💡┃suggestions-web

[sweet jasper]

upload your own music and extend it?

(also first)

[sharp venture]

• Build Architecture that allows for Highlighting and Changing sections of a song

• Split Vocals from Music

• Split different instruments that can be edited individually

• Ability to add and remove instruments/vocals

• Change song artwork

• Add pauses, tone shifts

• Allow us to choose where the song starts. Aka, At the intro

• Generate full songs with desired length.

[cloud crane]

• Every instrument as a sperate file
• Midi!

These two are enough to make this a boss app

[spring iron]

• Acapella only
• Instrumental only
• Tempo

[fringe knoll]

a cappella and instrumental already work...? I have been using them.

[cloud crane]

upsclae the sound quality for x tokens

but this would only work if the instruments are seperated, then one can master it properly

Instead of making it generate 30 seconds, make it generate so-many-beats so you dont get these weird cutoffs or necessity to make transits

[cloud crane]

for instrumental just leave the text blank

[fringe knoll]

not after lyrics, tho.

you'll get a 2sec clip

[tepid ridge]

@mental horizon Suno startup inception program 🙏

[cloud crane]

Ah yeah i noticed that, seems a bug

[shadow oak]

can i use the music generated to post on spotify and stuff? i am on the free plan , is it just for the paid plan?

[fringe knoll]

Spotify (and other music distro sites) is commercial tho.

[robust saffron]

You have to be on the Pro plan to use it commercially on Spotify, SoundCloud etc

[uneven yarrow]

Hello Everyone,

When I am trying to generate a voice from a text as of now it's taking longer time to generate rather than the generated voice length.
For example, if I am generating a voice of 30 seconds, it's taking atleast 30 seconds of time. Doesn't matter which GPU I am using. I tested this with Tesla T4 and A100 as well. There is minor voice generation speed increase.
But when I use Microsoft Azure api for voice generation for 30 second of voice it is taking hardly 2 seconds. Even for 1 min voice it's taking almost same time.

May be I am lagging somewhere, please let me know update over this.

[warm quiver]

link ChatGPT a knowledge base on meta tags

[strong shore]

I think it would be nice to add the possibility to make the music pieces "continue" with music files not generated by Suno ai 😃 that would be revolutionary !! not just being able to enjoy the possibilities of unfinisched music works, but also it would make it much easier to benefit the best qualities of other Music ai generation models. That would be great

[feral aspen]

Suggestion: Discord voice channel with a livestream of Chirp songs generated in real time. (Also YouTube, Twitch, etc.) With Bark/ChatGPT as the DJ, of course.

Would this be a moderation nightmare? Yes. Would this be awesome? Also yes.

[digital tusk]

А бесплатные кредиты когда нибудь могут появиться????

[sudden nymph]

Suggestion: make it possible to add a song as a refeence

[prisma sigil]

Suggestion: it doesn't understand the r&b music style or when we make songs in Turkish, whatever we write, he raps at once

[delicate inlet]

Can you guys create a feature similar to the Midjourney described feature, where enter a picture, and the chatbot creates a prompt based on it? Only this feat would create or capture the vibe or theme of the track you submit

[fringe knoll]

image to music 😮

[feral aspen]

Yeah I like this and I think it could work well because it doesn't have to be accurate, it can just be an influence on the mood of the song, the overall emotion or style, which is probably pretty doable.

[fading crater]

my humble suggest: Instead of just limiting the number of freebies per account, why not give a few freebies a day to maintain more engagement and connection with a wider audience, while increasing quality and functionality? potential for the product, there will be a stronger community

[delicate inlet]

That would be awesome to lol

[bright badge]

Suggesstion: make it possible to always use the same singer voice. (With a voice seed or something)

[jade stone]

Would love to see more direct control, like [bass drop], [segue], [bridge], [screamed], [whispered], things like that. It seems to semi recognize stuff like this, but not consistently

[glass fractal]

the voice of singer changes alot after each continue, can you make it to be fixed... I made a 5 mins song, and the woman singer become a man by end of the song

[cyan olive]

Hi. Can it be possi le to make the continue option base on previous data with the same voice. I try 2 time input the prompt and it came out the same verse and chorus. I was just following a youtu e video i saw and it didnt came as i was specting an this was a great idea

[quiet hawk]

Make it possible using metatags to repeat certain parts of a song multiple times (with exactly the same lyrics, style of background music)

[stray yacht]

/chirp

[mental horizon]

Hey @stray yacht, check out #🐣┃how-to-chirp to learn how to make music using the /chirp command. The TL;DR is get started by...

1. Going to #🐣┃chirp-beta-1
2. Typing /chirp and pressing Enter

[stray yacht]

Yeah I know, thanks. Must have typed enter before the prompt appeared.

[jade stone]

It would be nice if when using "random," chirp would include what style it decided to go with in case we want to use a similar style in the future.

[astral plover]

A website for suno

[formal bear]

There is one -- https://www.suno.ai/ -- guess we should have more content up there

[jade stone]

An ability to repeat sections during stitching, so we can actually have coherent verses and chorus lines would be excellent - just would need to change the intro of each section for flow purposes. Also, maybe a way to match lyric and tone from a certain snippet and lay vocals over a new one that just introduces new instrumentals and backing.

[frank wraith]

It would be nice if the content of Style didn't leak directly into the chat gpt lyrics (meaning lyrics that directly quote what is put into Style)

[leaden marlin]

The continuations don't match the song?

[soft vigil]

Instrumental only

[jade stone]

It would also be nice to be able to port projects into private servers that we've done in public.

[leaden marlin]

Download the songs in audio format

[solar sable]

Thanks yeah should definitely add a way to at least continue Workflows privately, maybe through a video ID or something

[solar sable]

Yeah should probably add a button for that. In the meantime just use an online conversion tool or something

[solar sable]

Was this in English or other language? Usually it should work seamlessly but sometimes there are weird issues

[solar sable]

Should already work by describing the style as ‘instrumental’

[solar sable]

Yeah right now the model only has context/memory of about 40s so for longer generations it kinda drifts along, so repetitions become difficult. Definitely planning on extending that soon

[green raven]

que tal no estilo musical random colocar qual foi escolhido para cantar tipo random(Pop), para saber qual da para ficar legal

[cloud wadi]

Had some ideas

1: control chord progression/key of songs.

2: Control tempo and time Signature.

3: Control vocal style(talking, singing, scream/growl)

[fringe knoll]

[Sprechgesang] and [Singing] work as prompts:
#🐣┃suno-showcase message

[cloud wadi]

What was that first one?

[fringe knoll]

it's the fancy word for 'talk-singing'

I think people have used [Spoken word]...

[cloud wadi]

Ooh ok

[cerulean coyote]

we need this

[feral aspen]

Suggestion: Embed the lyrics prompt text as soft subtitles for each song segment in a rendered out ❤️ long mp4 clip versions. Doesn't need any time synchronization (though that would be sweet, of course) but just a single subtitle for the entire 30s clip, all the lyrics text dumped in it (even the tags, no filtering, keep it simple) even if half of it wasn't sung at all. That would still nice to have readily available and lay the groundwork for fancier videos to be rendered from the songs later by users without having to manually input the original lyrics. Update: For example music videos #🐶┃bark-showcase message

[unborn token]

add an option to continue with the same music/singer/beat but as a new song. many times I get perfect music but some small stuff makes it useless, then its lost forever never to be replicated again

[gray vapor]

Character.ai and such services have these 'characters' set ups , I suppose they are like ChatGPT custom instructions, theoretically creating an AI personality or band and maybe even an album profile could be used the same what to accomplish this , I feel like there needs to be another AI to compare/edit the outputs for quality control. Or a reprocessor you can tell "remove hiss", "auto-tune flat vocals", like Stable Diffusion's instructPix2Pix

[fast tide]

I had an amazing singer in one of my generations that I wanted to make a duet with once I got it into my daw edited..but she's gone forever. Maybe if we had a "seed number" to reference that would help in consistency

[gray vapor]

This technology in a single box, so I can go into the woods and never come back.

[jade stone]

Well, what I mean is the ability to go back to a previous clip and heart it or something similar, and it'll insert it into the next section, only changing the first 3-5 seconds so it matches the section you're attaching it to.

[solar sable]

ah no, that we don't support (yet)

[gray vapor]

100% need an full service app like iTunes or a website like leonardo.ai. This is freakin' HUGE.

Mix some videos of 'Best of Suno.ai for (the week or month)" that mix the best tracks you pull from the daily theme on some repeating date, give the 'best' users tokens for their participation. Build a big library of tracks on youtube/tiktok/etc, could be 2 and half minutes each or...30 whatever your schedule permits. You can probably have an AI handle this from tracks you add an emoji to.

[royal cove]

An event to choose the 3 best songs made with suno would be incredible. Maybe worth some premium like a special role, or even nitro for example

[fringe knoll]

AI Eurovision sponsored by Suno!

😋 make it happen!

[steep dock]

Paid mode to get full length songs and/or continuation that is actually coherent

[delicate cedar]

A button which when submitted, allows you to somehow "quote" a previous section of a song. This could be a boon for people who get a chorus they really like and want to get either that specific arrangement again in the song or want to get as close as possible to that specific chorus again.

[solar sable]

slated for next model release 🙂

[solar sable]

very cool. hopefully you will get that for free once we have longer context in the model. internally we are prepping the next model release with longer context and there it will correctly pick up previous [Chorus] when you use the tag again. hopefully that would solve most of that use case

[delicate cedar]

Also it might be nice to be able to give higher emphasis to certain words in a song. Something like:

Oh, I *said*

to indicate the emotional weight or emphasize certain words.

[autumn shell]

is it possible to get constant bpm for long track? now it's floating a bit

[silk iris]

It would be better if it were unlimited, right?

That way it would attract more people, you know?

[delicate cedar]

lulz

I wouldn't mind a larger subscription though. Something that was more than the current amount and had unlimited but maybe slightly deprioritized generations. I won't name a price for that but something like that would be nice.

realizes how self-serving this suggestion is.

[silk iris]

Yes, it would be very good and better for those who don't have money

[delicate cedar]

A way to "tip" people via specific reactions using credits.

This would allow people to show appreciation for well done songs or song segments, and also when making a suggestion to someone, this would allow testing that suggestion to be free if the person making it felt like tossing a credit towards the attempt.

[covert oar]

An idea of when the next update is would be nice. 👌

[pliant rose]

Im thinking a section where if you dont like at all what was brought up. then one should be able to reject it without it counting against credits. Also some suggestions like why the prompts made the rejected clip sound the way it did.

[ancient igloo]

yes please!!! or even train a model to differentiate between reasonable music and unintelligible garbage so it can return our credits 🥺

[delicate cedar]

This is going to be a bit of a mess so please forgive me and bear with me a little as I meander to the point. I recently began to get really frustrated with some of my attempts to generate lately and feeling as though everything I made wasn't as good as the earlier stuff I was doing while still learning how things worked. Then I realized that my problem wasn't anything changing in Chirp, but in me changing how I prompted that was getting me results I liked.

I used to basically dump words in, pick a genre or music description and get the two options, picking between two until I got the "voice" and music I liked. What changed? Me trying to get something with a decent intro so it doesn't just begin with the song vocals itself so abruptly!

So my suggestion is this: Is there a way to create a 'remix' mode something like what Midjourney allows for images to take a song segment, select it for a retroactive use as an intro, extrapolating from it the style and music type and basing that to be the prompt upon which an intro is made?

I'm not suggesting that the segment become the intro, but rather it be some kind of special or super prompt that allows an intro to be based on it. If that makes sense? I'm not a programmer but I know enough to understand a lot of what might seem simple is anything but when it comes to coding or implementing such a thing. Thanks very much for Chirp.

[fringe knoll]

A similar discussion suggested the sampled bit of music could be divided between 2 clips... (it wasn't suggest this would actually happen, I think it was more like speculation what COULD happen)

As I understand it (from being a nosy person who reads the devs' comments too closely), the first clip is 40 seconds because that's the actual memory length.
the Continuations are 30sec because that extra 10sec is used to sample the previous clip.

What does seem certain (because it keeps being mentioned) is they intend to increase that memory length... so with more memory, this kind of dual sampling from 2 clips seems more do-able....?

(if that helps to make your suggestion better somehow?)

[feral aspen]

It's a good idea and along the lines to things I've personally suggested. I especially like that it's simple, just pointing at other samples as references, something anyone can do without deep music knowledge. Hope to see it at some point!

[delicate cedar]

It just seems like a limitation built into the thing that you get easier to work with results from a direct lyric prompt, but then you lose out on the intro so all songs seem abrupt. Being able to extropolate backwards would appear to be a solution that lives within those limits. Again I'm not a programmer, I'm sure there are all sorts of reasons why this won't work, but it was a suggestion in case there's something that would work this way that is accessible.

I don't know. This could all be entirely subjective to what I think works best for me and not apply to anything else anyone else is doing.

[delicate cedar]

Another thought I had...

Since we're all supposed to be learning from each other. Is there a way that the output of the various #chirp-beta-x channels could have a filter for languages? Possibly just needs a role added somehow? I'd love to listen and watch how other people do more stuff but it's harder when there are so many languages I don't understand mixed in with everything else.

