit unpredictable and may take many generations to achieve what you want.

[chrome jasper]

https://app.suno.ai/song/cd82e75d-b2d0-4238-ace1-726de112a3b7

[kindred warren]

v3 is great. but it tries to repeat the chorus to fit my short lyrics in 2 minutes. is there something I need to watch out for?

[fallow gull]

The model seems to favour verse;chorus;verse;chorus or stereotypical song patterns. Maybe because after being trained on thousands of songs, that pattern has been reinforced.
You can continue your song from before the repeated chorus and try to send it in a different direction if that is what you want.

[neat field]

i've found that not including any names in my lyrics helps keep vocal clarity

my best guess is in order to align their model, i believe Suno et al made Chirp make using names more likely to "muddle" vocals

ie people should not be making awful songs about people by name (which i agree with, cyberbullying etc)

[sacred brook]

how iss this working what kinda ai models afre used for this afre there any published papers about it or google collabs

[graceful condor]

Hello. Just trying out for first time. Does "remix" use anything from the original song (even passing it in autogressive style?) beyond the style prompts?

[bright root]

Bark was open source, I think. Chirp isn't. But both have spaces in GitHub.

[bright root]

I doubt that's the reason. I never used any names. But maybe you are up to something - e.g. if using a name definitely, in 100% cases, breaks the quality, then it might do something with pronunciation, language mix etc.

[dark inlet]

Ah. OK. And is it the same for Bark? Or did they retire Bark?

[errant lance]

/chirp

[flint haven]

The /chirp command is not available in this channel

Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.

[potent sage]

Sure! "Can you please help? I've got a song, but I want to remove the vocals. How can I do that?"

[eager topaz]

Try: https://vocalremover.org/

It does separate the music and vocals and you can download either or both.

[broken lotus]

Bark was released as open source, there is a community around Bark, but the Discord Bark Bot is maybe not a priority for them now.
Still up on github, and I think also Huggingface.

[potent sage]

Thank you, that's a fantastic idea, but unfortunately, the quality might suffer a bit through this website's manipulation

[eager topaz]

Yes, it does suffer. I use RipX Daw Pro and unfortunately, with Suno tracks the quality of the backing tracks is not good when ripped. The voice is usually ok, but the music tracks are not. Perhaps someone else can offer a solution?

[graceful condor]

wtf again we cant generate music lol

[neat nexus]

Right now they are prioritizing the people who pay for the service and make it possible.

This is not the end of free accounts being able to use the service, just a tempoarary situation.

[verbal pulsar]

You know, it's just another day in a row. I'm not surprised that free users are upset about this situation.

[neat nexus]

I mean...yeah. I get that to a degree.

But there's also the option of supporting the system they're missing so much by getting a paid account and being part of the solution to the problem...

¯_(ツ)_/¯

[verbal pulsar]

Well, not really, if you take into account the technical condition of V3... Personally, I used the entire pool of credits, I bought a package, among others: for tests, and the generator's results are very unsatisfactory. The situation is probably more complex and cannot be described in a simple sentence: "if you want to generate, buy premium".

I apologize on behalf of the translator... 😅

[fallow gull]

They are just temporarily a victim of their own success. Also maybe marketing got ahead of development/infrastructure. It seems like the recent Rolling Stone article and exposure from some influencers brought in a lot of traffic.

[graceful condor]

So after a few iterations, I can generate a song I like with quite a short generic prompt (e.g. one genre). Is there a way (not necessarily through Suno?) that I can fingerprint the track similar to how Spotify does so I can identify the vocabulary of other attributes of the track (e.g. more specific genres, characteristics of the voice).

[fallow gull]

If your song is in on YouTube, you can use https://sonoteller.ai/ to describe it.
There are other tools like https://cyanite.ai/ that can also describe music, but you have to sign up.
If your song had lyrics, Gemini and ChatGPT can describe them.

[graceful condor]

Thank you! Yeah it's more the melody and beat that I'm looking to describe. I got quite lucky on "80s synthwave" with this one https://app.suno.ai/song/c178d347-df34-45d4-8d53-8d39adcb0937/ and I'm trying to replicate the sound - not exactly mind, just the tempo and the vibe - for another song with different lyrics. I'll give Cyanite.ai a shot now.

[fallow gull]

You can also look through the suno community wiki for the styles that describe your song: https://www.suno.wiki/faq/style-and-lyrics/styles-and-genres/

[graceful condor]

Yep, that's what i've been messing with recently. Certainly better than shooting in the dark, but still feels a little... undirected.

[fallow gull]

Even if you get the tags that describe your song, there is no guarantee you could replicate it. There is a lot of variability in Suno song generation.

[graceful condor]

So I've discovered! Replicating definitely not the objective... but something with a similar vibe would be great. I've had it happen a few times but then something else has been off like clarity or pacing through lyrics. I can fix up those with changing the lyrics, but then I lose the vibe.

I expect once the tools around "inpainting" and similar make their way into the app the amount of wasted credits will drop substantially 😄

In case anyone's curious what Cyanite.ai looks like...

Thank you for the linking me to it CG!

[still knoll]

Hi there! Currently creating tons of metal songs, just trying out some stuff :D

Is there any way to make the voice level with the instrumental? Usually the guitar is louder than the voice itself.

i tried clear voices as a tag, sadly didnt helo

[tulip granite]

Can I have narrator in the web interface?

[bright root]

There are no tags that would make anything clear, crisp, noiseless etc. It's not Stable Diffusion with "masterpiece" tag. And answering your question - no.

[still knoll]

thx a lot for the heads up!

[gray valley]

anyone had luck in narrowing down what songs you want it to pull inspiration from, is there a limit to tags

i use about 3 or 4, but feel like it will randomly weight a different tag each time

[bright root]

Like with all AI models, the weight of a tag is random. However, you can't define the weight here. So if you say it's "doom metal, synthwave, gospel" - it will rather be one of them, or a mix of 2, or a Frankenstein with verse in one style, but chorus in another. But you don't control how and if it will be actually included. So yeah, stay focused.

[gray valley]

so where you place a "," can have impact on the weight?

[bright root]

I don't think so. You can even forget about commas.

[gray valley]

oh

[broken lotus]

Comma+space should create a balanced-weighted list. Items in a comma list have an equal chance at influencing the prompt, according to a Dev.

Items without commas should attempt to merge according to grammar rules (adjective adjective noun).

Loud rock guitars
Loud, rock guitars
Loud rock, guitars
Loud, rock, guitars

Each of the above are different… but HOW different?
This particular prompt may not show a lot of difference because all of three words are going to work together to make recognizable genre music.

With another prompt where the words are not so closely related, those differences may result in some clips sounding very different, as some words may be super-prompts the AI leans into

Disco, Mambo, Punk
may end up randomly favoring one genre…
Disco Mambo Punk
May sound like butt, if there is not enough common threads between them, since it is trying to invent a new sound

[versed valley]

Welcome! For help, check out Suno Help

[dapper scroll]

/chirp

[flint haven]

The /chirp command is not available in this channel

Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.

[opal kayak]

I am brand new to Suno and Discord. I'm stunned by this technology. Forgive me for just a couple of very dumb basic questions. Is the term chirp and Suno one and the same? Should I be using the Suno site or the discord method for better results or are they the same? Does premium membership transfer here as well?

[unique stone]

The site is better. Chirp is the name of the model. Suno is the company that makes it. There are 3 versions of Chirp - Discord only offers v1 as far as I'm aware. The website offers v2 and v3. Pretty common to just refer to Chirp as "Suno" since that's what the website and discord server is called, and the only other model is Bark (used for text-to-speech and much less talked about.)

[opal kayak]

Thank you!

How do I share finished songs here?

[unique stone]

Most people post them in #🐣┃suno-showcase but often they do it in #💬┃general-chat (usually if they want to talk about them or it's relevant to an ongoing conversation e.g. trying out a prompt)

[opal kayak]

So just share the url link?

[unique stone]

Yeah it turns it into playable media automatically

[buoyant cargo]

Just chiming in that I have also noticed with brand new generations that V3 has a tendency to muddy the vocal mix with an overlayered chorus effect. Hope the devs can fix soon

[opal kayak]

It sure does, lol!

[quick wigeon]

How can i make more smooth endings of the song?

[hot cobalt]

putting "intro and outro fading" in the Style of Music helped me

[quick wigeon]

Can you show me an example how you used it ?

[hot cobalt]

also i write [outro] in the lyrics

[quick wigeon]

ah oki, im making custom lyrics so i guess i add outro 😄

[hot cobalt]

It's not a permanent solution, but I noticed a difference in the songs I generated

[quick wigeon]

It got a bit better, not that "wall crashing" ending 😄

thanks.

[deft minnow]

https://www.youtube.com/watch?v=ZGl88Us7lE4

[graceful condor]

Suno’s modern sounding music sounds much cheesier…in writing and production…

If I say 2010s modern rock song - the song sounds like a crappy low quality commercial liscense track in terms of production and song writing

If I say make a 1960s Motown song it sounds like a 1960s Motown song in terms of production and song writing

Did you train suno on generic commercial music for music later than like 2000s?

[remote trail]

Hello, I received the payment subscription for Suno. Do I lose all my points after the 30 days if I choose monthly? Or do I get the new monthly points added to the already existing points?

[deft minnow]

If you have 200 when the period ends, you get up to a 1000 again, you dont get up to 1200. Old sub credits expire(unless they were bought separately)

[remote trail]

thanks for the info

[untold briar]

My gut feeling right now is the quite often (maybe the majority of the time if your priority is pure clarity) the best style prompt in a continue in V3 is blank, because it may minimize problems being amplified more. Not high confidence (didn't test this specifically, yet) more of a vibe. So the a blank style may have something to do with it.

I'm usually aiming to push the music to do strange things (hybrids, genre switches) so I usually don't use a blank style prompt by raw quantity myself, since you can't use blanks for that. So not super confident in that claim, just not enough time yet.

[iron hamlet]

Now I want test any stuff.

[raven turtle]

I'm getting varied results with "remix" but generally it sounds very little like the original. The ability to say, keep the vocals and lead lines like an actual remix would do, would be cool

[autumn perch]

Super-distorted

https://app.suno.ai/song/674d42b9-0928-4cd3-9fed-ce3285715cea

[dusty parcel]

Hey, just discovered Suno yesterday, made a couple of songs which I really liked the outcome of. However I was wondering, I noticed almost all vocals have a auto-tune, which is probably to clean up the AI voice, however almost in all songs there are backing vocals, echo's or some sort of layered vocals. Is there a way to generate music without those? I want to create some music with a more "intimate" feel to it.