[rose seal]

This is a good one, we're working language specific channels - check out https://discord.com/channels/1069381916492562582/1154766285960392715

[fringe knoll]

agree, I like a long [Intro]...

being able to reference a clip outside the current continue chain would be great

(or upload an audio file of our own)

[delicate cedar]

Oh yeah, that'd be great too. I was mostly thinking of for intros, but yeah as a general feature, being able to tak a clip generated from something else and rework it in would be awesome.

[delicate cedar]

OK, how did you make that happen?

Because I don't see that link anywhere else.

[rose seal]

Yeah tricky with Discord but it's in this section of the sidebar

[delicate cedar]

I feel like I'm missing some stuff?

[fringe knoll]

yeah..., not seeing that entire category either

[delicate cedar]

Moderator or Team channel?

[rose seal]

We'll look into it!

[delicate cedar]

Suggestion: Silent callback.

Since sometimes the AI needs to see or hear what it did in an earlier part of the song sometimes in order to keep its context alive, maybe it could be possible to prompt a command within a lyric as a meta tag to callback a line silently?

As in it's there, the AI sees it and uses it but it doesn't actually appear in the song segement it generates. It's silent in that way?

[rose seal]

Yeah, in general we're working on longer context which should help a lot with this and continuation issues!

[delicate cedar]

I apologize if it feels to you devs who know the internal workings like I'm beating a dead horse with these suggestions. To me they're all different things, but what I am hearing back is context, context, context. So I apologize if that's where things keep up ending.

[rose seal]

It's a bit of a research initiative still so TBD on when. But we also run into this when generating stuff ourselves

[rose seal]

Definitely helpful!

[gleaming compass]

There could have been Turkish language support, it was also very excellent.

[rocky lichen]

Is there any way to generate non musical voice? Just people talking?

If not, that would be cool. Maybe setting the style to None produces talking instead. It would also be fantastic if you could say things like Wisper or Woman with a frantic voice.
#very cool app. I have been looking for something like this! My attempts to clone the bark model and test were less than impressive 😆

[feral aspen]

Try formatting it like a script, or a musical, or opera. silly example. You can get one that are 90% non music, though I don't have one handy. Edit: Actually maybe this one from Chirp V1 is good example, second clip. This will overall be somewhat hit or miss, but it is quite fun.

Oh and for the music style, try using blank or - in style field. That means it uses the lyrics as the style hint, which I think makes it more likely to be speech only here. When you use the word 'random' it's using a style selected from a random set I believe, which is different.

[mint agate]

I suggest specfic channels for the most popular lanuages used.

Oh it already exists?

[feral aspen]

Yes, I think discord doesn't make it clear when new channels are added, but there's a channel with threads for each language now

[feral aspen]

Much funnier example #🐣┃suno-showcase message

[rose seal]

Hey, you can check the language threads here https://discord.com/channels/1069381916492562582/1154612901030678578

[rocky lichen]

It looks like the #🐶┃bark-beta does standard voice talking

[feral aspen]

Oh yeah, I assumed you were trying to use Chirp, but of course you use Bark for straight up voices.

[rocky lichen]

The ability to download MP3 would be cool. I’d also love to see cross integrations with MoonValley and Chirp / Bark. That would be cool to be able to make a song or generate voice-over and pair it with a video.

[mental horizon]

@delicate cedar, can you try checking Show All Channels and let me know if you see the Language-Specific Channels category?

You should also see some other channels appear under the Docs, Guides & Resources channel

You can also go to Channels and Roles > Browse Channels and select other non-default channels, which is another option

If anyone knows how to fix this, please shoot me a DM! Would be much appreciated

[delicate cedar]

I'll try when I get home again.

[feral aspen]

Hmnnn lots of general Discord complaints in the reddit that users don't see new channels, causing mass confusion sometimes. I'm not familiar enough with Discord myself to understand what's Discord policy and what's a bug in the comments here. Seems like you should be able to set all new channel in the onboarding to 'default' and they should be on by default.
https://www.reddit.com/r/discordapp/comments/126k3eq/the_show_all_channels_feature_is_literally/
https://www.reddit.com/r/discordapp/comments/16qahfb/new_channels_are_not_added_to_existing_members/
https://www.reddit.com/r/discordapp/comments/16owppr/onboarding_seems_brokenforcing_users_to_reselect/
https://www.reddit.com/r/discordapp/comments/163bz4h/this_update_is_so_awful_why_do_i_need_to_manually/

[feral aspen]

Having every single channel visible by default to every new user wouldn't actually be that awkward in this particular Discord. Not too many overall channels and they are all basically about the same topic. Even the non-English channels are just two total channels, not a big deal if English users see that category. I've actually had to check and enable new channels a few times myself because they were somehow defaulted or flipped to hidden, though at the time I assuemd I accidentally clicked something.

[delicate cedar]

First to prove that I have it checked.

This is what I've got.

The English under language-specific channels is from after someone linked me to the section and I made sure to follow from there.

[compact lark]

genre-specific channels we should add:

• pop
• rap

[compact lark]

❤️ I upload an audio of me singing a song. Suno analyzes my performance and makes his own version, improving my voice and correcting my singing mistakes.❤️

[frozen trail]

About music style, How many styles have?

[delicate cedar]

All of them.

Seriously, if I was able to tell it to use James Blunt as a style...

Just try it and see if it works for you.

[fringe knoll]

unknown... it is an LLM (like chatgpt) but not so big... and it knows many languages.

It was not trained on bandnames or music artists, but it does seem to know many of them....

[leaden marlin]

I posted this in general but I was asked to post it here instead: I’m not sure how to word this because I’m not a musician but I’ll try. You see how there are now apps that let people transform their drawings into more complex art? Like those apps that recreate a person’s doodles into artwork? Would it be possible for Zuno AI to do this with music in the near future? I’ve always had ideas for songs but I never knew how to make the songs because I don’t know a lot about music but I think AI can solve this problem?

[austere vale]

how do I find the songs i have created again

[broken flint]

how to create SFX and background music

[jade stone]

My biggest issue here, and I feel like I'm repeating myself, but seriously - coherence and consistency. I need to be able to create standardized choruses. Music can change a little but it needs to more or less have the same tempo and voice. There also needs to be a way for the AI to "lookback" and sample the original voice, as the longer a song gets, the more and more synthesized it begins to sound. The degradation is unreal sometimes, even with consistent style tagging.

Also, every song begins so abruptly. [INTRO] only seems to do so much, and only works at random. It does the same with cutting off at the end, whether I have [outro] or [fade] or not.

[feral aspen]

You can log into the Suno website, of you check your history of Discord mentions on this server (the bot pings you when done). It's the 'inbox' button in Discord.

[feral aspen]

What are you imagining the 'doodle' is here? Humming a melody? Or just a more detailed text description of music?

[vital laurel]

If their code only was open source, I'd be able to contribute a lot.

I have a similar, albeit drastically different and I believe far superior system. I also have the right music to train it on.

I'm creating an ensemble of 50-100 autonomous Sumo musicians each with their own style.

They'll be able to recommend Sumo to others, and also sell services/stems/MIDI on its behalf.

[fringe knoll]

🧐 not sure how this is a reply to anything I said (or how Suno benefits by giving their technology to facebook), but ooookay. 👍

[random python]

It would be nice to upload an audio file (let’s say just the vocals) and suno can fill in the background music

[leaden marlin]

That’s a really good idea.

[rose seal]

Yeah, we feel this too when we're making stuff - coherence/consistency will be a focus for the next major versions!

[covert oar]

Rock is the #1 Genre in the US and is missing from the Genre channel. Also Pop. Mini updates are meh. When can we expect a real update?

[jade stone]

Really guys? Censoring "Fuck you?" You know how many death metal songs have fuck you in it?

[fringe knoll]

and this is a problem with Suno or with death metal...?

[delicate cedar]

So? What's wrong with phuck or Puck instead? Be creative!

[jade stone]

It is when I keep getting "This prompt has been flagged."

[jade stone]

Yeah that was my workaround.

[delicate cedar]

🤣 🤣 🤣 🤣 🤣 🤣 🤣

Oh, God...I can't breathe...

[jade stone]

Lmfao! Nice! I actually got it through by adding some "oooooh haaaagh!" Between fuck and you lol

[delicate cedar]

Yeah, well, I had no pucks to give so...😂

I didn't really care if it worked or not but I'm impressed despite myself.

Wow. I love this tool.

[zenith bramble]

"continue" producing full silence or almost silence
#🐣┃chirp-beta-5 message

[delicate cedar]

Hmmm...

Actual suggestion: A tool to analyze a completed song and determine what genre its resulting song is.

I know that we tell it genres and styles of music when generating, but I'm not entirely convinced that the resulting song is completely of that genre by the time it's done. It'd be nice if it were possible to push a button at the end when the song is stitched together and have it give percentages by genre. Just a thought.

Incidentially, does anyone know anything that will analyze a song and determine by the waveform or some other method intrinsic to the music the genre a song is?

[zenith bramble]

thats a cool idea

[delicate cedar]

It seems like some work has already been done to make something like this too, this github seems to show a tool that might work: https://github.com/Pedrohgv/Music_Genre_Classification

Of course it's three years old at this point... 😵

This is a bust: https://sonoteller.ai

It couldn't identify any music.

And it seems to think there's a bunch of explicit language in it too.

[jade stone]

I had a similar thought the other night, but more internally with chirp - sometimes I don't really know what genre to put my music too, so I'll use random - instead of random showing on the output, it would be nice if it gave a result for what it actually chose so I can switch styles and then go back if I want to.

[delicate cedar]

Oh that's a good one!

This right here. That's definitely a good idea. One of those "papercut" ideas that should be easy to fix one would think.

Hmmm...I suppose it was reading 'Funky' as some other word?

(I just went through my lyrics to make sure I have them.)

[delicate cedar]

Sonoteller found one, couldn't detect music in another...sigh...

Another one hit, so the pattern would indicate this next one will fail?

[delicate cedar]

I take it back! sonoteller.ai may not have been able to tell me everything, but it definitely gave me hints towards what my style and genres are. Much appreicated and useful

[fringe knoll]

Yeeeeeeeeaaaaaaahhhhhh [guitar shred]
is going in every song now.😆

and if the rule is that we can fuck..., but not fuckYOU... that sounds like a better world to live in, lolol

[delicate cedar]

Suggestion: Monthly or weekly statistics.

I'd love to know generically what the rate is for continuing users to discard something before making a final song. There'd be some false positives due to people stitching stuff up to listen for changes in the song connectivity but it'd be interesting to see how many times people reroll before they're happy with the whole piece and how often people create something and simply discontinue it.

[frigid timber]

https://sonoteller.ai/zqRyRxUYAfY

this is awesome, I have never heard of this tool before, but its super spot on. Thanks for sharing it

[idle glacier]

yo how can I throw a complaint about someone else's generation where there is a humiliation of honor? I will be grateful for the answer

[delicate cedar]

I'm kicking this back up for visibility. I just heard something by a guy who impressed me and I'd love to toss a token or two their way, as a kudos. I don't know if they're a free user or if they have a pile of credits to use and it doesn't matter that much. But I know there are a lot of people who for various reasons can't get a subscription. And I'd love to be able to 'boost' their efforts with a 'tip' from our tokens to say "Hey this was cool, go make some more cool stuff" and have it work out.

I'm sure there are some issues which need to be worked out, as tokens can easily be considered money, since we pay for some of them in subscriptions. But since AFAIK there's no way to cash them out, isn't that a non-starter issue for stuff like money laundering and other issues of income? The only economy this should effect is the creative one, which is the only one I care about right now.

Thanks.

[feral aspen]

I'm no expert but I don't think there are legal issues just from this, gifting subs on Twitch for example is common place. just 'annoying to make work in Discord' I think. I checked and Midjourney doesn't even have gifting... but if it can be done absolutely super cool.

[delicate cedar]

I would assume you could just set up a special react that when given to a post with a video attached would debit a token on your backend?

I'm not a programmer so I don't know how outrageous my thinking is here or if even Discord itself has a bunch of restrictions on stuff like this?

[feral aspen]

could be. i was surprised it's not a thing in midjourney, so maybe some discord policy. quick search didn't turn up any stated reason

[delicate cedar]

I wouldn't have thought Midjourney as a usecase, myself. Pika Labs on the other hand...

(Though Pika Labs is free^1 so it really doesn't matter.)

^1 For now at least..

[fringe knoll]

I don't think this would be an actual reflection of overall Suno quality.

I expect the life cycle of a Chirp user is something like...

1. discover Suno and burn through all credits quickly, results not comparable to longer users because experimenting, learning, easily amazed by initial results

2. buy subscription and usage changes slightly, becoming MORE experimental, just to see, with many intentional rejects... and more exploration into other genres, also just to see.... Also starting to craft longer songs, and tolerant of mistakes.

3. veteran user trying to hone in on a specific sound, and more critical of results when it veers off. More critical of starting clips, and more critical of self-mistakes. Also slowing down usage as newness wears off...

My suspicion is that ❤️ is more frequent early on when it's all fresh and amazing... and rejects more frequent later as user becomes more sophisticated and knows what they want.

Therfore everything is skewed by the adoption rate of new users, and retention of veteran users

[delicate cedar]

Oh I wasn't thinking of it as a reflection of Suno, more of curiosity of what the life cycle of generations is.

I have followed that same pattern you describe, so it might entirely be correct. The request for statistics is to see if the suspicion is correct and if I've simply become more critical and more experimental or what the process is.

[delicate cedar]

Besides I don't really know much about the whole Midjourney Discord, since I cannot post in their discussion channels for some reason. I've tried getting their attention to ask about it but since the only places I can comment are in the new user generation areas...

May end up just cancelling my subscription there as there are all kinds of free image generators I can use. I don't need MIdjourney as nice as their stuff is.

[feral aspen]

I'll make sure gifting feature suggestion not forgotten. No idea how practical or if there are complications, but personal opinion, when your users like your stuff so much they are asking for more ways and reasons to send you money, that's a high priority feature.

[delicate cedar]

No no no...not cash. Never money. Tokens. Tokens which can only be used to make more music.

This isn't about generating wealth or a way to extort people. This is a way to kick a few tokens someone's way in hopes of encouraging them to experiment or to make more good stuff. I already have a short list of people I'd love to kick a few tokens at.

Sorry if I sound pedantic here, but I really am not looking for monetization to become a thing in this DIscord. Outside it? I think it's great that we can do so. Here? Here it should all be about cool people making good sounds they never thought they could do.

[feral aspen]

Understood. I don't think there's too much risk of that, but "more ways to use your service" is perhaps better stated.

Though now you have me trying to imagine an episode of the "The Sopranos" that somehow involves Chirp songs in a complicated scheme...

[leaden marlin]

Humming a melody type of thing and then having the AI recreate the song I’m thinking about? Sorry, I don’t know much about music terminology to explain what I want to do.

[delicate cedar]

Sometimes you can put in the text the kind of sound you're looking for and the AI will do it?

I know you're making a suggestion for future function, I just wanted to mention something you can do right now.

An example.

It kind of works...

[misty herald]

would be great to have an optional MP3 audio only button for file downloads

[vernal holly]

a bark update that can do heavy metal singing voices without the music background

[graceful rock]

how to choose the voice of singer? male/female

for example first verse male, second verse - female

[wicked trout]

Would it be possible to generate an intro? Kind of a reverse continuation? I've found that, while I can start things off with an instrumental intro, I usually get better results for the song overall if I start with a verse or chorus. It would be great if we could start there and then continue backwards to get a good instrumental intro.

[delicate cedar]

Suggestion: A feedback group. Where people can post a work, completed or unfinished and in progress and get feedback and suggestions.

[fringe knoll]

I like the positive... but unfortunately see potential negatives... Would need a moderator.

[delicate cedar]

True.

[mental horizon]

Just made a rock channel here:
https://discordapp.com/channels/1069381916492562582/1160714716046438532

Want to do the honors and make the first song in it?

#🚀┃updates will be small, minor changes and server improvements. We'll reserve larger announcements for the #📢┃announcements channel, so we're not spamming folks with notifications.

[mint thorn]

Can we change nothing and set the price of orders to 50 credits instead of 100 credits?

[foggy ether]

/chirp

[feral aspen]

For now it's okay to use general channel for that, it's not too crowded. Or even chirp-technical if it's a more specific/technical question about how to achieve a certain output.

Both @delicate cedar and @fringe knoll have some sick Chirp prompting skills. Really appreciate you both helping other Chirpers in here.

[covert oar]

Done and done. 👌

This is a tip for new and veteran chirpers alike. For a better chance at a clean Outro whatever the last word of the outro is use () 3 times. So if for example, the last word is smile. (smile) (smile) (smile) it will either echo the word out or instrumental out and sometimes even fade out perfectly. 👍

[delicate cedar]

I'll have to try this in my next one.

[delicate cedar]

Eh. Didn't work this time, but given the Frankenstein's monster I'm putting together, I guess that's alright.

[river falcon]

Can you add a pop category in genres ?

Can you add a English chat

[feral aspen]

I guess it wouldn't hurt but I think for English chat the regular general channel and all the regular Chirp channels should cover it, everything is English except that category.

[river falcon]

Yeah but they upload videos in other languages, English chat you wouldn’t have to worry about that

[feral aspen]

Yeah, good point. For browsing Chirp samples still could make sense. Oh English gen channel is there: https://discord.com/channels/1069381916492562582/1154766285960392715

[river falcon]

Where …? Can u screenshot it

[feral aspen]

Got it https://discord.com/channels/1069381916492562582/1154766285960392715

[river falcon]

Thank you so much 🙏

[fringe knoll]

There's a https://discord.com/channels/1069381916492562582/1156615585552150549 thread in the https://discord.com/channels/1069381916492562582/1101325881059840040 channel....

[cloud wadi]

Continues kinda suck now...

And I keep getting gens that are like 2-3 seconds, so it's wasting my credits

[random python]

Its would be great to add more options when you chirp and edit after

Sometimes the music and singing is great but the ability to change the lyrics after would be great

[fringe knoll]

Agree on this one. Being able to regen the same music with different lyrics would be sweet.

[hollow fjord]

Hi! Chirp seems to have a noticeable bias towards generating male voices when not instructed otherwise, with some genres like pop and children's music being an exception. Is this something Suno has looked into?

[crimson sentinel]

There’s probably more training data on male voices for certain genres like rap than female voices, could be a guess. Pop songs and children’s music typically has a balance or more for female voices, which is why there’s exceptions there.

For example, men usually cover the majority of rap songs, but not all of them. Women usually cover the majority of modern pop songs, but not all of them. So when the bot trains on a large library of these songs, there is gonna be more of a bias towards certain categories.
This is a guess though, if anyone actually knows why, they can correct me.

[hollow fjord]

Yes, of course, I'm also assuming that this is due to a bias in the training data. But I would like to know if the team is aware of this bias and taking some measures to counteract it

[compact lark]

when you finish suno, will you make all generation paid?

[rose seal]

Definitely always want a free tier

[compact lark]

like free tokens every day/week ?

[delicate cedar]

And a pony!

...actually that makes me think. Do the subscribers also get free tokens every month? Just curious. And with my usage, I'll take any tokens I can get! 😄

[fringe knoll]

Thank you for pointing out that the AI doesn't need to be stunted just because the music industry is....

As a practical workaround... I have been able to prompt 'thing-A but as thing-B' –– specifically 'funk but all instruments are tribal percussion'... etc

so something similar to: 'hard rock but female lounge singer' in the genre might theoretically work (not suggesting that prompt exactly, but something in that vein)

[fringe knoll]

the last free tokens I saw were when Suno upgraded the AI...

So we should all encourage them to constantly upgrade the AI, so we can get free tokens to try the new features. ... 💸

[delicate cedar]

I meant like how free tier gets 250 tokens (IIRC?) every month. I just wondered if subscribers also got those?

[delicate cedar]

But yes, they should upgrade constantly with only improvements and give us all tokens! 😉

[fringe knoll]

we can pretend the subscription is for commercial license use... and the additional 750 'free' tokens on top of the already free 250 is just a friendly gesture....

[delicate cedar]

That works for me.

[delicate cedar]

Suggestion: find something better to do a stitch than the heart icons. People are misunderstanding the function of it and just liking their single parts and not actually using it for its intended purpose. UX issue?

[rose seal]

Definitely is double-purpose right now!

[feral aspen]

Agree, actually suggested this myself, maybe for now could add in a text word like the way "continue" was added to the emoji, and wait till fully change heart until there's other UI changes at the same time. Just because people finally got used to after weeks...

[delicate cedar]

I was just frustrated when I tried listening to some stuff that was 'liked' and thus presumed to be completed only to discover it was just a short segment from one post.

😦

...OK, second suggestion. Find a way to determine if a song that is liked is over two minutes and set that in a 'radio' channel?

[feral aspen]

Maybe like a 'reacted to' clips channel that doesn't include basic thumbs ups and hearts, only clips somebody took the time to specifically select an emoji. Might be a good high quality filter.

[covert oar]

A problem that all music generators seem to share is holding an annoying note throughout the entire generation. Fixing that would not only create more coherent generations but also add inflection to the singing voice as a byproduct. Not a coder so not even sure if it is feasible.

[delicate cedar]

I propose we use a record vinyl icon if there is one...

📼 lulz, the only tape in the defaults is a VHS...

[tender light]

Can you add a genre for sailing songs? (salty sea captain, pirate, etc.)

[mental horizon]

Sure! What's the best word / phrasing?

Pirate? Sea shanties?

[feral aspen]

Probably Sea Shanties, but do you know any good Chirp samples? That term may not be what is working best, I'm curious myself.

[feral aspen]

Did you make any songs like this, what did you find worked well to use for the style?

[outer spade]

one for jazz please

and one for like

niche genres

just call it niche ig

[rose seal]

Funny timing reading this discussion earlier today but hope the updated plans scratch some of this itch!

[rose seal]

Funny timing reading this discussion earlier today but hope the updated plans scratch some of this itch!

[covert oar]

Thank you ever so much for Premier! 🫡

[limpid dragon]

Can we get an abbility to edit some text after its been created? Some times I miss spell a word or something and i like the track but just want the words edited. I lose a whole perfect track over a typo.

[tender light]

Sea Shanty makes sense, but I haven't made one yet. I wanted to make one for Captain "Alice Babybeard" who has a back story of adventure on the open seas and an image of a baby sea captain with a beard, made in Midjourney.... I'll probably make a 15 sec. spot in Runway when I find the time, lol.

[dire frost]

Great suggestion! Don’t forget that once it’s added to post what you make in #🐣┃suno-showcase 🙂 Sounds like we may need to make a new thread in #1154790975240155247 too 🏴‍☠️

I love sea shanties

[rose seal]

We want this too and we’re thinking about how to do it but will take some research work!

[delicate cedar]

Suggestion: A preview mode? Something where it doesn't stitch as good and presumably doesn't take as much effort from the system, it just plays the segments in order as they already exist so you can detect drift in vocals or music without it needing to render the whole thing.

Not sure if this would do anything good for you or not? It just hit me that if everyone is making previews ahead of their song actually being complete this probably impacts things.

[fringe knoll]

SUGGESTION:
Jonathan Fry's technique of spamming the genre description with a repeated prompt:

"Male Singer, Male Singer, Male Singer, Male Singer, ...."

could maybe follow Stable Diffusion convention of multiple parentheses:

"(((Male Singer))) ..."

[delicate cedar]

Suggestion: a stop button for when you did the stupid and accidentially entered something in the ChatGPT field and want to stop from wasting your tokens and the systems resources.

[sage pasture]

Suggestion:
Paid API so desktop & mobile apps can use this tech

[errant gale]

Suggestion: allow an option to copy the seed from one generation to another

[glass forge]

Bence arka plan şarkısı arttırılmalı daha güzel olur.

[delicate cedar]

Is it possible to have a collaborative Discord Server account? A way for a Discord owner to just designate an area the creative section and set up a fund for tokens to that section anyone with the role can access? Probably there should be some controls on how much a person can use in a day and other things of that nature...

[rose seal]

This is a great idea and we've been thinking about it

[delicate cedar]

Glad to hear it. I don't know how well some of these ideas are able to be implemented, but I can imagine a group of people wanting to do something like this so it just kind of makes sense.

[broken raft]

Would it be possible to have a subscription that would us to download the stems?

[misty solar]

This might already exist but.. negative prompts? At least 10 out of 20 times I’ve used this, it has “darkness” or “shadows” or “despair” in the lyrics - “shadows”, twice in one song, actually. I’d really like a way to ban those words. I know its vocabulary is bigger than that.

[broken raft]

/chirp

[digital thicket]

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.

[thin night]

La possibilité de spécifié au travers d'un parametre l'absence de voix serait intéressante. J'aurais aimé donner des liens youtube de musique pour définir le style.

[distant horizon]

EXPORTS

• midi for instruments and vocals in separate audio track
• notes, tabs
• all tracks as separate audio files, zipped

[tidal pilot]

documentation on website, where can I find a guide on chirp?

[fringe knoll]

Docs don't seem to be navigable from the Suno front page...
https://suno-ai.notion.site/Suno-Docs-38e5ba5856d249a89dcea31655f4fb74

Discord channel is not displaying for all members for some reason...
https://discord.com/channels/1069381916492562582/1151542418991239168

(I had to select SHOW ALL CHANNELS:

[north lantern]

Suggestion: Can you add to the /info command how many hours left are for the next usage?

[delicate cedar]

I think it's a great idea. Should be relatively easy to add (says the non-programmer) and is the type of thing they used to call a "papercut" on the Ubuntu Forums back in the day. Something easy to do that adds polish.

[mental horizon]

Great suggestion, @north lantern. Thanks!

[green raven]

que tal um novo canal para as melhores musicas do #😍┃v1-songs com mais de 10 curtidas, muitos videos otimos é perdidos naquela infinidade

[mint thorn]

Hi ! i hope you're good !
I think I have THE only thing Chirp is missing! 😄

It would be so perfect if we could adjust the length of the clip. For example: 10scd 20scd 30scd and I saw that it was possible to have 40scd so why not add an option with these 4 choices? that would be awesome! 😍 🥰

this would allow us to adjust the possible length of the verses (like 30 or 40scd) or for example interludes without vocals (10scd or 20scd)
this would also avoid clips of 30 seconds for 2 lines and vice versa, clips of 18 seconds for 6 lines ^^

[fringe knoll]

🤔 maybe an addition to this suggestion: count the measures (bars of music) instead of seconds.