[fossil ore]

I have a country song in process that was going great for the first two verses, but Suno is having a hell of a time with the chorus and then the final verse. Almost impossible to get a consistent result. Often it omits a section of the lyric or inexplicably repeats an earlier line on its own. Any suggestions for how to maintain consistence beyond the standard 2:00 limit on the first recording?

[lament cove]

I quote this, if you find a way to clean the constant backing vocals (most of the time in chorus) write it!

[ivory frost]

Hi, I'd like to watch the tutorials for making professional songs using suno by using the same voice with different styles of music and where to find good prompts that can help me generate better professional songs and Can I make the same voice sing in more than one language?

[kindred warren]

What should I do for a small wait between lines?

[broken lotus]

Try finish each line with “…”

[hybrid wave]

hi new here, i wanted to know how to generate a duo

[rough pawn]

So I feel like this is the best you can get with chatGPT https://chat.openai.com/g/g-dVkdZV2wQ-professional-songwriter

[limber terrace]

Re STYLE in a CONTINUE. Copy and replace whatever is in the STYLE with that of the part being CONTINUE'd

[twilit stream]

I encountered the same issue and ultimately generated over 200 versions of the same verse, selecting the best four to combine using an external program. Based on my experience in addressing this challenge, enhancing the distinctiveness of each verse is crucial. Incorporate specific, context-related words or phrases for clearer differentiation. Utilizing explicit markers in the text can aid Suno in recognizing the commencement of new sections. For instance, when dealing with two chorus parts, Suno may reuse rhymes from the first and overlook the stanza from the second. To prevent this, label them as [Chorus 1] and [Chorus 2]. Additionally, adding punctuation, such as commas and periods, after each stanza in the second part can improve clarity. Additionally, Suno tends to reuse older text more frequently if it rhymes better within the current stanza or if the current stanza's length is too long or too short for Suno's preference. In such cases, you can simply add the last words in (parentheses).

[twilit stream]

You might want to check out https://chat.openai.com/g/g-Bd4Armuqc-musicai-tune-composer-suno for generating music style descriptions suitable for Suno. In my opinion, it pairs excellently with Suno and effectively addresses the issue of Suno not recognizing artist and song names

[versed valley]

Welcome! For help, check out Suno Help

[verbal pulsar]

Hmm... I would like to report that Suno's censorship seems to work selectively, namely perhaps only in the English language. Writing correctives of scary and disturbing things in other languages shouldn't be a problem.

To moderation:
Sorry for putting this into Suggestions...

[river geyser]

I can't log in via discord ( Help

[tawny solar]

Wanna share 2 findings about the vocals:

1. Short reverb helps with the artificial grit of the vocals
   Adding a slight reverb with high-pass filter smears together the grit that comes from the generated vocal/noise reduction on the vocals.
2. Filtering everything below 100hz (50-100hz taper) helps make it feel less muddy with lower vocals.

[stuck marsh]

Hi all. New here, so forgive me if this has been asked before, but has anyone noticed that 2-3 continuations in the audio sounds like the AI has become ... breathless? Like it ran out of air? Hard to phrase past that.

[neat nexus]

::sigh::

I thought I could make this one work. It opened with true stereo and that made me jump at it to try to go with it. I like the music, I thought it worked well but the lyrics are just too repeative and hard to make work without some serious vocal control so I gave it up. Posting what I have here for discussion and maybe educational purposes if it can be instructive to others...

https://app.suno.ai/song/2c8b8117-bf31-4711-9080-f85157fd3f14

[tawny solar]

Yeahh, I think it starts to rely on the past "Part"Instead of the first part. Generating from the original, changing the verse/chorus helps a lot

I generated my outro from the first verse, made it much better

[stuck marsh]

Hmm. How do you piece them together if you do that?

[tawny solar]

I use the Custom mode, so I can change the lyrics manually. But I am not sure if thats a paid option or not....
From there you can continue and change tyhe lyrics and parts to what you need

[stuck marsh]

Totally using custom too, so we're good there. When its all said and done and you request the full song, doesnt it go through and pull all the 'parts' together in their order? Can you customize how they fit?

[tawny solar]

I put it together manually, thats not a thing poissible in SUno yet

[stuck marsh]

ah, got it. Not a terrible idea, actually.

[tawny solar]

and if you are good with the timing they slot together pretty well. This is5+ parts

[tawny solar]

And it makes you able to generate seperate interludes/solos if needed :)

[stuck marsh]

I'm worried that I'mma screw this up

[tawny solar]

Nah fam, if it feels right it probably is. Imo with music

[stuck marsh]

About 5 songs in i realized that after verse 1 all my vocals were ... 'faded' more and more each verse.

[tawny solar]

Ngl, I went through my 2500 credits in less then a week 500 left until 30th

[stuck marsh]

haha. I feel that. I started Friday.

[versed valley]

Welcome! For help, check out Suno Help

[stuck marsh]

constantly running into the music trampling over the lyrics, drowning them out almost completely. Anyone have any tips to get around that?

[tender eagle]

wow it sure is glitching, I've lost 100s of credits already to this shoddy buggery

it lost half the song's beginning, which i heard DURING creation, but now only the end is there... once it completed

as well, about 1/4 of the songs are choppy and hiccuping...

not cool

I go through credits so quickly already, to lose them to stupid clearly buggy mess like this...

I'm so mad... cos it was perfect for the continuation when it was generating and I could preview it, then it is missing the part I previewed. it paused at 33 seconds in, and then when it finished actually generating, it starts WHERE it left off not at the beginning ofd that generation >>>

[chrome jasper]

Joseph Haydn https://app.suno.ai/song/1fb067ac-1b03-47eb-9db6-17ff796b6554

[candid arch]

Hey is the chirp bot the same as the app? and which version does the bot use?

[near patio]

Any idea when suno will be up for all users again? I'm normally a subscriber, but my plan expired and I don't get paid till around 2 weeks to be able to get it again. Been wanting to create songs all day lol

[polar phoenix]

/chirp

[flint haven]

The /chirp command is not available in this channel

Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.

[river lodge]

Hey guys! Maybe silly question: I what ai or models is Suno.ai trained? Chirp? Bark? I would like to know for further prompt learning

[night pelican]

Wondering if there is a solution to this - whenever I build a song by continuation, sometimes I have entered too many lyrics to fit into the one or two minutes. When I then continue where I want, the written lyrics from the original remain and once a full song is generated all the lyrics just get added together making it look like an ugly mess. Example see this song: https://app.suno.ai/song/1825b75e-8e6b-48c6-afeb-4b070db8cddf/
Is there anything that can be done about this? Could the written lyrics be made editable?

[kindred minnow]

Is there some sort of protection with suno against real bagners pop song ?? I produced music for like 17 years.. im use to feelwhen a song is great, with the real flows of lyrics, the catchy progression , the build up... i used a lot of suno for the past month i noticed something when you get a very dynamic pop song whith great vocals with certain dynamic, it always end up the same... a weird(almost feel like an internal Ai feedback) from som reverb echo of some sort that always the most part of the song... its always there almost in avery great pop song built with suno. here the example that sound is killing me.

1:25 the infamous sound

this sound is very very disturbing

[versed trail]

might be due to the bugs going on rn

[kindred minnow]

ive been hearing this noise in song for the past 3 weeks

[real folio]

I'm getting a bit of that as well but with some fiddling and finetuning I'm getting somewhere. This tool is epic and but also has a lot to improve. My mind is blown that this is possible at all tbh.

I did not expect it to continue well enough to allow me to just stitch multiple continuations together like that but... here you go 😄

For some reason it really likes happy hardcore though and it's difficult to steer it away from that

[versed valley]

Welcome! For help, check out Suno Help

Accepted terms of service.

[craggy sentinel]

I can't sign in on the site. I've been a paying subscriber for months, but only use discord. I can't sign in with discord on the site. Very frustrating

Where can I get tech support?

[eager topaz]

It's not good ever since V3 full release. I've tried all that I can to get rid of the noise and I've been a producer for 30+ years and have all the software known. I'm giving Suno a break until it improves.

[eager topaz]

The noise that bugs me starts at 1:28 on your track not easy/impossible to get rid of it. I can't label it, but it's some sort of phasing issue, I suspect.
I Hope you don't mind, but in the interests of research I downloaded your track and ripped it in RipX DAW.

The biggest contributor to the noise on your track was the percussion track. There was noise on the vocal track and an overall noise that wasn't on any of the tracks, so I'm going to assume that it's some sort of phasing when generating effects, but can't be sure.

[humble thunder]

What is the memory length of the AI? Seems like successive generations forget motifs from the beginning of the song after a certain length

[kindred minnow]

ok

[light cosmos]

you can however, use things like 'acapella clean vocal' and get a more isolated version i found

[rugged tendon]

is there going to be an API access so that I can write code to slice and merge lyric pieces. It a bit tedious to slice out sections of my lyrics to past into the ui to then continue.

[robust river]

At 51 seconds, the sound was too piercing and I couldn't hear it anymore. The one in front was very good

[kindred warren]

v3 doesn't stay true to the lyrics either. It repeats previous similar texts.

[alpine venture]

fun prompt 😛 https://app.suno.ai/song/0cc2b98d-f280-4ff1-99ab-f4a073bc90dd

unfinished though, and likely needs more variation very early on, but with credit budgets toward the end of my gen period, I get less willing to do massive variation

[bold nexus]

#📢┃announcements

[ivory peak]

Same here!

[river lodge]

ANYONE would like to share knowledge on ADVANCED prompting or workflows to make your songs with Suno sound godlike would be much appreciated. I'ma put all together to create an in depth GUIDE 😄

[broken lotus]

contact:
[e-mail]

[final hazel]

Maybe wait a little? There is a fully new faq coming from Suno very soon with way more indepth knowledgge than it has now. Up to you ofcourse, I just would hate you've been doing a lot of unnecessary work.

[regal sparrow]

what is soon ™️ in this context? 🙂

[final hazel]

As soon as they have time.. it's extremly hectic for the Suno team.. But I am pushing them gently to get this up asap.. I want this badly too!

[regal sparrow]

ok, ty 👍

[river lodge]

Oh, didn´t know about that, when is that going to come do you know?

[final hazel]

read above

[river lodge]

So could be 1 day, 1 week or 1 month... it´s goo tho

[final hazel]

I have no insight in their workload.. I just know the FAQ is ready but just needs some checking, 🌼

[river lodge]

How did you find out about the existence of it? Do they have some kind of newsletter or are you a pythoness? 🔮 hehe

[untold briar]

I'm not super convinced that prefixes like anti- are doing something that different than randomword- where the random word happens to be one that has an effect in the output. While some random words do little or nothing, maybe 1 in 3 or 1 in 4 that I tried have about as much an effect as a prefix like anti. For example I think oatmeal-country is at least as distinct as anti-country https://app.suno.ai/song/2a8cd5ce-311d-4866-a371-c6763fc080d9/ (sort of old fashioned)

[final hazel]

I am a moderator and have short ties with Devs?