[mint thorn]

I don't know enough about it but if it's similar then yes 😄

[scenic oasis]

Please create a "how to" channel

[rose seal]

Welcome, check out #🐣┃how-to-chirp

[ancient igloo]

suggestion: mass transcribe melodies of chirp outputs to midi and use it to train a model where you can input a melody, profit

[digital cape]

pls add Belarusian🇧🇾

[scenic oasis]

That channel dissapears after first use I believe, but coming back(some time) after first use it is not available

Also thank you for your reply

[gleaming onyx]

maybe an addition: once you combine two clips together, you should have the option to carry on with that combined clip
AND, please please have it post what keywords, phrases we used in the song generation before it spits out the AI/song
i had a good set of words (sent to the GPT option) - got stuck on a phoen call - came back and the song was great but... i couldn't remember what I used prior.

[delicate cedar]

Since I really have no idea how this stuff works, I don't know how unrealistic this suggestion is, so I apologize if it's completely infeasible with the methods you use to generate sound.

Is it possible at some point to create a multitrack recorder type of system where the vocals and instruments are automatically split and can be requested as such with a button?

I imagine if the methods by which sound is created allow it, this would be a nice feature and much less "noisy" than some of the methods we use now to try to strip out vocals or to pay attention to specific instruments.

[fringe knoll]

Lots of requests for this: "Stems"

Suggest this could be a (paid) add on (or extra feature for a 'pro' tier)... the technology exists, but it takes some processing power and a longer wait... and doubles (at least) the download file size....

[oak echo]

make bark longer 🙂

longer audios i mean

add it so we can edit the audio togeather on bark, like chirp

[delicate cedar]

I thought stems were like seeds for images\video. I was suggesting split audio generated that way from the start so people could edit sections and still have the whole fit together.

[fringe knoll]

https://blog.native-instruments.com/stems-in-music-production/#what

[delicate cedar]

Thanks for the reference materials.

[umbral silo]

being able to feed in seed from past generations have been brought up few times. Would this be technically possible? I'd love to use something like this to adjust the lyrics a bit and hope it wouldn't change the output dramatically if you kept the same seed

I know things like that work on other AI systems to some extent, and sometimes the tiny change on prompt just steers the output totally off from original. would still be nice to have option for it, even if it only works most of the times

[void panther]

this quote of psalm 23 in portuguese ("a tua vara, e o teu cajado, me consolam" or in English: "your rod and your staff, they comfort me") is blocked by bark "safeguard" locally, please, review your censor methods, because it is a pure text, and i don't know how to allow this and block similar sentences with bad intentions

this quote is about correction and leading respectively

not about bad words

"vara" (rod) commonly is used by majority of people to refer to masculine organ, and this word solo works, but if given a context, bark start to spoke random phrase not understandable to block the prompt

[fringe knoll]

just misspell the phrase phonetically as a workaround.

[void panther]

yes, is what bark do to block some content, but I think it should be removed, at least for non commercial use, also, someone already reverse engineered it, and people made the same with llama and llama2 and many others, so why continue to try blocking AI this way? I think that the bad thing with censorship is that it broke non bad intentions too

and how about stable voices to TTS with Bark? I'm using https://github.com/rsxdalv/tts-generation-webui UI and would be amazing if some explanation were given on how to train a voice? My best bet is clone the semantic of real people and do a bit of prompt on how the tone should be, but with some amount of steps doing this, bark npz start to lose semantic

[leaden marlin]

what would be really cool if u could enter the name of a singer and it would sing it in there style

[void panther]

would be amazing if possible to retrain bark voice like concatenating a new audio to a already done .npz file

[clear thorn]

Finetune ADA OpenAI gpt model on a text sample, I made myself a few finetuned models to mimick few of my favorite authors

[void panther]

can you make a simple guide about how to finetune the voice? I always lose the semantic

[clear thorn]

Semantic = ? So we can agree on the definitions

For me a sentence both has syntactic (spelling apparence and grammar rules) and semantic (meaning)

You can mimick one's way of speaking by downloading transcripts of YouTube video interview for instance

[void panther]

in the ui there're two voices, semantic voice and audio voice. The semantic voice i discover that is how well the phrases are spoken to guide the audio voice, which is the how the tone of voice will be

[void panther]

oh, so i have to input the uh and ah too?

[clear thorn]

Yes allegedly

[void panther]

my huge question is about these options:

Use old generation as history
or Use history prompt setting
or Clear history

what of them should i use to train?

[lean basin]

Suggestion, make the tracks available so we can mix our-self, the ready mixed track alone is not usable for a musician

[void panther]

use demucs

the AI generate already mixed

you should use a tool to split them

or generated with style "acapella" then use your sample

[tulip summit]

Hey what if there was something like ticket tool for you to create your own thread

So it’s not clogged up when other people are generating

[void panther]

thread system

just create a thread for you

🙂

[white tide]

https://github.com/arkrow/PyMusicLooper this looks like a interesting feature to have in chirp

[random python]

It would be nice to see more options/suggestions in the style section

[oak echo]

maybe put famous peoples voices on bark 👀

[white tide]

Alternative monthly/yearly subscription plans, with a daily amount of credits instead of monthly, might be more attracting to some.

Maybe with options to switch to a monthly credits plan with same pricing at any time and the other way round.

Or simply let a user set an optional daily limit for how many of his montly credits he can spend per day, so he can keep track of how much he uses it. Just a warning which can be ignored alternatively by pressing a button to ignore

[fringe knoll]

the app.suno.ai Home Page should feature our completed ❤️ tracks.

[mental horizon]

Great suggestion. Thanks, @fringe knoll and @pearl elk!

[white tide]

Not sure if it's already a thing, but it would be great being able to select older bark/chirp versions, if one does not like a new update or so, or needs the results more consistent

[misty solar]

I haven’t actually tried this yet because I don’t have chirps like that (free plan) but say I wanted to collaborate with the AI on lyrics? Like if I put 4 lines in or pasted them from a previous generation, then ALSO put a topic in the bottom box, will I it write the other 4 lines? If not, could that be a possibility? Though I don’t know how needed this is because I’ve been splitting my parts up into 4 lines anyway just to give the AI some “breathing” room while it’s singing (it doesn’t always get to sing all the lyrics if any notes are held - and I want it to hold the notes). Anyway, just a random idea.

[fringe knoll]

Could we get an Instrumentals Thread in the https://discord.com/channels/1069381916492562582/1154790975240155247 channel?

I feel like a lot of the non-lyric prompts could end up in there as examples.

[novel ferry]

There needs to be something better than #😍┃v1-songs something like a starboard maybe

[rose seal]

We're working on a web showcase! But open to discord-native suggestions too. Starboard is basically the more filtered version of that right? Like if something gets N stars it gets pushed somewhere?

[rose seal]

👋 welcome back, yeah, we will keep this in mind

[novel ferry]

Correct, web showcase would be cool, always nice to see how people create the best songs

[crystal bridge]

Why does suno censor people from creating songs with profanity. Its so annoying. Most popular rap song lyrics would be rejected by suno. Big L.

[feral aspen]

The auto moderation is pretty barebones now but hopefully better eventually. Sometimes it flags lyrics that look like a Victorian novel. I will just note that if it's just specific words and not overall content, well, words have many possible spellings...

[crystal bridge]

Auto-moderation.... Okay this is the problem. How on earth are you going to auto-moderate someone's artistic expression, regardless of how vulgar it is. Sure its not music your grandmother will approve of, but whats the actual incentive here?

If there was a paid membership which i could create songs with profanity id take it

[feral aspen]

Yeah, I think devs wouldn't really disagree, I've seen them mention they know that a lot of real music wouldn't pass the automod. It's not the goal or incentive, it's just a thing not yet done yet, and the team is small and working on the music ai itself, website, etc

For context, it was only 6 weeks ago that Chirp had NO MUSIC STYLE controls. You just had to hope and pray and try to bait the AI to make the music style you wanted.

Back in those primitive days, we couldn't pick our music, but we made it anyway! #💬┃general-chat message

[crystal bridge]

I'm just confused why they added moderation, you could message the chirp bot if you want to keep it private, so why not just allow potenitally offensive songs to only be generated through the chirp bot?

And its kinda sad because in my opinion Suno really shines in the rap genre

[feral aspen]

I don't know what their plans are but I've seen Suno devs chirping lots of great rap. I bet the moderation gets smoothed out eventually and finds a good middle ground, as well just being smarter overall.

[crystal bridge]

What is a good middle ground for moderation?

how many people are we allowed to rap about getting stabbed? 2 per song? 3, 4?

Screw moderation

Its not needed

Also consider the sort of things you could communicate in a song whilst staying within moderation, I'm sure there would be a way to write a song positively affirming all of Hitlers actions using similies and metaphors etc and have it pass Chirps moderation.

And there we go

Too fucking easy

Suno allows anti-semitic content GASPS

yet i cant chirp "Its god’s will not mine, I woke up in the matrix.
I can't rescue lost minds, gang life come with great risk.
Heater by my waistline, cause niggahs play tricks
No regular suit, but i make corporate killings
Not a young thug, locked up caught, for stealing.
Full spectrum hash oil, yeah, it’s all for healing
"

what a joke

I guess ill stick to rapping bout hitler cause thats allowed

[grave elk]

cant say fuck which is lame lol

but also it would be cool if the model was able to take [intro/outro] commands a little easier. most of the time it goes to end and at the last secnd starts up again. I assume so that contine works, so maybe another button to "finish" the song?

however other words make it through the filter so it would be nice if you didnt eat my credits if you dont like tho no-no words

otherwise pretty cool app 🥇

[grave elk]

A way to save your chorus would be cool.

[steel shuttle]

/chirp

[rare obsidian]

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.

[fringe knoll]

so many poo-poo boys getting blocked this morning.

🥱

[feral aspen]

ChatGPT Suggestion: Let ChatGPT see the *lyrics * field. That's it.

So instead of ChatGPT starting from nothing eeach time, each start has a more unique context. Seems to increase diversity a bit. Doesn't fix rhyming or writing quality though. 😔

Mainly it'd be fun. More context, details, verses as conversations, reactions. My instructions here #🐣┃chirp-beta-3 message were Explain why this is terrible in detail. so it's a song about how terrible toast is. But you have to jam "yummy toast" in style field or it can't see it now.

"A verse passionately disagreeing with the previous."
"in the opposite style of the previous." etc. Sounds fun./

I guess ideally ChatGPT consider how many lines of lyrics it should write, so this could be more than a simple tweak. But it already adds 8 lines no matter what so it's just the same as it already is now.

only did quick test, and actually surprised it didn't seem to even try match the rhyming scheme of lyrics when continuing them. 3.5 isn't the brightest bulb. Maybe a little prompt engineering grease could get it over the hump there

[delicate cedar]

Relating to the recent conversations in #💬┃general-chat, is there a way to isolate the behavior that is allowing us to reproduce specific sounds instrumentally? Some way to tell Chirp: 'Yeah this text? Use it as a guide for the instrumentation, but under no circumstances do you try to sing this, ok?' so to speak?

The idea is to make it triggerable on command. Is this possible?

[feral aspen]

Is this referring the bagpipes clip where the bagpipes (or the instrument, whistle) are played according to how the words would have be sung, if they had been?

[delicate cedar]

More to wetcircuit's guitar attempt.

Also to my clip I posted there for a reminder to try to continue it later.

#💬┃general-chat message

#💬┃general-chat message

Those.

I've known you could subtly guide Chirp with audio clues, I'm asking if there is a way to isolate it and make it a command in the future.

Think talk box. Think Peter Frampton.

[feral aspen]

As far I know not anything specific, or directly in there on purpose, to do that. Overall I think not many *intentional * things - like things hard coded into the code, things people deliberately made extra prominent or important in training data. These are some:

1. Commas in style field, separating concepts. Intent was to make the order of words not be too fiddly.
2. [brackets] in lyrics field, words in brackets seen by the AI as a different than lyrics. That doesn't mean it never sings them but just generally.
3. Some tags in brackets are deliberate. [chorus] [verse] and some tags (a list would be good...) had extra effort put in to try and get Chirp to pay more attention to them and work more reliably. Most other tags the AI learned by itself, though that doesn't mean they can't be surprisingly useful or even reliable.
4. Some common music styles had a little extra elbow grease too, not sure what. I think the big words, country, etc. Probably instruments? And reportedly more in next version.

That said there could be a way to prompt it, even maybe somewhat reliably. As far I know [chorus] [verse] in Chirp V1 was actually just one of the random things the AI learned, only V2 was it more formally made into a feature. I'd be kind of surprised if was able to be easily and reliably prompted, but not too surprised if it could doable with lots of effort and luck.

[delicate cedar]

That makes sense.

I was just thinking it's create a whole slew of new possibilities if instrumentation could be directed like that.

Instrumentally epecially.

[feral aspen]

I see a lot of requests for "melody guidance" which means the user uploads some audio and the song roughly follows the pattern. But I think I'd have more fun personally with your suggestion here. And it'd be maybe more accessible to more people, just going wild on the keyboard and the AI doing it's best to interpret it as cool music.

[delicate cedar]

Yeah, I get why they're asking for that, with their own audio, especially if they have their own audio. I think it's detrimental in the long run to creativity though, compared to this and more collaborative with the AI to try to see if music can be generated via text this way.

Plus, the upload of audio opens entirely different kind of doors. How do you prove it's your own audio you uploaded and not a fragment of something that is copywritten by others.