[quick valve]

בלב המזרח, ישראל עומדת גאה,
צומחת ופורחת, כמו פרח במדבר נאה.
מתקדמת בטכנולוגיה, בחינוך וברפואה,
מצליחה וגדלה, כמו עץ תמר בטוב תואה.

ברחובות תל אביב, עד חופי הכנרת,
כל אחד תורם מהלב, עושה את השינוי, זה הסוד הגנוז.
מהצפון ועד הדרום, מהמערב למזרח,
ישראל קולטת כולם, כל קול וכל פרץ.

אז למה לגייס כולם, אם יש דרך אחרת?
לבנות את הארץ בידיים, ברוח, בלי קרבות.
חרדים וחילונים, ערבים ויהודים,
כולם יחד במאמץ, לבנות עתיד טוב יותר לילדים.

כי ישראל זו לא רק ארץ, זו בית לכל נשמה,
מקום שבו כולם יכולים לצמוח, לא רק בכוח הגופה.
ברוח, באהבה, בכבוד ובשלום,
ישראל תמשיך לפרוח, להיות אור, להביא תקווה וחלום.

[chrome jasper]

https://app.suno.ai/song/37cd427e-a42d-4a57-8529-78a431ae3042

[empty pecan]

There's a melody that I'd love to use but isn't in the Suno app, is there a way to give Suno your own recordings, and if not, is that something that may potentially be added in the future?

[empty fiber]

Hello everyone, by any chance, has anyone managed to create a music piece using a talkbox voice? Thank you!

[carmine minnow]

i tried V3 yesterday right before they disabled free use so I wasn't able to try continuing this.
how well do continuations work in V3 though? are they influenced by the last ten seconds of the previous clip? like where i start the continuation might have a really big influence or does the whole clip influence the next one?

https://app.suno.ai/song/dd258016-22d4-4500-8a91-b3ae79fe062c

[versed valley]

Welcome! For help, check out Suno Help

[glacial scarab]

Does anyone know if say said anywhere how much resources generating uses? I don't mean all generations just in general. I suspect they didn't but i would be interested to know if theoretically this could be used on "prosumer" hardware (think 4090 and above) locally. I know it probably won't be released in that way i am just interested in the knowledge

[rapid bluff]

I’d have to imagine it’s quite a bit, if you have ever tried to run bark locally, probably a much more enhanced version of that. But nobody knows for certain

[glacial scarab]

Hmm.. yea.. I can imagine. I think the most resource intensive thing i did was train StyleTTS2 locally. Maybe some day we will find out 😄

[tame oar]

hello

i have problem with sub

my credits desapear.

[dry kestrel]

spend your time or money:)

[tame oar]

anyone to help?

[worldly juniper]

It seems that credits do not roll out to the next month, they just renew and you have to use them all in the same month

I also would like to know if the daily renewal of 50 credits also applies to subscribers?

[final hazel]

Credits in Suno's monthly plans are designed to be used within the specific billing period and do not roll over to the next month. This is because Suno allocates computing resources based on expected monthly usage, ensuring optimal service performance. Rolling over credits could lead to unpredictable demand, potentially causing service disruptions and increased operational costs. The non-rollover policy helps maintain a stable and efficient service for all users.

[final hazel]

They are looking into that.. when they have stablized the platform.

[worldly juniper]

It hasn't been exactly stable for free users, but I understand, it must be very expensive to run these models

[final hazel]

That's because the demand grew in the last few days.. Can you imagine if they also let free users come in too atm? 🫠

[worldly juniper]

I don't quite get it... So if I subscribe, I won't have the daily refresh of 50 credits anymore?

[final hazel]

No.. I explained above

[tame oar]

so the credits is only for a month? and how i suppose to know that.do you write this in your sub descryption???

[worldly juniper]

Yeah, that's perfectly understandable. I'm grateful that we still have a way to use for free, even if I have to login in odd times to generate my songs haha

[glacial scarab]

Uhm yes? @tame oar

[worldly juniper]

I see, thanks for explaining that :)

[final hazel]

It is

[tame oar]

i dont want my daily 50 i need my 2k bak

[final hazel]

For any concerns just mail [e-mail]

[tame oar]

where is that ? in what tab?

where to find this i need to read it!!!!

[worldly juniper]

https://app.suno.ai/account/

There you go

[tame oar]

thanks i read at questions categories. dont see that down there 🙂

[worldly juniper]

Anytime! I also wished that the credits did stayed with us

Hopefully when they upgrade their servers 🤞🏻

[tame oar]

only patato stay with us heheheh 🙂

[worldly juniper]

Potato is forever 🥔

[hidden cedar]

So, what's the deal about vocals getting distorted later in the song? I've noticed this a lot recently since the V3 beta went live. I thought it was something to do with a [chorus] tag because it almost seems like subsequent vocals have a chorus or multiple voice effect on them that gets "stuck" after the chorus?
But then I just made this instrumental one-shot song which has a pretty clean slide guitar riff at the start, and if you skip along the song it gets more and more distorted (in an unpleasant way) as it goes on. Which is possibly what is happening with vocals?
Just wondering how everyone else is dealing with this. Maybe continuing the song after only 30 secs to avoid it falling into distortion? But that seems annoying.
https://app.suno.ai/song/b6ca13b7-8eab-490e-a9e4-e17dd4e64deb/

[charred lake]

My favetimestamped snippet from this new 17+ min house / drum & bass mix: https://on.soundcloud.com/wesZgobTYHzY918u9

You need to click the link, embed doesn't work with timestamps

[ionic wind]

The Explore page has been completely frozen for the past 24-48 hours for more people than just me, right?

[worldly juniper]

If you mean in the sense that musics aren't being updated, then yeah... Would love if they implemented a search function

[eager topaz]

I did a song this morning and I notice that the distortion on voices has improved and now much cleaner. I used your Style and got one really distorted track and one really quite good and clean.

[eager topaz]

That's good how you removed the reverb from the vocal. How did you achieve that?

[rough pawn]

I used audimee, to effectively do a voice changer on the voice after isolating it

[eager topaz]

It worked well, thanks for sharing.

[rough pawn]

Apparently, you can also do instruments

although it sounds like someone singing through one of those plastic things

[lament zenith]

Why are the generations on v3 back to 1 min now

[empty pecan]

As a singer, I'd love if maybe for Version 4 we could have a tool to pull the voice out of an audio we've made. I've made a few songs I really like, but can't recreate them w/o the voice. Is that possible for the future?

[opaque wasp]

DUDE

I HAVE 50 POINTS IT SAID

and it still wont let me do any music

[flat field]

Can you provide more detail?

[opaque wasp]

i did /chirp and it said i didnt have points so i clicked manage account and it said i had 50 points

but it keeps on saying i dont have points

[eager topaz]

Did you try doing a Refresh on your Create page?

[opaque wasp]

yes

and ive checked my account multiple times

it says

that i have 50 points

i dont get it

alr it started working

[wanton reef]

how do you change the language?

[lunar cloud]

I am so creeped out right now!
This is the 'song' that Suno spat out based on a super-tame prompt (anthemic indie folk-pop, male vocals, dance, acoustic guitar, piano).
It's like the AI is laughing maniacally at my lyrics.
😬

Listen till the end...

[cobalt warren]

is there a way to import or continue songs from copilot? i have a song powered by Suno that i would love to continue.

[eager topaz]

What a shock!

[kindred warren]

Is there a way to make suno say riddles and answers instead of singing?

[polar olive]

How can I tag/create songs with the voice of specific musicians?
I tried to use tags but they dont seem to get recognized, does suno auto detect lyrics of a song and try to match the voice with what it knows as the creator of it or how does it work?

[rose ore]

I figured out how to break the v2 verse calculator in a really interesting way, and it's something I haven't seen anyone else do yet.
Keep in mind this doesn't work at all in v3.

So, in v2, you can change the verse calculation by moving words around, or small phrases in different lines, which gives you the ability, essentially, to mark emphasis and control flow. This doesn't work in v3, and only sort of worked in alpha

For example, rendering in v2, there's going to be a difference between

And when the mood is right, we've always got tonight

and

When the mood
is
right, we've always
got

tonight

V3 would usually read something like this robotically

and skipping a line for emphasis still works in all models.

[ionic wind]

Interesting observation I've made while experimenting with funky/groovy music these past few days: Adding "choir" to the prompt (as long as the prompt is otherwise meaty) will often not actually add a choir but will instead enrich the music and add pleasant textures to it. It makes a huge difference to some genres.

[rose ore]

oh that's interesting

[ionic wind]

Here is an example of what I mean. While there is no before/after, you can hear the effect of the "choir" part of the prompt, especially in the flute's reverb and in that the flute is "sung" rather than played in the traditional way. There are brief sections where you can hear actual choir, but it's brief: https://app.suno.ai/song/bef5beb0-6665-4a63-9b84-68b0f18fb0ad

When I drop "choir" from this particular prompt, the music is much simpler and tighter, which can be both good and bad

[versed valley]

Welcome! For help, check out Suno Help

[undone kelp]

Anyone please know how to specify language for the lyrics? It always starts with an English accent. 🙏

[versed valley]

Welcome! For help, check out Suno Help

[polar olive]

Can u see who created this or what was used?
https://app.suno.ai/song/2f0c9dec-26d6-442e-a7fb-e7c87ce02dfb/

Since everything here is pretty nice but I doubt that "Hardstyle" and "big beat drop" is everything that was needed as tag to create this

[final hazel]

You cannot see who created it.. but you can open de link and you will see the lyrics and the style of music then. @polar olive

[polar olive]

yeah checked that, tried same tags but the songs I tried never came close to this

[final hazel]

Maybe.. but that is the only thing this person used.. You can see if you look at the video.. the style of music never changed

[gusty yacht]

اغنيه للطيور

[versed valley]

Welcome! For help, check out Suno Help

[sacred swift]

The songs being made are short, is there any way to make them longer songs with fewer lyrics?

[lilac anvil]

I find that the sound is too saturated and the voices are too noisy, can you do something about it? on rap or hip hop it's unusable thank you.

[cyan matrix]

hi

[hybrid wave]

just sharing some (is it one though) a discovery with making a duo (but still not clear) (ande the tempo:fast does not work sometimes) https://app.suno.ai/song/75ee4d69-c496-4643-ab20-a6ddc968bf1f

[stray loom]

Any method known to get around the very weird... voice trembling while using V3?

[neat field]

how many prompts honor these lyric tags ? did you have to cherry pick ?

[solid swift]

I was just playing around just doing random crap lol and I think this is a banger what commands can I do to improve this omg!!!!

https://app.suno.ai/song/3770edec-0377-40b0-bee0-bc806773c2b7

[neat field]

you could remix the song with lyrics or click the 3 dots near the share button and "Continue"

[raven turtle]

I've tried isolating the vocal using vocal Remover, and then running it through Audimee to create a new vocal. Then blend this over the top of the original vocal in Logic. Had some success, but it's not easy given the underwater vibe the MP3 is kicking out. It'll get there eventually but TBh just thankful it's an amazing ideas generator if nothing else ATM

[stray loom]

That's a great idea for a pipeline.

Can you show me any results?

[raven turtle]

No worries. I'm basically taking the idea that the vocals aren't really up to scratch yet so using them more as backing vox, dropping them lower in the mix with mix rebalance in RX9 and adding the Audimee vox over the top, with a healthy dose of StudioD and ADT!

[raven turtle]

Will do once I'm done on it. If I'm happy with the results I'll detail my workflow

[stray loom]

i tried the first part, even on the free version

Suprisngly good results.

[stray loom]

Would be great!

[broken lotus]

Try...
Breaking up...
the lines...

Like this....

[verbal pulsar]

I have a question... Premium and pro users have full rights to their songs, right? What if someone shares the song publicly on Suno and someone else continues the song from the last second, adds a new verse, chorus and creates a full song - who has the rights to it?

[dense owl]

is it just me or the quality of the song is declining? it seems flat. no matter what prompt and lyric i use. i tried remix a previous song from a month ago using the same prompt and lyric and the differences are very clear

[versed valley]

Welcome! For help, check out Suno Help

Welcome! For help, check out Suno Help

[broken lotus]

the concept of building on another person's song is written into the ToS, since everything started on Discord when it was more social/sharing....

By agreeing to the ToS, we agree to make an exception on our rights to our songs, so legally it is already handled by the ToS. Your song is yours, but your song continued by someone else is theirs – assuming everyone has a paid acct.

Free accts could continue a public song too, but they do not own the song.

to keep your clips private:
Keep your song as Link Only ... not Public.
Do not share the song url here, or on social media (download the mp3/mp4 and share that instead)
Do not share a Playlist with any private songs.
Publish to your platform before sharing on social media. Ok to post Youtube, Spotify, Soundcloud Suno songs here. there's no way to trace back to the song's page without the url.

Songs on the Trending page for example are all Public and Continued at least one more clip.... Link Only songs do not appear on Trending, the long scrambled url is essentially content protection.

[versed valley]

Welcome! For help, check out Suno Help

[lament cove]

Question: but so this applies to building on a song made with a free account? Because hypothetically the person can't know if is building on a song i made while i'm on a free subscription. So that new build nullify the "not owned copyright".
Is that so?

[broken lotus]

I'm not a Suno employee, so I can't split hairs that finely.

Suno can track a song backwards through its creation clips, and they've recently said there is a watermarking system, but did not go into details.

[broken lotus]

But if I had to try to interpret the ToS, the original 'Free' clip is owned by Suno, and a paid acct continuing on it would not be a problem

[heady egret]

What is the max context window on v3? Like if I have a 6 minute song and continue, how much does it look at?

[broken lotus]

They haven't said...
(but consider the longest clip we can make is 2min..., and a continue clip already takes up 1min.... )

[heady egret]

Yeah, I've noticed that if you continue a clip, say at the 10 second mark, it doesn't retain the context of the original clip (because it only has that 10 seconds available). This is really useful for changing up styles, voices, etc. in a single song, then you can 'get whole song' to bring the rest of the context window back. Context seems to peter out after 90 seconds or so though, wish there were a more direct answer

[plain spear]

If anyone is trying to create instrumental music for film or generally emotional compositions, here is a graph of my most used styles that were successful in pushing the bar of the best quality I've seen from Suno.

Every style you see was used to create a piece that I considered to be successful.

[night pelican]

Why does Suno insist on ending my song after the second continuation? I've tried three times now, every time choosing an earlier time to continue up to 30 seconds before. But it's refusing to generate a full minute after that.

[untold briar]

That's an unlucky clip if even 30 seconds doesn't do it. You can try completely changing the style prompt and adding additonal length into the lyrics (using only [brackets] if it's an instrumental). But sometimes the song just really really sounds like it's ending and you have to back up even further.

[verbal pulsar]

Ok, But no ToS is above the law, and copyright laws in different countries may interpret it differently. What you wrote basically does not solve the problem I described in any way. Because you either own the rights to the track or you don't. So what's it like?

[gusty urchin]

Looking to get some good solid chanting.

Tried "Guttural Throat Chanting."

Doesn't work.

[verbal pulsar]

I've been through this before. No effects. Suno doesn't seem to have trained throat singing. The closest I could get to is a choir humming in unison in the background and a singer gently singing in the foreground.

Interestingly, with the choir in the background, Suno did not try to duplicate or "improve" in any meens the lead singer

[gusty urchin]

I had it in one song. It may have something to do with the overall theme having been medieval fantasy.

[eager topaz]

Try: Tibetan Monks Chanting in Style

[gusty urchin]

Will try that.

[eager topaz]

It might not be what you want, but it's good for me

[covert ledge]

Hi everyone, will there be a way to commercially integrate to integrate chirp into an application? Will there be an API or open source release of the model any time soon?

[verbal pulsar]

Does it work?

[agile veldt]

that "chirp-beta-7" channel... which version of chirp is that? it doesn't sound like suno v3 (the one on the website)

[alpine venture]

OK, so I thought this might be somewhat relevant here.

given the prevalence of people's certainty that they can tell a qualitative difference between v3 and v3a, I'd be interested to see which of these people believe are from which:

https://open.spotify.com/album/2z3tPbsG8FlIY8N6Dwuptx?si=DapiX0VlRLaxrLUK-gVHoA

it's a mix of alpha and standard v3

I can easily confirm by checking dates

but I suspect nobody can actually tell

also, as an aside, I need to make an album cover for an instrumental album I'm making

and if anyone is interested, I can use that as an opportunity to help anyone else who wants to learn how to use existing ai tools to produce decent cover art

if enough people are interested I'll do it

[agile veldt]

01 = v3, 02 (doesn't play), 03 = v3alpha, 04 = v3, 05 = v3alpha, 06 = v3, 07 (doesn't play), 08 = v3alpha, 09 = v3, 10 = v3, 11 (doesn't play), 12 = v3alpha, 13 = v3

that's my guess

based on listening to the voices and the general quality

[uncut escarp]

is this possible when creating a song to just update 1 text ? I mean sometimes AI says some words wrong or you want to change 1 line, can you update it without changing anything else ?

[agile veldt]

not that i know of, and that would be really useful

[broken lotus]

Devs are calling this 'in-painting'... like how we can spot re-generate in an image-gen...

They have said it is being worked on, maybe for v4 (which is probably several months away)

[agile veldt]

wouldn't it be easier to just keep the random seeds though? and let the user re-generate the same song but with the edited lyrics?

[brisk kelp]

I am trying to make a song in the genre of "ankara havası" or "oyun havası" or "roman havası" in Turkish in Suno. But Suno doesn't accept the word "havası" in "style of music" box because apparently there's an artist called "havasi"? I literally can't type in the music genre in there :D Anyone got a solution for that?

[agile veldt]

sometimes AI is good at ignoring typos. what if you use a normal english "i" instead? like "havasi"? i know it's not the correct spelling

[brisk kelp]

I will experiment with different typos thanks

[neat field]

what's the source for this information ? random sample? what was sample size ? just curious

[plain spear]

It's all the song's I've liked and published. It's obviously biased and not scientific at all. I do however have a steep bar for published pieces. It's 51 songs, most of which are attempts at Indie Pop on cello but also a lot of film pieces.

[neat field]

cool, thanks for sharing! you might want to try prompt "math rock" with your cello sometime 🙂

[plain spear]

I've made 3 playlists on Soundcloud if you want to see the best ones.
https://on.soundcloud.com/bhRmw8CK2W2H3ZSA6

[broken lotus]

From what the Devs have said, a random seed would be something like having car directions (turn left, turn right) but with different lyrics – which seems to be the desired reason, to fix lyrics – it's like starting the drive from some other part of town... now the 'driving directions' don't make sense (turn left, turn right) because it would be pulling up different parts of the neural net....

[plain spear]

The stuff from the Symphony of New Beginnings playlist is my best. It's mostly cello indie pop. Most people see it as classical, but honestly, its more pop to me.

[agile veldt]

@broken lotus that's a rather odd analogy

it doesn't quite tell me what would happen if you really did fix the random seed and only changed the text

[broken lotus]

Well, I tried to make it something non-techies would understand.

[agile veldt]

now it got harder to understand

[broken lotus]

😅

[agile veldt]

so what would happen? what would it sound like?

i mean, the same input and same seed would always yield the same result, right?

i'm assuming it's working somewhat like stable diffusion, but with samples intead of pixels

[broken lotus]

probably a random mess.... it would be fun to find out...
But not all AI functions are deterministic like that... some build on the steps before... so if we change those earlier steps it ends up going to a different unexpected place.

(not sure how far different)

[agile veldt]

so one tiny change to the lyrics (like replacing "dog" with "cat") could result in all the music being completely different? even with the same seed?

[broken lotus]

idk about 'completely' different, but it probably undermines what most people would want a 'seed' for

[agile veldt]

not sure

even if it only becomes 80% similar, i still think it's worth giving access to the seed.

but of course, i haven't tried (for logic reasons), so i cannot know if that's the case

in stable diffusion, changing the resolution just a little bit is the same as picking a completely different random seed. i wonder if that's also the case here.

but if there's just some coherence left after changing the lyrics, but keeping the seed, then i think it would be a good feature to allow editing the seed.

[broken lotus]

The devs have said 'in-painting' is a better solution...
And that 'Seeds' in this context won't work the way people want it to...

Our lyrics have a huge influence over what the song does... And the music model has some randomizations built in... (again I am not a Suno Dev, so I can only try to understand and repeat what they have said)

[neat field]

i have my own theories about the workflow behind the scenes...

we do know a major part of the workflow is similar to bark, aka GPT == changing one token of a prompt may or may not influence the outcomes of some parts of the generation pipeline as much as changing a single word in a large language model prompt

[agile veldt]

in-painting sounds like a rather risky/unoptimal alternative to just fixing the typo in the lyrics and re-generating

but i guess if that's really not possible, then there's no other way

[broken lotus]

I admit I have not seen this in action... lol

[agile veldt]

i know how inpainting works in stable diffusion. in the context of audio, you would need to be able to mark/select a range of the audio that needs to be inpainted. or if you don't care about usability at all: let people enter start/end in milliseconds... which would be horribly cumbersome.

so realistically: inpainting requires a selection tool

and that seems like a lot of dev work for a feature that might not even be that popular.

so i'm deliberately not making that feature request. 😅

i think there are other issues that need to be solved first... like the choir-tendency for example.

one last question though: which engine does the chirp-beta-7 channel's bot use? because it seems to make better voices.