Look, I'm not a stickler on copyright by any means. For one thing I like Bastard Pop\Mashups way too much to pretend that is a stance I take. But for this type of thing I'd definitely err in the side of having a way to show the source of everything created, and to me this is one of the best ways to maintain that

[feral aspen]

Actually, add-on tweak. Less of a tiny tweak but would be quite useful to just instruct ChatGPT to do something not just create new lyrics. I would have used this many times today to ask: "Rewrite lyrics and replace all [bagpipes] tags with [flutes]. And every place you see two [instrumental] tags on one line ([instrumental][instrumental]) remove one tag leaving only one on that line."

[feral aspen]

Album Art Suggestion Maybe last frame of video could skip lyrics, art only, so you can extract it. Full image even cooler but maybe just this is easy tweak.

I was testing Chirp (interleaving music/lyrics try to get Chirp to mix them up so lyrics guide instrumental) and this test prompt happened to be a song about "bagpipes ruining breakfast" and some of the song images are amazing, be cool to have better version.

[fringe knoll]

How about a self-promotion channel

(and possibly a rule about only self-promoting in that channel)....

[delicate cedar]

https://tenor.com/view/wait-what-confused-um-squint-gif-13025511

[frigid timber]

any chance of the pricing for credit top ups get updated to match the new monthly amounts?

[delicate cedar]

Of course you say this now just after I bought a bunch of credits....

[frigid timber]

lmao i almost mentioned it yesterday but still had like 10 left but im trying to join that kaiber challenge and want to create my own horror song to go with my video, more difficult than i thought

[novel ferry]

Allowing to create threads in #🗳┃mini-polls

[misty solar]

So whenever I put “jpop” or “K-pop” as the genre, it usually matches the style but in English. Is there a way to get the song in Japanese/Korean but with an English translation? Or-while I’m here-without one? It did once, but just the once. I think it’s luck of the draw. :/

[fringe knoll]

Song in Korean but with English Translation means...?
it is sung in which language? Both?

[delicate cedar]

Yes.

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

Posting this since S.E.S. was my introduction to k-pop

(Also it shows the interplay relatively quickly in a one minute clip.)

At least this us what I think the OP was asking for.

[fringe knoll]

The Language-Specific https://discord.com/channels/1069381916492562582/1154612901030678578 channel has many examples under the various threads that switch between languages (no Korean examples I could find).

I can't say how well it works, but always helpful to see what others have tried.

[white tide]

I'd love to save the voice & seperately/optionally the melody style of a generated chirp song, is this maybe already planned?

[rose seal]

Yeah, a bit of a research project but we want this too!

[cyan ingot]

Some of my chirps are only coming up to 22 seconds and it's cutting off the last lyric. I mean... it'd be one thing if I was putting a ton of lyrics in there and it was cutting off right at 30 sec... but it's not even coming close. I'd just like to see this improved so that it really truly gives you 30 to 40 seconds without cutting lyrics.

[mental horizon]

Added. Thanks for the suggestion, @fringe knoll!

https://discord.com/channels/1069381916492562582/1166166196467548280

[misty solar]

Sung in Korean with English text. The text wouldn’t rhyme because it’s just a translation of the song, just the actual vocals would.

[void panther]

I think it's wrong to do songs like this, but shouldn't be blocked, because with many restrictions, AI starts to block some not offensive kinds

[delicate cedar]

I suppose this would relate to stems? But there should be a way to share a private generation so that others can use it.

I've got a great intro for a song but I'm struggling to do anything with it. It'd be great if I could share that intro and someone else could get a chance at it in the event it proves too much for me.

[novel ferry]

Can probably implement midjourney style ai mod, that get manually reviewed if you contest the ai decision

[fringe knoll]

Is anyone volunteering to read poo poo and FU lyrics all day? 🙄

The TOS is clear on offensive content. If you want NSFW lyrics, open a DM with the bot and get creative with spelling. No one NEEDS to share that in a SFW discord.

Sure, false positives are annoying, and the censor is inconsistent, but the majority of censored content is DELIBERATELY juvenile and offensive.

That's NOT an ai problem that needs a human to babysit free accounts.

[delicate cedar]

Now, w-w-wait a minute, let's not be h-h-hasty here...

There might be something to this...

[delicate cedar]

Full disclaimer: I am not a mod or any way affiliated with Suno.

But...would you be willing to pay for a special professional plan which included human scanning of every lyric you wrote in order to be able to generate uncensored songs with Chirp? Pricing to be delcared later of course....

[delicate cedar]

Don't misunderstand me, I agree with you in principle. But hey, if enough people are willing to self-fund, why ignore the revenue stream?

[frigid timber]

Not sure which channel to ask this in, but I won a Premier plan subscription code via a chirp contest (go figure). Currently out of credits on my pro plan but the options to upgrade are prorate my current plan which ends on November 6 or wait until November 6th for it to activate. Any chance I can just start a new plan?

[lyric sluice]

Path visualization tool to be able to see all the continuations from a single piece in a tree structure.

[delicate cedar]

Is there no option to just add credits?

Like this I mean?

[frigid timber]

No it’s specific to the premier pro plan, essentially a free month code

[delicate cedar]

So there is no buy credits button at all?

[frigid timber]

The button is there

But there is no 10000 credit option either if it did work

[delicate cedar]

This button doesn't work?

[frigid timber]

One sec let me get in front of a pc and I’ll go down this

[delicate cedar]

Might be because you have no payment attached to the account, if it was entirely a free month?

(Not affiliated, just a guy trying to help.)

[frigid timber]

lol no you’re good

My current plan is paid (pro plan)

[delicate cedar]

I just wanted to make sure there was no confusion on that.

I like being here, this place is awesome, but I don't want to misrepresent myself in any way.

[frigid timber]

I’m out of credits though when I try to upgrade it tells me to prorate the remainder of my days

[delicate cedar]

Huh. I've always been able to buy more credits. I've gone nuts if I couldn't have.

[frigid timber]

Which is far too few to get through 10,000 credits

[delicate cedar]

Yeah.... ::shifty::

Well, maybe...

One hopes at least...

[frigid timber]

I can buy more credits but $30 for adding 4000 credits, vs $30 for 10,000 credit plan

[delicate cedar]

Yeah. Having just done that I get it.

But then again, those are extra credits and persist beyond the monthly window.

The monthly credits are use or lose, AFAIK.

Which I don't anticipate on being an issue...

[frigid timber]

Oh I’ll use them given a whole month lol

[delicate cedar]

🤞 I'm hoping not to but....yeah...

[frigid timber]

I’m always cautious every month but now with this many I can actually test some new things out

[delicate cedar]

I've got to get the cash together so I can switch to annual billing though, that will save me some money.

[frigid timber]

Yeah I hear that 100%

[delicate cedar]

Right now though, that total for the year is just over my monthly car payment so priorities...

But once I whittle down some debt it's definitely the plan.

And worth it.

[frigid timber]

I haven’t even checked the amount but 1000% it’s worth it this is basically how midjourney used to be for me when I first subscribed, just creating any little idea and see where it goes. But them credits man really hold me back from exploring too deep

[delicate cedar]

Well I started with the Pro plan just under a month ago and have gotten extra credits twice now. So I really hope the Premire plan is enough room for me to move.

And that was before they upped them both....

[lyric sluice]

50k option would be nice, probably could even go through 100k easily if I had an api or some other programmatic way of doing it

[frigid timber]

I’m roughly at around 550-600 chirps a month

[lyric sluice]

I started two days ago

[delicate cedar]

H-h-how would that even work? You'd set it up in a branching type of deal where you did several different paths on a single song?

[delicate cedar]

You only have the two ears, you'd have to listen to stuff before editing it.

[lyric sluice]

No just go through a ton of files, linearly probably, but branching would be useful probably because theres two returns you could just continue each

[frigid timber]

Daaaamn!!!

[lyric sluice]

So far I have no trouble keeping up and because you would only listen to snippets, you can discern which paths become pruned

[frigid timber]

You got here after it was free too LOL

[lyric sluice]

So probably a dfs

[lyric sluice]

:((((((

[frigid timber]

Wasn’t nearly the same quality as it is now when it was free

Still impressive

[delicate cedar]

I've managed twenty two songs though in that time. Some better than others. A few that I only privately acknowledge. And a bunch of bad rolls...it happens...

Though I've lately been looking back at some of those to see if editing can save them..

[frigid timber]

Oh yeah me too I give up after 4 or 5 rerolls of the same verse unless I know I actually have something then I tweak the styling a bit

[rose seal]

We had a little prototype UI that was like a tree viz of continuations, was interesting

[lyric sluice]

O then I'm not too worried, right now I get about 1/100 generations that would be suitable for what I consider releasable, quality gets worse on each continuation from there, most of my songs are around 2.5 to 3.5 minutes. So at say 20 seconds per snippet we get lengths of like 10 snippets. Which honestly I havent reached in length, my longest continuation is only like 3 snippets, and each tier got worse.

[rose seal]

Let me take a look here

[lyric sluice]

This would be useful for parameter tuning. Because you could go through branches based on parameters

[frigid timber]

Appreciate it!

[rose seal]

Yeah, next release which we're working on.. should help a fair amount with consistency / quality / length

[lyric sluice]

Consistency is definitely number one in my book, but lack of hallucination would be good, most of the time, I can tell the recording is a hallucination fairly quickly so maybe there's something there, where you could automatically detect in streaming, if that's how it works, that it's a hallucination, and undo certain span of time dependent on depth of hallucination

Detecting hallucinations would also help synnergistically with consistency

[delicate cedar]

Might even be a good feature if it were somehow possible to specify a timestamp where an hallucination appears and specify it for removal, keeping the rest of the segment intact. A sort of audio version of the way images can have parts removed

[lyric sluice]

Plus a hallucination detector sort of goes hand in hand with a consistency detector.

[lyric sluice]

Inpainting would be great

[delicate cedar]

That's what I meant. Thanks!

Not sure how that would work but yeah.

[rose seal]

Agree, this is probably release after next - tricky research problem (although we have some ideas)

Definitely want this though - like start with a chorus /hook and then get verse, and then put an intro

[lyric sluice]

I don't know much to anything about audio gen, but I know nlp fairly well, are there any good resources for learning the technical details of how something like chirp v1 or the next models work/will work

[lyric sluice]

Yess, that would be awesome, then that leads directly with consistency into automatic mastering basically. Detect part of song that needs inpainting, inpaint to fix, compare versions, continue

[frigid timber]

If easier for you guys I can shoot an email or DM the code I received if it’s easier to manually change the sub

[rose seal]

I DM'd

[tired yarrow]

Would be cool having different languages for the songs, like italian, spanish etc.

[rose seal]

Welcome! You can do that. Are you doing GPT or putting in your own lyrics? If you put in lyrics in other languages it should just work.

[tired yarrow]

Oh alr, thank you so much for the tip!

[lyric sluice]

A usage graph like open ais daily, and cumulative

[lyric sluice]

A number somewhere in the mp4 that indicates which try of that prompt it was

[lyric sluice]

Time taken is sort of like a hash, but doesn't count up.

[misty solar]

Where are you guys going to get to see this? Whenever I click on manage account, this is all I get…

[rose seal]

Hey, you might need to verify discord email (which our system uses for billing) https://support.discord.com/hc/en-us/articles/213219267-Resending-Verification-Email

[misty solar]

I just resolved that problem but I’m still getting the same page 😦

[vocal oak]

Hello
I am super new here.
I am interested in ai voice generating. I believe there is a GitHub repo associated with this technology ( Suno ) right ? If yes , has anyone been able to make it work locally?

[tranquil raft]

Would be cool to have a seed to be able to reproduce the style of a piece of music without having to continue that piece. E.g. when I like the music, but there is a problem with the lyrics.

[void panther]

upgrade bark to safetensor please

[mental horizon]

Yes, Bark!

https://github.com/suno-ai/bark

You can post in #🐶┃bark-technical with any Bark-related technical questions

[vocal oak]

Any py can be transformed to safetensor easily? 😁

I meant .pt files

Although I think you responded on the wrong channel AT

[void panther]

.py it's not a model, it's a source code, summarizing: plain text, like .txt

[void panther]

no, mate, i suggested them to convert .pt and .pth to .safetensors

[vocal oak]

Yes I asked about that in the other channel haha but I love that anyway I appreciate the response

I did not have time to try it out

Was working in a nother project

As I said will make sure to thank you if I make it work

[void panther]

i remember, hope it works

also, there's a bark channel here

but it's broken now

you can test there when they fix it

[random python]

Suggestion:
The ability separate the instrumental form the vocal

[lyric sluice]

hotshotXL for gif generation instead of just a still image

[graceful hatch]

Midi export? With separate stems

[leaden marlin]

omg i just found the best type of music if u enter pop rap u can here the voice so clear and it sounds sooo good

[delicate cedar]

A way to take a finished song and have Chirp restructure it with better bass. Not sure if this is something that can be done or not.

[lyric sluice]

A way to search and easily manage past productions within the app.

[tough lintel]

recording voice structure (to reuse the same tone of voice)
longer generation time 1-2min

[inland pivot]

I think people shouldn’t have to pay so much money for this, maybe just 10 dollars per month

[inland pivot]

I would love to have that

[fringe knoll]

😐

[void panther]

I think if you underestimate the awesome work they did, you should do it yourself, and they are so pleasure they release Bark for free, so you have a high quality point to start build your own Chirp.

[inland pivot]

Your right. My apologies

[fringe knoll]

Maybe you missed it...

[jade stone]

It amazes me that people will willfully pay 20 a month for something like Netflix that makes money hand over fist with zero innovation, jacking up prices quarter over quarter for no reason but greed, but here is something new and inventive that you can actually make money off of and people think they should have a lower price point. It's ridiculous.

[fringe knoll]

Not to mention the prices dropped/everyone got more barely 2weeks ago, and the $30 plan was introduced that more than DOUBLED the number of songs over buying a Pro plan with top-up credits for far less money.

One thing to say 'I can't afford Premium'..., another thing to say 'it's not worth it'.
Compared to what other music AI on the planet?

#📢┃announcements message

[delicate cedar]

Yeah, I set myself in for the Premiere plan when it was announced. Should kick in some time tomorrow.

Also, I am also curious about this alternative music AI generator, who else is even close to what we enjoy here?

[void panther]

take easy guys, he already said "my apologies"

[delicate cedar]

At this point it's less about that specific individual and more about the multiple posts of that type.

Also, I admit it, I'm hoping someone posts a link to something that is actually competitive with what Suno Chirp has here. I'm thrilled with what we have here but I'm pragmatic. Competition usually spurs things forward. So I'd personally love to see something come close. So far as far as I know no one comes close though.

We generate songs with music and original lyrics and can specify voice types, style of music, etc...I don't know of anything that is even in the same book never mind on the same page as Suno Chirp. I feel blessed every time I use it that I somehow came across it. If you knew me personnally you'd understand just how unlikely the whole thing is and how wonderful it is to me to be able to do what we can do today.

But yeah, no dogpiling on ThePawOfDoom is intended.

[void panther]

ok so, sorry. I agree about competitive across this theme of make songs. Someone sure will make something close to chirp for free locally, can't wait hahah

[delicate cedar]

I don't know about "free" but I'd love to see something come close. This is actually fairly good as is, you know? And we have been told in various discussions that the next generation model of Chirp is being planned and trained as we speak so once that comes out...well anyone competing on the current level will need to step up again.

But that's the whole idea. 😁

[void panther]

Chirp is awesome, but not everyone have money to invest in it. Dollar is 5x more expensive in my country currency. In dollar it's cheap, but not for us in Brazil, and I have a GPU to run some models, so, it were my pleasure to pay for chirp if possible, but it can be better if someone make a free version of it. And about the new version, can't wait, because at now it is awesome tool, a new whole level will surprise us too much

[fringe knoll]

Bring back the option of a 30sec chirp. 40sec is too long for an introduction (for typical pop/dance tempo ~120bpm)

30sec tends to cut just shy of starting a new 16bars, allowing the verse to begin neatly in the 2nd chirp.

40sec is long enough to start the next 16bars – almost always an exact copy of the 1st 16bars, forcing the 2nd chirp to waste more than half its time duplicating the same 16bars before it can get to lyrics (which are always cut off).... House music is ok with 32bars of repetition, but pop songs need to change every 16bars to sound like... pop songs.

The difference is almost always self-corrected in the 3rd or 4th chirp being much shorter, which gets the bot back to it's usual timing (cutting just shy of a 16bar phrase) but also tends to End the song.

TL;DR the 40sec chirp works against continued (pop) songs

(alternately, break chirps into 16bar segments, rather than an arbitrary amount of clock time.

[soft oasis]

How about a 60sec chirp? Because somehow it still cuts off long lyrics

[delicate cedar]

I don't know what exactly the expected window will be but longer segments is something being worked on.

[fringe knoll]

🤔 if they do mark-up to allow us to continue clips from any time, this will be moot.

[formal bear]

yeah I think continue from any time should be doable even now -- will def need to be an optional argument though

[covert oar]

The only part of this whole program that hurts my soul is how it starts breaking down after 1:40 or so. The average song is 3:48. I can get a 3 min song very rarely and it always has a static hiss or is whispering the lyrics or some such nonsense. I do also understand that all A.I. programs have this same issue. Can visualize it as a geyser of water spewing up from the ground. The higher it goes the more energy required to keep it there much less get it to go further. Might be that it requires the new neuromorphic tech to achieve this. Then energy use and heating become a moot point.

[delicate cedar]

The thing to bear in mind is that there is a ten second context window which begins with the first continuation. If there is noise or anything in that context window it is going to carry over from that point. What you can do is return to where the noise began and try to tweak things to break it before it influences the rest of the work.

Which means giving up on some awesome vocals or instrumentals, I know. It really sucks. For now that's the only way I know how to do it.

The developers have said they're working on the issue and I believe them. We can use workarounds for now but I hope to eventually see this fixed one day.

[covert oar]

https://tenor.com/view/you-shall-not-pass-lotr-do-not-enter-not-allowed-scream-gif-16729885

Yes I have also found that revisiting just before the sound helps but it will just create another soon after. There is a barrier that seems to say

[delicate cedar]

Hence why I said tweak not just regenerating.

Sometimes just rolling again can fix things, other times you need to take an epic side quest to fix it.

[terse sun]

Are you guys considering a chirp API so devs can expand this beyond discord into apps? That would be awesome!

[delicate cedar]

I thought @chilly trout had a great idea here:
#💬┃general-chat message

If it would help? I don't know how Chirp does what it does. Does it take as much work each time or would it be posible to save resources by putting a pin in the music and only generating the voices? It seems like a great idea, there are times the music is just fine but that voice just ain't going to work.

[chilly trout]

And vice versa, to pin the voice but change the beat/music
thank you for the suggestion @delicate cedar

[delicate cedar]

Nah, this is all yours, I think it could be a great idea but we need someone who programs the system to tell us if it's something that would be a good useful idea with how things work in the backend. It's likely to be one of those 'why did none of us think of this' type ideas if it can work.

[mortal oasis]

Able to add your own voice recording and the ai will turn it into a song depending on the genre

[short ermine]

how we do this?

[feral aspen]

we can't, it's a suggestion for new feature

[short ermine]

i see.. thanks

[shrewd relic]

/chirp

[rare obsidian]

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.

[minor anvil]

cancelar

[thick meadow]

can you have rooms which are ONLY for full tunes, and can you also categorise them rooms please. the idea is, so people who are making full tunes can connect with eachother

then all the people messing around with a single prompt or whatever, can socialise in their own area too

Categorise, Rock, HipHop, Reggae, Dancehall, Trap, House, etc etc etc but ONLY for fully professional finished tunes in their entirety

this would be superb and allow people who love music to explore other musicians full music & ideas

can you also consider a spotify style app for full professional tunes made on SUNO, charge the public for the service then use the same business model or a slightly more generous one so artists can make money from this

then also, can you consider if you havent already, building up the SUNO social media then sharing the best tunes made on the platform. not the prompts, but professionally finished music in its entirety.

cheers v much

come on SUNO pls don't rest on your laurels - help us artists to progess if you are able too in any way

[tardy citrus]

/chirp

[rare obsidian]

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.

[visual wedge]

Hi, we should know when our credit is reset in /info or define a hour at reset all free user credit (like 12h00 UTC). in hope of your consideration on this suggestion.

[tacit idol]

When you leave the "music type" / genre at random, the videos we generate list the type as just that. "Random"
But the AI has obviously chosen a specific genre from its available data.
Could you have it display the random genre that the AI randomly chose?

Random: (Classic rock)
Random: (Heavy metal)
🤘

[soft oasis]

Can we make chirp make continuations of audio snippets we provide (for example make a continuation of "song-snippet.mp3"). If you add that, the lostwave community will flock all over it, probably making continuations of EKT (ekt was found)

[compact lark]

320 kbps

[tacit idol]

Other genres I've found that work: children's music, grunge, polka, psychedelic rock,
(will add more as I find them)

[compact lark]

you type artist and the track name, and Suno generates exactly the same style.

[oblique sedge]

Is there any way to end the song ? After a few chirps it would be good to have the option to conclude the piece 🙏

[fringe knoll]

yes, try [Outro] [Fade Out] or [End]
find more prompt tips and ways to manipulate longer chirps here: https://discord.com/channels/1069381916492562582/1102676217687195679

[delicate cedar]

This seems like a easy thing to do that would make a huge improvement in quality of life. A "papercut" as they used to call them in the Ubuntu Forums back in the day, something with a easy fix that had a bigger impact.

I was looking at Midjourney the other day and I noticed something they had that we don't...

When you select an option the color of the button changes.

As you can see I selected to stitch the parts together so I could check to see if there is consistecy between the parts so far...

But which one did I pick?

The only way to know is to listen to the parts all over again.

Is there a way to have the color of the buttons change after being used to indicate which option was used last?

[rose seal]

Interesting! Maybe only if it's you liking the clip (as supposed to others?)

[delicate cedar]

Yeah that would be best. Forgot for a moment that not everyone generates in private.

[narrow steeple]

A suggestion would be, if there's something like a seed value, that could be looked up after a generation and used to recreate a similar feel from one song to another and across generations, that would be more specific than just genre.

[covert oar]

Sorry, the bot is currently under maintenance. Please try again later. I hope this means an update!

[delicate cedar]

Sadly this is related to this: https://status.openai.com/

Soon though. Hopefully. We're all eagerly anticipating the new version whenever it comes out.

[dire frost]

@ashen drum please remember that solicitation is against our community #📜┃rules. Please do not advertise in the server. Thank you

[ashen drum]

you lost a potential customer,
have a good one.

[fringe knoll]

'Negging' after a shallow come-on...? 🙄 Yeah, maybe leave that technique in the pick-up artist manual.

[compact lark]

**where updates ❓ **

[mental horizon]

Soon 🙂

[ripe mango]

Wait we're getting an update later today or later in the future

[delicate cedar]

https://tenor.com/view/shaq-shimmy-excited-smile-gif-2048765052329044460

[lethal matrix]

Is there a way to save a voice, tone, speed, style once you find one you like? I made one song I love but have burned 50% of my chirp credits looking for same style sound on the next song. This would be a great addition.

The ability to lock in male of female voice would be nice too. I've tried adding male voice to the song style box but doesn't work.

[drifting mica]

can i upload audio and then add on to it?

[compact lark]

Allow to change the lyrics of an already generated song (like 🔁 but "Edit")

[lethal matrix]

Will try.

[long apex]

Where to see upcoming update ?

[zenith gorge]

Chirp - Have the ability to add pure instrumental sections, and have an extra second at the end of a clip to allow better transition to the next clip of lyrics. 🤘

[fringe knoll]

Chirp v2 is working at app.Suno.AI (with a different web UI and no Continue, atm)
devs said v2 should be working with the Discord UI soon (days, not weeks)
when it's up there should be info in https://discord.com/channels/1069381916492562582/1083140371980234852

[astral hamlet]

We want Arabic language for Bark. It is greate but the only thing missing here is tha supported language for Arabic. Many thanks for your greate effort.

[zenith gorge]

Chirp Website - Add the ability to retrieve lyrics back from the clips page maybe a button under share that says Copy Lyrics or Download Lyrics

[umbral silo]

I just want to second this suggestion, this could fix many issues where things got cut too early

there is difference between continue with half verse vs giving full verse though, sometimes it works, sometimes it just starts doing it's own thing at the new clip 😁

I do wonder how this could be implemented nicely on discord prompt though.. maybe if you use continue option, you'd get extra field for "continue from" where it would autofill the last clips full duration there? or alternatively some way where you define how far you rewind from the previous clips end and it would fill 0 as default

input unit could just be in ms or s if you can still parse floats efficiently there

[zenith gorge]

Chirp - On MP3 files before downloading can we have the option between what the bitrate and khz is set to? Also, the option to choose between Mono and Stereo audio would be pretty nice to have!

[zenith gorge]

Discord - in https://canary.discord.com/channels/1069381916492562582/1154790975240155247 maybe there could be added Synth and Dubstep posts

[formal bear]

you will have to fill in the start time from the clip (probably has to be greater than 5s or 10s), but that should be it

This will likely to be a pro feature though -- features like this can be very confusing for 99 percent of the users, if it is sth on the UI it will be very different

[umbral silo]

yeah that's totally understandable, also nice extra perk for pro tier

[formal bear]

we have a lot on our plates but I will cc @rose seal to consider adding this

[umbral silo]

if it's meant for more advanced users, it could be just some tag in too, like [start:24.454] at the start of the genre or lyrics

[formal bear]

nice try but that will be conflicting other features we are cooking 😉

[umbral silo]

well isn't that a tease

but yeah, I'd love to have weights for the genre too

[formal bear]

I don't think you have to wait for long 😄

[umbral silo]

I get the bandwidth issue but could we have option for higher bitrate mp3's on some of the paid tiers? I just found out that the website actually gives notably higher fidelity audio if you download the mp3 audio-only version, but the bitrate is still the same as on the AAC track along the video

mp3 version is still notably higher fidelity due to different compression algo but mp3 is also currently only 128kbs which suggests the original chirp gen is a lot higher fidelity itself, but gets compressed down

getting 320kbs mp3 would be very nice

[zenith gorge]

Yea and also an option to change the kHz to 44.1 for example

[umbral silo]

I actually think it's 48kHz now

it's just getting compressed away

[zenith gorge]

Well some sites require 44.1 for uploading audio for whatever reason ;)

[fringe knoll]

Just give us the best quality possible. Any workflow to other software is my business.