[broken lotus]

all the Chirp-Beta channels use v1.

[agile veldt]

oh! ha ha

thanks.

[warm acorn]

idk if this is the right channel but is there a way to do negative keywords? like do a drone song without drums? -drums doesn’t seem to work

[broken lotus]

I would LOVE negative prompts, but i don't know a way to make that work

[warm acorn]

im new here, are you on the team haha

[agile veldt]

maybe the #💡┃suggestions-web channel. i want negative keywords too btw.

[harsh kestrel]

Is there a prompting guide anywhere or other resource I can get a quick download on how suno likes to be prompted?

[teal raft]

tony had asked about if we'd like negative prompting even if it cost twice as much

Everyone that responded said "yes"

So my guess is eventually we may get negative prompting but it will cost x2 a normal generation, probably because you'd be using more resources with the use of it

[agile veldt]

i'm surprised why negative prompting would cost extra though 🤔

for instance, it doesn't seem to slow down stable diffusion

[warm acorn]

yeah thats what i was thinking

also another q - is there ways to get over 2mn for v3?

for a subscriber?

[agile veldt]

yes

click the ... menu and pick "Continue from this"

[warm acorn]

ah lovely tysm

[alpine venture]

inpainting is coming? Oh man I will lose my sh--

[alpine venture]

that doesn't sound that awful to implement once the actual inpaint equivalent is running. And the results would be amazing for small things like correcting areas of a given song that contain minor inconsistencies or errors (which is really what it's for).

[agile veldt]

oh okay

cool 🙂

but again, certainly not on the top of my wishlist by far.

[rose ore]

So something that never occurred to me before, is that the music layer could be context sensitive too. Never even thought about it until the system made this. Check out the prompt, and squint at the hallucinated words. It's super interesting
https://app.suno.ai/song/3c3ec5ff-d92c-44d5-b03e-6877286ad732

I swear it made a thing out of repeating "waking up" in the hallucination

which is cool

This might actually warrant getting fleshed out

[manic shadow]

Just wantet to ask, is it possible to Train a New Voice and use that with chirp?

[broken lotus]

Not inside Suno...
you'd need to use another app like RVC Replay to strip the Suno voice and layer in the cloned voice.

[manic shadow]

sadge

[ornate badger]

Hi there.
I don't understand how is it connects two parts of the song when I want to continue?

[autumn perch]

Tried to generate a classic organ piece, it sounds very noisy and bad quality, Suno must be trained better with organ voices....

https://app.suno.ai/song/8ffce32a-3321-4959-b3d7-3b55ae12c45f

[eager topaz]

Not just organ. I'm finding the same problem with most instruments that I use. This is a part 2 and the noise/distortion gradually gets worse. The sound shifts left to right as if out of phase. Not really useable.

[proud merlin]

any tips how to make losing interest like almost like the song itself

with the shiloh dynasty singing type

and the guitar

[mint quiver]

https://app.suno.ai/song/29bb0818-644f-4286-a1f9-026826112fab

[rose ore]

So it turns out, you can composite emotions the same way you do with genre on suno. This is extremely interesting

I'll have a demo in showcase in a couple hours

[broken lotus]

there is an option in the [...] menu, only on the continuation clips, to Get Whole Song....
use it on the last in the chain (Part 4 or whatever) it will create 1 seamless clip with all the clips stitched together. This feature is free, so it's ok to do it at any point when you want to review how the song is going, or you need to Continue From Time and want to let it hear as long of the song as it can....

[azure crane]

Is there a tag that removes the melody in an instrumental, just to have the background music sounded?

[raven turtle]

@vladsn7821 that would be really useful TBH

[azure crane]

Yeah, I have tried with drop beat, break, breakdown but nothing seems to work.

[eager topaz]

No, sadly, and even when ripping to stems/tracks, the quality is not good enough to make the tracks very useable

[azure crane]

Well, drums, bass, instumental/vocal works sometimes great but not super of course.

[daring lava]

This is a great idea, if you’d like to please add it to the #💡┃suggestions-web channel so it doesn’t get lost here

[rose ore]

Anyone else having trouble with suno dropping packets today?

Like, just click on the generate and watch it freeze?

[shy cosmos]

yep

probably, heavy usage again

[gentle rose]

Suno is great with acoustical instruments but it shits the bed with like electronic lead sounds and lead parts. I wish I could train it with my own sounds and midi parts.

[alpine venture]

absolutely, it's actually highly effective to get it pinholes into a genre, possibly more effective than genre tagging

I'm fairly sure there are "mood" tags in the training data

I can't be certain, but it would seem a reasonable assumption based on my results doing the same.

you can get mood tagging ideas from rateyourmusic.com

where basically all songs are well tagged

[alpine venture]

omg my child

my sweet boy

my Anti-

[shy cosmos]

???

[alpine venture]

anti has become a popular prompt lately (I coined its use) and so it was sort of an inside community joke

my prompts often contain a litany of anti-X'a

X's

[shy cosmos]

Oh))) took this nickname many years ago))) but do you use this word? In genre specifications? In which genes? Like anti-rock or something like that?

[gusty urchin]

I'll wager that's in the pipeline.
In the meantime, what I want to do, is make isolated instrumental tracks; and then to splice them together in external software. Record vocal tracks over the top.

[untold briar]

One thing I posted earlier, I find oatmeal- about as strong a style word as anti- and I doubt songs were tagged with oatmeal. It's quite unpredictable what works sometimes! Well, there are for sure mood words, I mean mostly that it's hard to infer that words work well without just testing beyond the basics, very easy to be surprised I'm not sure that a lot the style prompts are very intentional.

[alpine venture]

here is an axemple https://app.suno.ai/song/b3b6ae12-bfe3-481d-86ae-b16f2d7ef6af

and another

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

[flint dock]

oh dear, I only want to change the lyrics of an already generated section, but suno keeps changing the melodies or even the motifs and most of the time it doesn't even use the new lyrics, whyyyyyyyyyyyyyyyyyyyy 😭😭😭😭

it is my only option to keep trying to generate the sections again and again and be lucky? please let me use the melody seed, voice seed and rythm seed for more control and FORCE suno to use my lyrics, please 🙏🙏🙏🙏

[azure crane]

Trying to work with instrumentals & it’s frustrating that I can’t find a metatag that stops the melody and just having the background music featured. Wouldn’t the breakdown be a good one for let’s say do this: [drums bass keys breakdown]? Seems though that it doesn’t work. I have tried intro but if the melody starts immediately then there is no way to continue with an intro further away in the song. If anyone here had an idea to manage to stop the melody like you can do with vocals then it would make my day 😊.

[azure crane]

Hehe, will do it now 😊

[autumn compass]

which style of music can do a disney channel type music?

[ocean salmon]

Hey, does anyone know if someone figured out how to switch between a male and female voice in a song? Aside from using other programs

[exotic fiber]

Check the wiki... U have an explanation there on how to create duets... However, it works sometimes... Try and error

[rare blade]

Where is the wiki?

[little sedge]

I am looking for information on technical prompts, do you know where to find this information?

[ocean salmon]

@exotic fiber Do you have a link by any chance? I'll start looking for it too

[rare blade]

I tried a bit, and got it to sort or work, it was interesting. I started with the male part, then did a continuation a few times and it sort of started the next section with the female voice which morphed into the male voice sort of. Curious what they say to do in the wiki
https://app.suno.ai/song/d4055b8a-7c20-4bcd-acd1-874476ff3675

Kind of picks up again with the female voice properly past the both section. I guess the model struggles with switching context between singers without reference to the other singer idk

Got smart and searched the discord for wiki

https://www.suno.wiki/

I assume this is the wiki

[exotic fiber]

@ocean salmon

[rare blade]

Duets seem like a lost cause...

[exotic fiber]

works 25% of the times (for me),, but also depends on the genre, for example, i have managed to get 4 different female voices when making kpop...also u can try using boys band or girls group in the tags...

does anyone knows when the page will be available for free users again?

i need my suno daily dosis before bed

https://tenor.com/view/byuntear-baby-cry-crying-gif-25019897

[opaque karma]

Ye fix it please

[desert basin]

anyone noticed that vocal clarity drops the more you string tracks together? I'm wondering if i should try to bridge in quieter sections

[exotic fiber]

thats a ptoblem of the model itself... degradates the quality over time.. nothing to do with that at the moment... try to keep your songs up to 4 min to avoid granular noise

[desert basin]

I'm doing a 30 minute epic so.. oh no

This is still absolutely incredible i'm shocked at how good even reading hackernews sounds

[karmic lance]

I thought suno was a god, but now it's become the devil of capital.

[solar garnet]

omg this so hard

[primal stream]

Hey guys! Quick question: is it possible to modify the lyrics of music already generated on https://app.suno.ai/. Basically keep the same rhythm, topline, but just different lyrics?

[mint mirage]

Did you try this?

[primal stream]

Yes, but it doesn't give me the same style / topline / rhythm / instruments. It creates new music. Whereas I indicated the same style: "French variety". And I used the "Remix" tool.

[exotic fiber]

use the: continue from this song option, and select 00:01 on the time stamp.. u will begin from the start and get the same sounds and melody, the downside, u lose the 2 minutes first generation because the contiues are up to 1 min...

[alpine venture]

glad you like it, it's off this album, so you'll likely like others. Specifically, if you liked that track, you'll like "where are you going" and possibly "mother"

https://open.spotify.com/album/2z3tPbsG8FlIY8N6Dwuptx?si=DFsItx0bR2GZTgQHMCkWOQ

[solar garnet]

you made this with ai???

[quick peak]

how do i download a song in mp3 format?

[exotic fiber]

I'm not entirely sure how to do it, but perhaps someone else knows how.

What if someone collected every message here on Chirp Technical and Chirp Bugs, and train a Language Model with that? We could then attempt to create a Suno chatbot assistant for handling issues. Additionally, since all the information here is relevant, as many people offer their tips, it could also help with that... right?

[alpine venture]

yes, although the lyrics are all mine

my poetry is linked here

I am lucky to have a significant backlog

[solar garnet]

Insane

[alpine venture]

at a time when it's useful

[exotic fiber]

May I ask, How complicated is it to register on Spotify as an artist to upload your music?

[austere mauve]

why suno Ai dont is oonly for pro and premiere now?

[alpine venture]

not hard

you don't do it through spotify

it's through a distributor ie distrokid

or similar one

they're the most well known

it is a monthly cost of, if I remember, $3.99?

something like that

[exotic fiber]

Awesome, thank you for the information!

[versed valley]

Welcome! For help, check out Suno Help

[exotic fiber]

so.. no free suno today?

https://tenor.com/view/angry-angry-cat-mad-cat-mad-death-stare-gif-2007099240114134459

[raven garden]

I am wondering: Is it possible to give suno voice snipetts so that it will use eg my voice for generation?

[mortal vortex]

ohh,that's a pity.I try to make a music but I got this message today.

[broken lotus]

that would be cool, because it probably could imitate our voices (up to a point)
but Suno is strictly text-to-music...

Look up RVC Replay, free needs a GPU, can replace the Suno voice with a cloned voice model.

[raven garden]

Hm, thought so. Maybe start by generating a very short clip with vocals and continue from that for new songs?

[burnt charm]

Hi all,

Could someone guide me on how Suno works?
Are we in a purely audio generation, wave generation, or melspec.

Or is it a multitrack composition in midi then exported as audio?

Thank you for your answers 🙂

[broken lotus]

Suno is generative AI, text-to-music prompting.
They also have a Text-to-Speech voice model called 'Bark' which the music bot probably evolved from (giving Suno a big head start with the ability to sing expressively from our lyrics)... There are no samples or midi internally

AI uses a large language model (like chatgpt) neural network trained on all types of music, paired with text descriptions.... like a Midjourney for music.

Suno hasn't released details about their training process

[ashen cedar]

As I'm trying to modify a song without using a "continue" each time.. any Idea how to promt it for example to switch voice or style properly?

[untold briar]

It's very hard to switch music genres entirely without using continue_from custom times. I think the only time I've seen it is in genres that kind of already do that, with maybe a tag like [changeup] or [switch] or something (I can't recall exact name off top of my head). Voices can sometimes switch between singers though. If you prompt for a musical and write the lyrics like a script (different ways of formatting, one example might be JONATHAN: Hello, I am singing! and another name for the other sings. But usually the voices switch sort of randomly (not aligning with the names) so it's more like using the names is a signal to switch voices generally

[ashen cedar]

Thanks so far. I generally struggling with directing it where I want because no matter what I try it makes something else - first goal was one verse male, one female but.. well.. only with luck 🙂

[burnt charm]

Thx a lot for your answer.
ok, so if I understand correctly, the AI generates a wave which has the graphic form of a track of the style in question as would MaGNET from Facebook. We take a latent noise and the AI gives a shape according to a prompt, here a waveform. It superimposes bark's text-to-speech with our lyrics based on the harmonics of the generated track.
I ask this because it would be interesting to be able to have access to several stems, the drum line, the bass, the different instruments as well as the vocals, for in-depth editing. In this way, if the generation was based on multiple midi tracks, we could arrange the song in future versions. Like a mini DAW. 😄

[exotic fiber]

Nope u cant, sorry

[azure crane]

Wonder a bit over the Continue generated part. Shall we delete the metatags that was used in the first part and add new once for the continue part? Would be nice if there was some more info about the Continue part.

[ashen cedar]

As from my experiments it ignores all but the text. Can't change any tags for that.

[vestal beacon]

hi what is chirp? how can I learn about an overall summary on what chirp and bark do? I need to cut clips, etc (for background music in a game) and then would like to create voice audio sometimes and I imagine bark and chirp help with this.. do I need to use a discord bot like Midjourney? or is this all done in the web interface?

[shadow aurora]

/chirp

[flint haven]

The /chirp command is not available in this channel

Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.

[compact stratus]

the possibility for stems would be nice 👍

[untold briar]

Bark in Suno's open source TTS. Most of this Discord is about Suno Music, which used to be called Chirp. /chirp is still the Discord command for Suno music. As of right now, the website version of Suno music is V3 and the Discord is only V1, so web site is preferred.

[versed valley]

Welcome! For help, check out Suno Help

[vestal beacon]

thank you!

[vestal beacon]

thanks. please keep adding more and iterating on this doc!!!

how do I know what is or isn't a meta tag? is the Using Metatags Notion doc exhaustive?

do the metatags need to be in the Song Description?

[azure crane]

Well, I’m using Outro, Refrain, End etc. and it works but I don’t know if it’s any idea to delete the tags that is in the first part that also will be included in the Continue part if you don’t delete them.

[neat field]

experiment: remix a clean vocal v3alpha song in v3 and see if the clean vocals carry over from the alpha version

[cold zenith]

this album is maide with suno ?

[alpine venture]

yes! v3 and v3a

the opening of "all the winged sparklies. crush underfoot." is v2

[cold zenith]

nice, is it on youtube? ( youtube is available for everyone)

[alpine venture]

https://music.youtube.com/playlist?list=OLAK5uy_nUb80MGnkxY8cXvcHpgqxr9USeTioZC88&si=lwKf573zvkgaqWWl

appears so 👍

it should be basically everywhere, also on iTunes etc

I do have some art on Spotify for it if you want a lil extra

additionally, I have some music videos on YouTube

https://youtu.be/dBypbWk7BXU?si=ZpmNO2tSIH7WEKIn

this is off the first album, v2

if you want details on how I prompted to generate these hang on

[lusty orbit]

Is it possible to get video with text? I don’t see it anymore - and would be awesome if I got a more karaoke style video instead.

[cold zenith]

how did you make those videos ?

noisee ?

[lusty orbit]

This is what I get today - no more lyrics. Buhuu

[alpine venture]

https://app.suno.ai/playlist/f9cd0dd1-8e31-4878-a71e-55277cce419b

this has all my prompts for the album

it's a playlist

may help some people with guiding their production

[alpine venture]

video was via other generative ai

same with album covers

I use sdxl, lots of different fine tunes and lots of different techniques

that video was via SVD 1.0 (Stable Video Diffusion)

this one is via Runway (I think it's inferior so I didn't put it on YouTube)

my weakness is I am very bad at likeable social media presence/fan engagement, and I have very limited time/resources. But the quality, at least subjectively, I think is decent for the work I've put up. If you have any questions about prompting, just tag me.

[karmic palm]

I wonder if there will be an “upscale” mode in future (similar to midjourney) where we can improve the fidelity of our current v3 generations

That will be so awesome

[dusty whale]

good

[chrome jasper]

https://app.suno.ai/song/9038baaf-4783-46ca-a2c3-bba4f20e2fe4

[alpine venture]

you can put that in suggestions

my intuition is for stuff like that to happen it would need to be open source

as that's a bit niche in this case

and won't be high on priority list

in comparison with inpaint

and new models

[karmic palm]

I agree I think upscaling fidelity maybe not be as important as doing proper in painting

[alpine venture]

I think inpainting is on the road map

I mean, I don't know for sure

[karmic palm]

Trying to continue from a song in the same rhythm or cadence is a challenge at the moment

[alpine venture]

but from things I've seen, I think yes

[karmic palm]

I made a mistake of containing part 2 of a song and not putting more lyrics. Turns out it generated a banger and now I’m sadz

[alpine venture]

you mean like a new song?

you can just split it off

or what do you mean, it generated a great Seco halt but you wish there were lyrics?

*second half

[karmic palm]

Not a new song but I wrote a bridge and it generated a good riff

Sometime like that. Except I already wrote a 2 part bridge but made a mistake of just putting part 1 in

[alpine venture]

aaaaqh

I see

ya, that sucks

[karmic palm]

It’s a high tempo song so trying to splice in the second part of the bridge is a challenge haha

[alpine venture]

yea, nah, in any case, give both versions a listen, sometimes mistakes result in a better song

[karmic palm]

I’m not giving up. Gonna try again when I get home on my pc as doing it on the phone is quite a PITA heh

[alpine venture]

if you take a step back from your original intent

[karmic palm]

Yep some mistakes have generated interesting results. Loving it

[cloud cedar]

hello, is it possible to choose the key of a song before generating it, thank you in advance?

[untold briar]

Basically no. There is a decent chance prompts with chords work more often than random chance, but that isn't very practical. That said, some user may yet discover more reliable ways of prompting that (or even already have done so and I didn't see it).

[cloud cedar]

thanks

[raven garden]

Hm, Do you also encounter the problem that if you continue your song, it seems to never know how to proceed and just gives you a 10s bit?

[chrome jasper]

https://app.suno.ai/song/4133df70-d8a5-4558-a678-0b6a65baf6c3

[wraith raft]

this most often happens to me when I continue a song too late, when it's reaching an outro or some bit of silence. works much better if i continue from the middle of a distinct series of chords so it has more data to start with, so to speak

[south dock]

hey guys, do you know if I can get an access to a chirp bot through an api?

I have a service, which give users an access to text-to-speech, text-to-music capabilities, but without bark v3 in the public my sources are ... nonexistent? Ahah seems like it

[maiden monolith]

When will chirp be available without a subscription again?

[broken lotus]

We don't know... 🤕 they are scaling up to accommodate the traffic.

[topaz estuary]

!chrip

!chirp

[versed valley]

Welcome! For help, check out Suno Help

[viscid lintel]

I'm loving the Instrumental Toggle that was recently added.
Does anyone know a technique to get vocals only? Just vocal track, no instruments

[short yew]

@broken lotusany idea why /chirp isnt working for some people?

[exotic fiber]

Still no Free version?

https://tenor.com/view/tantrum-gif-12615339064861248159

[tired tulip]

yep still no free gens

[final hazel]

understand your frustration. Maintenance periods can be inconvenient, but they are essential for improving the service. We appreciate your patience while these updates are completed.

[neat field]

this is my go-to line for down free service

everytime the free version goes down, a developer starts work on an open source alternative

[short yew]

just use chirp

@final hazelis there a way to put a title on chirp songs?

[neat field]

can i save someone elses public playlist to my playlist somehow ? a button i missed ?

[final hazel]

Currently, Suno doesn't provide a feature to directly save someone else's public playlist to your own playlist. You can create your own playlist and manually add songs you like from others' playlists.

[shrewd charm]

I've been playing with Suno for a few hours now, completely blown away. I've work with a lot of voice over synthesis over the years, but never successful music synthesis. I've been trying to gather a conceptual understanding as to how this system works. When I've tried generative music networks, I end up with so much randomness and lack of structure, even if the training sets have strict consistencies within them. This system, however, seemingly makes no mistakes, whether we are talking consistency in tempo, keys, chord progressions, melodies, timing, vocal harmonies, etc. I'm going out on a limb at thinking that the training data doesn't just generate a single model, but instead Many models, each derived from a specific piece of audio or a song, separating out information about vocal melodies/harmonies, and separating out information about music. Then, using that model, it appears to inference a new stream for the music, and a new stream for the vocals. The ability for it to time lyrics to melodies in a Human-Friendly manner makes me wonder if after the lyrics are generated, if a song model is chosen based on comparing syllabic placements of the generated audio with the model's lyrical syllabic data. Of course, this is purely a guess as if I was to try to develop this (which I have failed at for years). All in all, I'm so thoroughly impressed; hats off to everyone involved. If there is a white paper, patent, or even an old post (wasn't able to find one) on the technical approach, I'd love to read it and learn more.