[zenith gorge]

What i meant is make it configurable for what mp3 output kHz and Bitrate should be in a prompt before downloading the mp3. (Sorry for the confusion)

[umbral silo]

also, please let us download the whole song audio on the website version, instead of the clips only

if the only way to get the full song through the service is the worse audio quality video version, it gives worse impression of the whole gen than it should

everyone should know chirp can actually do much better

[fringe knoll]

Devs have promised that web and Discord will match features (probably this week) – it's why they are doing a 'soft' launch of v2.

I expect once the full quality is on discord, and Discord features are available at web, we will get a big announcement about v2...

Then they said new features will start happening.... 😋

It's only been a few days – which is forever in AI years, LOL, but it takes time to physically code the webUI and stuff....

[delicate cedar]

This is not the first time I've seen this. Someone uses 'Random' for their music style--either deliberately or accidentally and falls in love with whatever it is they've created. But they don't recognize the style at all.

Suggestion, any songs made in Chirp with the style Random should have it listed as Random: XXXXX with the actual style used listed. That way people can rediscover the style they have used and use it again.

Recent example: #💬┃general-chat message

[umbral silo]

I've had this too.. but is the style actually input to chirp in this case? the whole suggestion is based on assumption that they randomize the genre for you and feed that to chirp model. but what if it's simply not put to the input at all?

on stable diffusion we can use clip interrogator to reverse engineer what might be in the prompt, it's not 1:1 same thing but if we could do same with music, it would help on deciphering what made up that song eventually

[delicate cedar]

That sounds good to me too. I'm just acting on the assumption that when randomly selected the AI still knows at some point what that random selection is. If it does then it should list that on the created Chirp. If there is something else going on that makes this troublesome, then by all means some other method should be used.

But it just makes sense to me (a non-programmer) that at some point it is known what style was selected and that should be used to indicate the style to the user. I wrote it as Random: XXXX because it should be indicated that the style was selected randomly when giving the type of style which was selected.

I myself try to make careful records of what exactly I inputted when making a song, but sometimes I get caught up in the act of creating that I forget to write stuff down. It's good to be able to get this information after the fact in the Chirp itself.

[rose seal]

we will probably have our own CLIP style model at some point (there are some decent msuic ones out in the wild but they are still pretty high level) . Re random unfortunately it's true random, so we dont have the exact genre text in generation time. We could do a random genre mode though

[feral aspen]

I think the web version basically does do this, since you see the keywords right? If I just write "A song about pizza" i don't get "random" I get 2 or 3 music style words

So maybe that essentially solved already? Can default to that mode, maybe, instead of genunely random.

[umbral silo]

this could also make the genre more random as well

I assume current random on discord bot still has heavy tendency to generate rap music

[feral aspen]

Just as long as we can still use a "blank" genre, like Chirp Alpha with no style field, I find that pretty useful myself still

[delicate cedar]

How does that even work?

Never mind I'm sure I don't have the background to understand it. I assumed that at some point it has to know what it's doing but I guess not. Computers are amazing.

#🐣┃how-to-chirp

[zenith gorge]

Chirp-The ability to input 10-30 secs of anything as a way to substitute the genre for a genre you dont know

[twilit nest]

Hey Suno Team,

I've been loving the AI composition tool. However, I've noticed that when connecting different segments, the transitions sometimes lack smoothness, affecting the overall flow of the music.

I was wondering if it would be possible to introduce a one-click optimization feature. This would be super helpful in ensuring that transitions between segments are more coherent and musical. It could significantly enhance the overall quality of compositions created using Suno's AI.

I appreciate the innovation Suno brings to AI music composition, and I believe this addition would be a great improvement.

[wicked hawk]

Hey Suno Team

Love it, love it, love it....
Might be obvious for you but a Music Based PC/Console/Phone game with this powerful tool might be legendary. Long ago there was a game called 'Kudos Rock Legend, . With Midjourney, ChatGPT, and Suno it would be awesome to recreate some game like that. I was a musician and graphic artist, now i work as junior programmer so i see now how this concept would work, if you have idea to get into it keep in mind I would be glad to join complete that project

Regards, you're awesome

[solemn verge]

Hey Suno Team,

Honestly its incredible that music generation is even a thing, and all the music generated is incredible considering this is cutting edge. I was wondering if you guys could refine music generated in Cantonese. Right now it can only do Mandarin well and I would like to try generating songs in Cantonese. The songs come out pretty messed up.

Other than that, honestly incredible work!

[icy root]

Consider adding an appeal option when a prompt gets flagged, like Midjourney does. I think if a human reviewed the prompt I just had censored that they'd agree it was unreasonable

Also, you should provide some context of what triggered it. It's pretty frustrating to have your lyrics rejected and attempting to guess what the problem was only to have it repeatedly rejected

[rose seal]

This is a good idea, we'll look into it. Re context, it's not a manual blocklist so it's a little harder but we can think about it

[icy root]

For a little more context, I eventually managed to edit a version that it accepted, and in retrospect I think it improved the song. I thought it didn't like a mention of nudity, but it turned out that it wanted me to remove a reference to "breaking windows" and also "burning all my clothes". It really toned things down, but it worked. Maybe those lines were a bit too much. I'm still not sure why it needed to be censored like that. If it had the appeal feature, I could have taken it as feedback and still generated the version I wanted and then made the censored version for comparison

[delicate cedar]

There's a bit of automoderating going on for song lyrics. (Consider how many people are here doing songs--no way we could go through every single generation and check them all!) But the problem is that ChatGPT is managed by scared humans so they went through a lot of trouble to make sure they coded into it instructions to be nice.

As a result it really doesn't like when people don't play nice and get along.

This manifests in stories where every bad thing turns into something good by the end of a story or gets put back the way it was. How everyone learns a valuable lesson. And in our case, it flags songs with stuff like breaking windows or in my own specific case a lyric calling someone stupid.

I'm glad it resulted in you being more creative and making a better song. I'm sorry that it was necessary.

[hollow otter]

Good afternoon, please tell me who is in the know, does the subscription allow you to make tracks with a longer length, for example 120 seconds or 180 seconds, maybe more? please tell me

[delicate cedar]

You can do that now. See #⏩┃continuation

What a paying account offers is more credits to make more parts and the right to monetize according to the rules laid out in the Terms of Service.

[zenith gorge]

Chirp - Metal screaming [I.E. Corpsegrinder, Cannibal Corpse, Lorna Shore, and many others] Metal singing styles

[blissful current]

I have a couple of questions regarding the new web app: (1) have I missed it, or does it not allow me to enter my own lyrics the way the Discord bot does? (And if not, why not?); and (2) will there be an API soon/eventually, so I can have my code generate songs programmatically? Love Chirp for sure, but the app seems less useful than the Discord bot without either of those features.

[humble hare]

Hi, can I enter my own lyrics in the Web App which is much nicer to use as a paying customer? thsx

[feral aspen]

Not at the moment but soon. Web app is a beta/preview of the new model btw, so even the music is different.

[delicate cedar]

So I was thinking...

I know that most of the time when an AI system is capable of doing something successfully the developers try to lock it so it won't stop doing the thing. So I don't know if this is something easy to add or if it would take a bunch of work or if it is even feasible at all. So disclaimer out of the way...

I was thinking...sometimes Chirp does the music just right but the vocals hit or miss on me. Or vice the versa, the vocals are spot on but the music lets me down. Wouldn't it be nice if there was a way to let Chirp know which way to aim when trying again?

So my thought was this: optional split feedback with vocals and music. A scale of 0-5, with 0 being neutral and 1-5 being a graduated scale of how close Chirp is to what you're looking for, which could be refined as a sort of meta-prompt in the followup generation. Basically do that again with emphasis on this right here.

I know something like this would if possible result in just as many bad chirps as good ones, but theoretically the chirps would be focused on the area which needs improvements with an eye towards hitting that mark eventually. A bit better than just tossing the dice and hoping for them to land in our favor, yes?

If this is even possible or feasible to do?

[rose seal]

Definitely in line with what we want to do, but need some R&D to figure out how to do

[delicate cedar]

Given the hopes of the webapp going forward, it may even work beter with a 'hot and cold' slider but you guys know that aspect of it. This was just a thought that I pondered.

[rocky geyser]

Stereo output and being able to pan

[umbral silo]

website version (chirp v2) is already stereo

[dire harness]

can we please get a way to just make the music. This is so good when it glitches out sometimes.

[leaden marlin]

On the suno website it would be cool to get music only. No lyrics
Or at least the option. The music is so high quality that I would like to use it for my own projects if possible

[blissful current]

I've found that including "instrumental, no lyrics, no words" in the prompt is usually enough to get the new Chirp model to ignore whatever GPT spits out. Got some nice chiptune tracks that way.

[hasty pagoda]

Any plans on adding a feature that lets you make a “karaoke “ version of the song you made, essentially separating the music from the audio somehow ?

[long apex]

Hey there! I have a suggestion for the future of this app. It would be amazing if it could incorporate advanced music settings like time signature, tempo, key signature, chords, and modes (major/minor scale, arpeggios, etc). This way, users with a background or understanding in music theory would have the freedom to craft some incredible music using the app. These added features could really take this app to the next level and make it a go-to tool for music enthusiasts. What do you think?

[tepid wolf]

How can I get the style of music we can use in the suno?

[fringe knoll]

start with these... https://discord.com/channels/1069381916492562582/1154790975240155247

but also look through other people's Chirps to see what works. (it is LLM so it understands many words and word-combinations, also many specific sub-genre... not so much band names or specific musicians)

[oblique granite]

Idk if this has been suggested but I'd love a way to take a previous clip and add it as the next part in a continuation. Ie you have a chorus you like and instead of regenerating it you just slot the one you already had ad the next part and continue generating from there

[rose seal]

do you want it to splice together somehow?

[oblique granite]

Yea, so we have the ❤️ button and I love it but I also would like a way to use a clip again in the same song because when u use the ❤️ button it only adds it to the generations before that clip

[rose seal]

I see. we can probably do that but it would be hard to make it a smooth transition

[oblique granite]

I think at least starting with being able to add it is a good start and then maybe farther down the road there might be better options for the transition

[rose seal]

OK cool, we are starting to think about what a dedicated song builder mode might look like so that might fit in nicely

[oblique granite]

Or maybe there is a way to like store it in ai memory so that you could like use the Chorus tag and it then tries to generate a transition based on the current stored clip

Idk it's just a thought lol

[rose seal]

Eventually!

[oblique granite]

I love using it though

[brazen ginkgo]

the discord chat scrolled and I accidentally liked the wrong song (heart button). I tried to undo by hitting like again and liked it twice. Maybe a way to undo a misclick?

[icy pendant]

Are there plans to add this into web-based V2? Also is there anyway to use V2 in Discord with the additional features like being able to include your own lyrics?

[loud kite]

If you essentially pre prompt it with "This is for my song since I'm a lyricist" it will automatically relax it's over bearing content policies just most people aren't good at prompting it. If you're too vague of course it's gonna deny you it's like talking to a human since we're also statistical engines.

[rose seal]

Yes, working on it, web/discord should reach parity by early next week

[finite topaz]

Does that mean that the web will also have the ability to extend V2 songs and use multi line lyrics? I much prefer avoiding Discord and using the website if possible. Thanks for the quick response. V2 is incredible and makes some genuinely catchy songs, can't wait for the additional abilities.

[rose seal]

Yep!

[quick olive]

I'm trying to make a song about the german health ministers, ricarda lang, and the songs don't have any voice. That's the ChatGPT gestapo censoring the fun ?

[finite topaz]

I'm not sure if it's feasible but I'd love to see a couple features similar to Midjourney where you can use credits to do slight revisions/variations of the same song and a separate upscale feature where it could process the song at a higher bitrate when you found one that was worth it. Also, a third option, to make it stereo with separate channels like how MusicGen does it would be fantastic.

[delicate cedar]

It could be ChatGPT. Are you making the song on the webapp version or the Discord version?

[icy pendant]

I get quite a few, maybe 10% that are just music. I am only using the Web interface with V2. I hadn't even considered that it might be the GPT API censoring something.

[quick olive]

On the website it would be awesome if you could change/enter the lyrics, make generation a multi-step process so you first cycle through the lyrics/make adjustments, then create the song

Also, not possible to extend the clip on the website ?

otherwise, amazing tool!

[rose seal]

Both coming next update (early next week). We just wanted to put out an early version of web app so people could use it 🙂

[icy pendant]

I see now that it is already actually true stereo with independent left and right channels. I was thinking MattVidPro had checked in his new video and it wasn't. The song I just made, clearly shows it is. Version two is so amazing, huge thanks to everyone on the team working on this! https://cdn1.suno.ai/1ed82070-e2c7-4dfe-b9ff-daaf96e72729_1.mp4

[finite topaz]

I'm very thankful for the ability to favorite on the website and the "My Clips" section. Hopefully they well be able to reformat the cards soon so they show the full lyrics. Right now it shows about half and there is a lot of empty space on the right side of the square cards.

[tame creek]

First time checking out the discord after just trying out the suno.ai app, which is pretty amazing.

I'd like to suggest a feature where you could split a generated song into an instrumental and lyrics, or even into smaller stems (drums, leads, bass, lyrics).

[delicate cedar]

This is something that has been requested before and the devs say they are trying to add it but it may not be in the next major release, but rather a release later on.