[neat field]

https://arxiv.org/abs/2402.15294 for your research pleasure

my best guess based on the literature 2024 is their model is actually 2-3 models in a multi-step workflow

it's def ahead of State of the Art research community. very impressive Suno work

[shrewd charm]

Thanks so much, I appreciate it!

[shrewd charm]

It's still hard for me to think in terms of problem solving via AI; 25 years of procedural coding is hard to shake.

I love how this community builds upon each other's work. I am still struggling to comprehend SeqGAN and it seems this has made such significant improvements beyond it. My implementations stopped at VQVAE and GAN for vocal cloning. Thanks again for sharing this!

[neat field]

I would be interested as your research journey unfolds, come back from time to time here or #🗣️┃chill-chat to update

[junior island]

not sure if there is anyone here technical person can help? i am only my 3rd day to learn. i think i lost many credits that just want to know how to use the continue

[versed valley]

Welcome! For help, check out Suno Help

[hidden night]

/chirp

[flint haven]

The /chirp command is not available in this channel

Go to the #🐣┃chirp-beta-1 channel to use the /chirp command, and check out the #🐣┃how-to-chirp channel for step-by-step instructions for making music using Chirp on Discord.

[broken lotus]

Maybe this will help?
https://www.suno.wiki/faq/making-music/how-do-i-make-the-song-longer/

[broken lotus]

@wise creek @chrome jasper please don't spam this channel with songs. This channel is for technique and troubleshooting discussions.

[static wigeon]

hello, I didn't find out too much, but are there any magic prompts to change the rapper's flow?
I'm around 20+ generations (so 40+ clips) and I don't have any changes, I've tried lots of things but I must not have the right prompts 😭 😁

[static wigeon]

you can try something like "a capella" ? idk if it's possible :/

[broken lotus]

Maybe stuff like:

This-is-the-part-where-my-flow-is-fast I fly, high... no-bummer-chummers-choking-my-flow I-know, suppose, it-goes...

[static wigeon]

thanks ! i go try this

[hybrid wave]

Try adding [Voice Only] at the beginning of the lyrics

https://app.suno.ai/song/b63cb4e9-d163-488b-bc73-531fe9e9f24c for the choir i found a prompt that works with " multiple choirs " in song description

[surreal urchin]

有人知道怎麼解決嗎

Does anyone know how to solve this?

[broken lotus]

I think some countries are having issues getting through.
the 'fix' has been to use a VPN.

[dusky skiff]

is there any way to pay for a subscription using paypal?

[final hazel]

Not atm. They are working on it.

[dusky skiff]

thanks a lot!

[mossy robin]

hey guys do you have any issues with the song never fully resolving itself? Is there anything I can do to make sure the AI knows when its time to end?

[final hazel]

To ensure your song has a proper ending, you can use the metatag [end] in your song description or style of music field (but then without the square brackets). This tag can help guide the AI to understand that it's time to conclude the song, potentially improving the resolution of your composition.

[mossy robin]

sweet thank you!

[lyric sonnet]

how does the 'New' page work on https://app.suno.ai/ Explore? I generate a song and make it public but it never shows up there?

[broken lotus]

it probably needs to be a continued 'Full Song'

and set to Public right away...

[lyric sonnet]

it is.

there is a time limit on publication after generation? ok..

[broken lotus]

Seems to be, according to people who have got on the New list.

[lyric sonnet]

I assume there must be a much larger amount of songs submitted per second than end up on the New page, i wonder if some random selection is involved as well

[final hazel]

this is what a DEV said about it yesterday.. Not all end up on new...

[lyric sonnet]

oh wait, New is curated? 😲

meaning, someone listens to these songs and selects manually?

maybe auto-curated.

[mossy robin]

likely auto-curated based on specific factors aka no profanity, and likely from persons that have generated lots of tracks and are using well defined tags

[neat lake]

Guys the quality is awful right now, like seriously awful and seems to affect both V3 and V2. Anyone knows why that is and do I just wait it out or can it be fixed somehow?

[covert socket]

Messing around with free version but will VERY likely sub. Question: can one take a "whole song" (a first song plus 2 linked) and then add another (started from link 3 but it generated as a stand alone)?

[broken lotus]

I don't think so, if I understand you....
but we can always edit later in a DAW.

[broken lotus]

try stripping your style down to minimal prompting basic...
too long of a style prompt seems to strangle the voice.

[covert socket]

Also, some feedback to the devs (LOVE this app): I put in "70's prog rock in the style of Yes/Genesis" and the result was more of the "modern" prog style, i.e. a harder more "metal" like sound, when I was hoping for a more classical/madrigal sound. I suppose no big deal but something for future development for a style that the engine can learn.

[ionic wind]

One of my biggest disappointments with Suno is that it doesn't seem to understand the keyword "mellotron". At least I've never had any success with it. This despite the fact that I've actually heard mellotron being used in some of its generations

[stray walrus]

how to make suno keep the same chorus in continuation of the song?

it just hapeened to be first thing because it cut chorus in earlier continuation

[spiral wigeon]

did somethign change? suddenyl the quality of the generated stuff is way worse

[final hazel]

No, Suno doesn't update or change its model without informing users. If you notice a change in song quality, it might be due to different factors like input variations, not a model update.

[spiral wigeon]

i'Ve been just doing over and voer the same prompt with (almost) the same lyrics, so i dont think its because of input, because yesterday ive been doing that too and it was fine

hmmm okay

[lament zenith]

Any luck with this

Trying to do the same to get the flow to be faster

[static wigeon]

Faster yes, but for me i just try more generation ^^ I think it's my first clip that caused everything to go wrong :/
Try some prompt to be more faster i think

[lament zenith]

how did you get it faster

what prompt

[static wigeon]

I chose the easy way and I took the version that I found the best x) So to rap faster I can't help you sorry x(

try this :
(wetcircuit)
Maybe stuff like:

This-is-the-part-where-my-flow-is-fast
I fly, high...
no-bummer-chummers-choking-my-flow
I-know, suppose, it-goes...

[lament zenith]

It just repeats the words

[static wigeon]

if it ever repeats a part then you have to make it clear that you want a quiet moment where there is only the instrumental

you can try this :
[Build up]

[DROP]

[FULL ORCHESTRA AND ELECTRIC GUITAR]

[Climax Solo]

You can change Buildup and try break, turnup or maybe some new prompt like faster flow, fast vocalise, i dont really know,

for me, I just wanted a change of flow, not necessarily to speed it up

[lament zenith]

I see fair enough I’m trying to speed up the flow but also changing the flow you said you just used a version you like

[neat lake]

Interesting, will check it out thanks 🙂

[static wigeon]

Yes I took the version that I preferred from the ton of clips that I generated, with the prompt i send above

[brittle escarp]

why one of the two generated song doesn't have vocal, even though I have provided lyrics?

[final hazel]

Hey Phillip... When you generate music, the AI considers the style and tempo you've chosen, which can significantly influence the song's structure. For instance, if you opt for a genre or style that typically features minimal singing, or if you're working with a slower tempo like in a ballad, the AI might interpret this as a cue to introduce vocals later in the track, possibly in the second part.

[brittle escarp]

thanks! I'll try the "continue" for the one without vocal.

[final hazel]

Next time.. please do not post your question twice.

[brittle escarp]

My bad. didn't know if my question would get any answer.

[final hazel]

I understand, but we have rules here. The count for all. 😉

[cold zenith]

Someone got a tips for the skipping verse even when u choose with the minutes ? ( i'm losing my mind using 50 credit for that lool )

it's skipping 16 sentences...

[alpine venture]

so, basically once it skips

[brittle escarp]

I tried to let chatGPT handle it. But the generated song only has vocal on the chorus part, and no vocal on the verse 1 that is before the chorus

[alpine venture]

it can be hard to recover; I recommend repeating that section in your lyrics on the continue; if that doesn't work, then you can run an instrumental continuation.

often it will read it in that instrumental

even though you haven't included it, it reads it off the previous

[cold zenith]

This is crazy bro, everything was perfect until it skips 16 sentences....

[alpine venture]

yea, it happens; you can always just run the continue from there as an instrumental, it likely will continue reading and effectively you may get a reroll of that section

the issue is once it has skipped a certain group of words, you may struggle to gst it to say that paragraph later on.

like, it will continue skipping it, I'm guessing something go do with the way continuations interpret the previous snippet

but if you try several different methods, one will grab it

but if it sounds good, consider just cutting those lyrics

we get attached, but less is almost always jore

more

[cold zenith]

it has cut a chorus and the outro...

[alpine venture]

ah, you have style tags?

[cold zenith]

The songs is less than 2 min loool

[alpine venture]

you may be able to run the continue then and paste that section, but change the tags. this may trick it.

[cold zenith]

i'll try

[alpine venture]

Like, use some other tag, like [verse, emotional]

or whatever

and then you can just do outro at the end. Youu generate an instrumental with just [outro] as the style

or [end] if you want it go close out relatively quickly

[cold zenith]

I just tried the instrumental solution, it works but he went into a loop. But i don't understand why it works in instrumental but not in regular way...

[alpine venture]

I have a few ideas for why, but yea, it usually works

and you'll get some new repeats or cycles that eventually lead to an end

basicslly I think without a lyric sheet, it correctly identifies that it didn't read a portion

or is more likely to

[cold zenith]

why do you think it works there ?

[alpine venture]

and the predicted pattern

ends up presuming it is the excluded part

[cold zenith]

but how does he knows that if there's not lyrics ?

[alpine venture]

yes.

😄

it's a bit obtuse with diffusion models and machine learning in general. Sometimes the answers to such a minutae can't really be answered

certainly without me seeing the exact backend, how the model is trained etc

[cold zenith]

hum

[alpine venture]

I'll say I think that, when there is an instrumental portion

after a lyric sheet

there must be some songs in the training data that then have portions of the lyrics repeated, without it being repeated in the lyrics that match them

and thus is proceeds to follow any patterns it identified in that data

and probably the increased chance of it reading wasn't yet read

is related to other errors

in the data

[cold zenith]

This is exhausting fighting against the machine like said Xzibit

[alpine venture]

but that's just a guess

[alpine venture]

trust me, it's a lot easier than the alternative

also, it gets easier

[ashen cedar]

If you change voices it's a way greater chance it misses 1 or more rows from my experience

[alpine venture]

much like music, there is an element of mysticism ie muscle memory

but of a semantic type

if you miss, you can almost always get it if you address it right away

[cold zenith]

i think i'm gonna give up that music, i charge for 1000 i'm already at 830, just because of the skipping... i liked the song

Don't want to burn more for "maybe" at the end the results. I remember sometimes i burned so much credit for the same reason

[brittle escarp]

when using discord /chirp command, how to specify v3? I just found out the songs were generated with V1

[cold zenith]

use the website.

[brittle escarp]

thanks!

[cold zenith]

i won

From 1000 to 630...

😆

[outer furnace]

I have a few questions (V3 specific):

• I specified these parameters
  [BPM: 140]
  [Time Signature: "5/4"]
  [Verse 1]
  and the video seems to turn it into [BPM: 140, Time Signature: "5/4", Verse 1]
  should I roll with this from now on?
• Is this going to be applied for the entire generation? Or do I need to repeat it for every verse, choir, bridge, etc.?
• Will the continuation of the song remember that I started with certain settings or will it ignore them unless I specify them again (ties to previous question)
• How can I write actual instructions.
  ex:
  [BPM: 140]
  [Time Signature: "5/4"]
  [Verse 1]
  [the singer is sorrowful yet determined and the tension rises higher]
  In a sea of doubt, I stand alone,
  Fighting waves that push me to the unknown.
  Voices clash, they scream and they collide,
  But I hold my ground, I won't be denied.

I tried several methods:
(content)
[content]
{content}
and everytime, the AI sings the lyrics.

• Will [Annotation - the singer is sorrowful yet determined and the tension rises higher] work?
• what is a full list of annotations we can write and what is the syntax for all?

Thank you!

[final hazel]

Hey PeaceSoldier! For V3, BPM tags might not always work as expected. It's okay to put all your settings in one bracket, like [instrumental bridge, solo guitar]. Instructions work best in square brackets, but keep them brief—too much detail might get sung as lyrics! Getting the exact results you want might require some trial and error, so be prepared to spend a few credits to nail it. Managing Expectations: The current state of AI in music generation, including Suno V3, means there are limitations to how accurately it can interpret and execute very specific instructions. It’s often a balancing act between giving enough guidance and leaving room for the AI’s creative interpretation.

[outer furnace]

Thank you! I will do some more tests soon.
A few suggestions that I have and think will be really helpful would be:

• public playlists
• the ability to add our own pictures to the songs / choose from any of the components that made up the song
• when the video generates / the lyrics are added to the final song, it would be nice if it will reflect the cuts done in the song as now it still keeps the ones from the full uncut song
  On the same topic (but probably extremely difficult to do) an idea would be to add a rescan button that would make an AI scan the song and remake the lyrics of the final song

I really enjoy using the program. Thank you! 🙂

[final hazel]

Sound like something for #💡┃suggestions-web
I am happy to hear you enjoy using the program 😄

[outer furnace]

thank you I added them there

[chrome jasper]

https://app.suno.ai/song/dfa12327-cfea-4848-8a6a-9d439789c5c0

[obtuse sable]

hello

[errant sapphire]

#📚┃suno-school What could be the cause that a continuation of a song is always failing?

[untold briar]

Try using the Continue From feature that lets you choose a specific time in the clip, and using times before the end of the clip. (Especially if you can hear in the clip that it fades to silence, then choose a time before that.)

[errant sapphire]

That seemed to work, thank you very much. Solution was 5sec earlier in the continue from

[gaunt scaffold]

校园，追逐梦想， 青春活力，朝气蓬勃，永不言败，男女对唱，女的柔情，男的阳刚， 汉语演唱

[broken lotus]

請使用新網站
https://suno.ai

[onyx umbra]

Hello. I've bought the Suno Pro plan and then added Suno server to Discord. I've already connected Discord account to Suno, but when I use /info command it shows me that my plan is Basic free subscription. Can you please help me?

[last grove]

Hi, where can i learn more about the underlying architecture behind chirp? I'm an AI student so would love to know more about the technical details

[mossy swift]

https://app.suno.ai/song/b7b1e3f9-f488-4a74-91b9-f7f206f598a8

[grave yarrow]

chirp

[versed valley]

Welcome! For help, check out Suno Help

[atomic garnet]

Are song likes and dislikes used to reinforcement learning this generative model?

[eager moat]

Chirp

Маленький мальчик играет в

[broken lotus]

The Devs have said 'yes', but did not elaborate.

[graceful condor]

how do u move your credits to this discord server to make more songs i have 40

😭

[alpine venture]

https://app.suno.ai/song/d8497988-9402-4742-91e5-2c28a2e562bd

Guiding instrumentals

using lyrics

[patent tiger]

?

[final hazel]

Do not ping the team.. It is not done..

[patent tiger]

I didn't see a role for that so I thought it was ok to ping them.

I could have just removed the @ if you told me

So is there a solution or a setting to fix it?

[final hazel]

Bugs go #🪲┃bugs-web And they read all.. If you ping them, the might feel there is a server down and they have already have a million things to do and they deserve also a weekend for themself

[patent tiger]

But I'm don't know if it's a bug? That's why I asked?

[final hazel]

That is for the Devs to decide. If you expect crystal clear sound, then I have to disappoint you. They are working hard on V4 to get a better sound. Each version is getting better.

[patent tiger]

ok

[static wigeon]

Hey ! i hope you're good !
You have an idea of the day for v4 alpha ? 😄

[final hazel]

https://tenor.com/view/ej-sweevilej-crystal-ball-my-crystal-ball-is-on-the-fritz-gif-12049040

[final hazel]

No clue.. sorry

[static wigeon]

eheh np 😄 i try my luck x)

[simple willow]

is it normal to get outputs that seemingly randomly switch back and fourth between songs within the same song? 👀

[versed valley]

Welcome! For help, check out Suno Help

[kind holly]

I couldn't find this anywhere online. Does the chirp model directly generate audio waveforms, does it generate midi which is then output as waveforms, or does it do something completely different to produce the instrumental part of the song?

[exotic fiber]

Audio quality decreased after the last update? getting more noise and granular noise as before

[final hazel]

there were no updates on the model

[exotic fiber]

weird then... do u know what have they changed?

[final hazel]

infrastructure only.. more room, more gpus etc

[exotic fiber]

this is the best i managed to get in the last 6 generatios, the prevoius ones sound even worst

listen onto 1 min 16

to much noise...

and im not overcharging the tags, i just have this on the style: kpop, upbeat

[stone spruce]

this is great

[warped cove]

https://app.suno.ai/song/cd580a17-bcf6-4728-bf93-a33382cf4b5d

[eager topaz]

It's hard to get good, clean vocals in any genre, I'm finding. Too much processing, adding harmonies that just increase the noise, a sound like double tracking that also adds noise. I'm hoping it all improves, but sticking to instrumentals for now.

[exotic fiber]

i was managing to get better and cleaner vocals... so, maybe right now is just a bad seed bad luck mixture im getting :p will try again tomorrow.. :p

[rose ore]

yall, 9/5ths timing is so beautifully chaotic, you have to try it to appreciate it

[fossil hinge]

Maybe this is better posted in technical:

I've been poking around the app a bit – things I've discovered so far:
Running on Clerk – https://accounts.suno.ai/user
https://accounts.suno.ai/create-organization

You can extract the User ID + "concatination IDs" (other IDs of songs that were joined to generate a full song) out of a song's page.

[ionic wind]

Has anyone had any luck getting Suno to generate Future Garage? It seems to only generate variations of house when I try it

[patent trail]

Im a new user to both Discord and Suno, what value is there to using chirp in discord if i can go directly to suno.ai and create?

[shut pollen]

seeing as discord's chirp is v1 and suno's running v3, none

[patent trail]

Thanks Shawn, that makes the decision easy.. 🙂

2nd question: I am trying to make Promo's and Jingles for streaming rock radio broadcasts. I want 10 to 20 second, how do I get it to give me short creations?

with that question, is there any metatags for different types of sound effects ?

If you want to switch from male to female voice do you put the tag in the lyric before the lyric ie. [female] he is my guy , [male] she is my gal or is there another way?

4th question: is there a correct way to tell it to match the audio suno created in the previous clip? I have audio music that is great but the vocals suck, and want to alter them, however use the previous audio, Suno usually doesn't even get close to the previous audio

Final question, I have tried to get it to create the lyrics using my lyrics. Making small changes off a lyric suno created. Sometimes it seems to understand and mostly not, so like the audio how to I force suno to stick to what I wrote?

[alpine venture]

Well, my experiments to clear the voice created a country song with a rock progression

https://app.suno.ai/song/692b3e5c-947d-4ada-9202-b47a0d1dd393

noise explosion

at around 1:"40

that blew me out my socks

[late dome]

I'm a newbie

[shut jetty]

https://app.suno.ai/song/a91c149e-ff4b-4545-90b5-dbd482082f8b

[full vale]

Got a song I've been working let me know what you think https://app.suno.ai/song/fb1e6302-250b-4c16-94d0-fd4ccab298c3

and another one that took me along time https://app.suno.ai/song/e96c85df-8f3b-4a95-9662-7de04e5efa97/

[dapper imp]

Is it possible to add the Suno Chirp bot to other servers?

[broken lotus]

yes, but it is just v1 on Discord.

[dapper imp]

Is v1 very limited or is it still worth adding it to other servers?

[broken lotus]

it is only 40sec max, and the quality is pretty low, compared to current model (v3 = 2min and stereo)

[radiant hill]

Always this disgusting background noise when the intensity increases Suno hasn't done anything about it yet

[ionic wind]

What do you expect them to do? The only thing they can reasonably do is train new and improved models, and that takes a ton of time and server resources (we're talking months to train a single model in many cases, without knowing for sure if that's the situation here), and even that is far from an exact science due to how these AI models work. It's not like they can flip a switch and get rid of the noise. Temper your expectations.

They might come up with a clever way to deal with the noise, spend weeks training the model for this, only to discover that the results are subpar in other, unexpected ways.

As someone who has been an AI enthusiast for several years now, I've learned to be patient. We're in the infancy of this technology yet, at a stage where we normal people would typically not even be exposed to it. Except AI tech is so revolutionary that we're all drooling after new prototypes, which is almost unheard of.

[radiant hill]

I'm just talking about something that was not generated and suddenly became so. So they can at least go back to the previous operation or look for where and when the problem occurred. I'm not trying to philosophize about AI...

[ionic wind]

I'm not sure I understand what you say. Are you talking about songs getting increasingly noisy as they progress? This is a recurring problem for basically all AI models at the moment. I was in a business presentation by a Microsoft representative who talked about how ChatGPT became increasingly confused and even argumentative the longer you talked to it. Generative AI "self-pollutes" in a way that decreases the quality of its output the longer you push it. All you can do if you run into this problem is to retry and hope that the dice land in your favor, but the longer the song is, the more the odds are stacked against you.