[quick olive]

very impressive. some further feedback: ability to generate the melody or the lyrics separetly. E.g. i had an amazing song but unfortunately you can't hear the voice at all (Example: https://cdn1.suno.ai/3e040a82-6acc-404a-b546-83b3d3cafc03_0.mp3) - it would be great if one could regenerate the voice or melodies (remix the songs)... keep it up guys, i'm shocked how quickly AI is moving forward...

[blazing heath]

Is there somewhere where we can learn syntax of how to generate - the basic stuff is documented - but clearly there are times people use [] brackets to cause laughing or control voice type

[zenith gorge]

#1151542418991239168 should have a section for it

[delicate cedar]

I will warn you that while, yes it is possible to call for that stuff occasionally, it is by no means always on demand. I've multiple times now gotten songs which incorporated a live audience clapping and cheering and I've no idea what causes it. I did nothing to make it happen, it just did.

[meager vale]

I would like to be able to input an audio file (a song i recorded on my phone using acoustic guitar and vocals) along with the lyrics and have this program generate a production from that.

It seems to me like it would be less work for the AI

[wise tiger]

Is there any chance of getting API access to produce songs? I'd happily pay etc.

[austere escarp]

three suggestions, for website allows us instrumental only option, and second, let us do longer than 1:20

also inputting an audio file

[gusty trellis]

thanks for watching!

digging the new model, but yea waiting for further implementations... good stuff!

[hot kayak]

Something I would LOVE to see on the new site: a section for browsing chirps people have made. If the user first hearts a song, it appears in the pool.

Random: Songs the initial user/creator/chirper hearted appear here randomly. If you heart a song you find, that helps it reach the top of the hour/day section, which can propel it upward faster.
Top section: Top rated songs go here and can be sorted by top of the hour, day, week, month, all-time. Maybe it randomizes the top a little bit too.

Bonus Request: Some sort of way to filter by song type/genre.

Thing to watch out for: Make sure the first songs ever upvoted don't just stay at the top forever, permanently gaining hearts and never being budged from that top spot. There needs to be a way for newcoming songs to have a shot at obtaining recognition.

[gentle sandal]

I'd love to see an tab to choose between Grok's various modes and GPT for lyric writing.

On random topics, such as those from HHGTTG ("Boy being meets girl being..." or "An infinite number of monkeys..."), GPT tends to be bardic / literal, wheras i get the feeling that Grok would be more humorous.

On more serious topics, such as Zukerberg's resemblance to a reptilian android and fact-checks, the lyric engine goes serious / verbose, with a preponderance on minor keys, which then results in the music output being more like spoken word poetry with musical accompaniment, rather than a song.

[shell quest]

Feature request : display the lyrics in karaoke mode / or tiktok word per word mode.

[quick olive]

Feature Requests: Download Split audio tracks (voice+music), ability to regenerate audio or voice tracks independently, ability to edit lyrics, ability to extend the clip...

[lyric moth]

Some tracks have vocals soundling low! pls fix that

[zenith gorge]

Extend clip is #⏩┃continuation and edit lyrics can be done already technically with the discord version.

[quick olive]

oh i've been only using the web version, the discord version gives me really bad results for some reason

[quick olive]

You need to put a disclaimer on your site that the product is addictive. I have been spending 500 credits now every day and laughing so hard at the memes i generated.

[leaden marlin]

The discord version is a entirely different ai model version. The web version is a updated version that produces way better results

[feral aspen]

Maybe the web version could have a line of text "The web interface is a preview of the next Chirp model" somewhere, it is a fairly high volume of questions lately. Even if it's going to be identical pretty soon.

[delicate cedar]

And by explictly labeling it a preview or a demo we can point to that and make it clear that not everything is fully baked yet so people should enjoy but also should expect limitations and occasional bugs--which they can help us all out by reporting when encountered.

[zenith gorge]

Or just label model versions in the top corner/bot bio :]
Ex: Bot Bio: v1.9.0 while top corner of the website v2.0.3

[leaden marlin]

Ability to only generate one version of song instead of 2

so instead of using 10 credits

you coud only use 5 for one song

And also instead of typing your own lyrics, you could tell it to make a song based on descriptions on what you want. Like for example lets say a knight and a princess

"The knight and the princess were once best friends when a dragon came and destroyed the kingdom. The knight sacrificed himself and the queen wishes it was her instead"

The ai can make lyrics based on what information you give it as an alternative to typying your own lyrics

[feral aspen]

Related: #💡┃suggestions-web message

[leaden marlin]

I mean idk if I'm stupid or something but

it's similar, I think

but i meant typing it by yourself without chat gpt

I never even thought of chat gpt tbh

[warm fiber]

Probably it was already suggested but It would be supercool to merge parts automatically in Discord version or at least make them stitchless. I had to remove pause between 2 files in editor.

[zenith gorge]

"❤️" does this in airquotes not the smoothest transistions tho

[warm fiber]

Sorry, I didn't get it. What can I do to try to merge it automatically?

[zenith gorge]

Press the ❤️ button on the last clip of the song specifically the one you prefer and it will combine them all from the start clip.

[warm fiber]

Thank you, I'll try. I only used "continue" button and thought that " ❤️" are just likes.

[zenith gorge]

Thats what it means on the website confusing i know

[warm fiber]

Ah, you mean the website. I was sure there was no continue option at all. I'll try tomorrow.

[zenith gorge]

There isnt one on the website yet but the heart there just means to favorite the clip lol

[fringe knoll]

🪡 🧵

[atomic spindle]

I'd gladly pay for longer tracks.

[formal bear]

how much longer?

[atomic spindle]

3-5 minute

total

and not using the discord interface and continue

[fringe knoll]

Do you just want 1 prompt and get a 3-5min song? No input along the way?

[atomic spindle]

Yes

[formal bear]

longer tracks are going to be more expensive for each generation though -- simply cause it is more compute. Also if sth goes wrong in the middle (which could happen) the whole gen is kinda ruined.

Given these facts you'd still want a 3 min gen? 😄

[leaden marlin]

ye
just 3-4 mins
cause thats the standard for full music
are longer tracks going to be behind a paywall or

[atomic spindle]

Absolutely. I would love a three minute track. If it screws up, oh well, try again.

[leaden marlin]

i wouldn't really say "oh well, try again"

remember, credits are being used up

although it could be lessened if they allow the option to generate 1 song only instead of 2, so only 5 credits are being used

quick maths will be: 10 credits equals 2 versions of a song, so what does 1 version cost

5 credits

[formal bear]

ok.

It should be possible to gen 3 mins.

We will get there (maybe soon ish). Stay tuned.

[atomic spindle]

Credits are relatively inexpensive. I'm fine with the price.

[leaden marlin]

yeah well

great

[fair hazel]

Suggestion: Improve the artificial intelligence to understand more details of what we ask for in songs. I like composing ideas with more refinement. It would be great if we could adjust parts we don't like. Each song has a richness of details, and it would be fantastic to build this within the tool; it would be the future of musical productions. Try training the artificial intelligence to refine details of any music it has created, of course. For example: the verse part sounds like hard rock from the 2000s, but I want to incorporate punk elements without losing the melody, etc. I would like it to be like this in the near future. Thank you for reading my suggestion.

[autumn pilot]

It's me again... I think a "Vocaloid" like software, with piano roll and text-to-voice generation would be a killer product, some of these voices sound so natural and it would hyper charge music production if we could generate our own lyrics to use in our own produced tracks... Just an idea!

[quick olive]

please tell upcoming features will be available for currently generated projects... have some real nice gems here i'd love to perfect

[last fiber]

It was probably said often but we need a way to generate only beats. The AI made such crazy beats but it sadly has really bad lyrics and flow lol

[random tundra]

A way to generate a specific section for a reference song

i.e.: /generate -s [section] -r [reference song id] -l [optional: lyrics]

so something like

/generate -s bridge -r A1...9Y -l anthemic metal about freedom

It should somehow also give us the ability to regenerate specific sections.
Sometimes the verse is weak but the chorus is great, or vice versa... sometimes they're both great and I want to keep both and wrap them up into a full song

[random tundra]

---

Some songs come out without vocals? (or they're so muffled you can't hear them)

Not sure if this is a bug or a feature... I'd like it isolated, so we can choose to produce instrumental songs

[zenith gorge]

You already can should be easier when the discord chirp setup gets moved to the websites for generation [The Genre and Lyrics sections]

[leaden marlin]

Ability to hum notes or upload our own notes/midi

[frozen saffron]

any chance generate only acapellas??

[leaden marlin]

yeah they workin on it

[frozen saffron]

and download wav not video

amazing

[leaden marlin]

you could download mp3 though

what you using

web or discord

[frozen saffron]

discord

[leaden marlin]

oh

uh idk bout that but they workin on it

I use web

[frozen saffron]

web i can download audio well

[leaden marlin]

yeah web is epic

discord idk

never used it and not planning too

[leaden marlin]

Ability to edit lyrics from the already made songs made before update (Web Version)

I have some banger songs that I wanna change/edit the lyrics. When the update comes out is it possible that we could do that too? Cause the instrumental is REALLY good

[rose seal]

There's not really a way to edit lyrics of existing songs right now (that's probably a few more releases away). We're thinking about ways to reuse instrumentals..

[leaden marlin]

Oh no take your time

it was just an idea

[rose seal]

a good one!

[leaden marlin]

aw thanks man

i just don't wanna lose that one song so I pray it'll be possible down the road. I wish yall the best journey fam 🙏

[frozen saffron]

how to put lyric in the webpage??

[rose seal]

coming later this week

[frozen saffron]

nice

Are all these generated songs exclusive?? this melody?

[supple lake]

Is the web version going to include extensions soon? The model is a lot better but you cannot do a full song

[delicate cedar]

The webapp is basically a limited function demo of v2.

[supple lake]

Yeah makes sense

[delicate cedar]

It is not fully featured yet. It is coming soon but not yet available.

[supple lake]

Cool! But I'm guessing anything we make now won't be able to be referenced later right?

[delicate cedar]

In the Discord bot version you can still do #⏩┃continuation to make a full song.

[delicate cedar]

Honestly? I don't know.

It's a good question.

[supple lake]

Yeah would be epic if so. Otherwise it's tricky. Cause the model is a lot better but on the other hand less control...Hmmm...LOL

[delicate cedar]

The idea is that the v2 full feature release will have those controls available.

[supple lake]

Yeah, awesome. Just figuring out what to do in the meantime lol. This is exciting though.

[delicate cedar]

Very.

[autumn pilot]

As incredible as a lot of the generations are I think there should be a dislike button for some of them lol

assuming that feedback is taken into account for future models in some way (?)

[quick olive]

I'm making an assumption here that the music generation is based on genre + seed, it would be amazing if one could get the "seed" at some point to generate a lyrics with similar music.

[tepid wolf]

May I talk to manager?

[fringe knoll]

I'm a mod if it's something related to the discord you can pm me...

[leaden marlin]

bro thinkin this a wendy's https://cdn.discordapp.com/emojis/1062901570678816808.gif?size=48&name=SkullSpin&quality=lossless

[atomic spindle]

Flamenco tends to have long intros, so by the time the lyrics cut in, the the song play time is over.

[leaden marlin]

It's not only for that

It's for any song really

Sometimes it makes long lyrics and it cuts it out

[fringe knoll]

we will be able to continue songs on the website soon (this coming week)...

[rancid flame]

I was just going to ask, the model on the website is really good

[vast vigil]

The ability to review the song lyrics before generate the full song

[rancid flame]

what about own lyrics on the website? (ok, see it was answered already above)

[feral aspen]

Coming soon. Devs said they are hoping for next week to have Discord Chirp and Website Chirp be the same quality and have same controls.

[rancid flame]

thanks, how is it able to generate 1m 20 sec? is it auto extending?

or will the base generation be 1 m 20 + extensions of 30 sec?

[feral aspen]

I think base is supposed to be single gen, meaning the 1m20s is created without using extensions/continues (so should be very consistent within that 1m20) and stay at 1m20 for launch.

No precise details on continues unless I missed them. It would make sense they will be longer than current 30s, but less than 1m20, some balance between space for new content and having enough history and memory of previous song

[rancid flame]

sounds good

[fringe knoll]

I was told there a little more leeway on the time (full context is longer than 1:20), but they have plans for new (unspecific) features (which may use up some of that memory)...

so it's not just doubling 30sec chirps, the new model has more 'capacity'

not sure how much of that is used in continues

[feral aspen]

Thinking to myself, it might make sense to do 90s first gen and add the extra 30s into the memory, probably something they are testing who knows

[rancid flame]

at least the new model make quite good intros as well before the lyrics

going back to old now sounds bad 😦

[fringe knoll]

riiite? 😓
quick post all your old songs in mono before the are obsolete

[autumn pilot]

Hey guys, could a Vocaloid-like editor ever be in the cards for Suno? I'd kill to be able to do my own vocals for my own songs, it would give so many people who may not be great singers the ability to have custom vocals in our tracks! WOuld love that.

[pastel skiff]

hello. i subscribed just several hours agao but it still says i need to buy more credits. anybody knows what's needed to do here?

[turbid narwhal]

[e-mail]
email them
there is probably a disconnect with your account and they will need to manually merge them

[pastel skiff]

thanks

[leaden marlin]

No mf

The lord is good

He say no lust

[iron acorn]

Add necrotrap style music

like HXLX or NEEDLEGUTS

[polar sundial]

Is it possible to combine the melody of an existing song but put lyrics in another language?

[feral aspen]

Suggestion for web app: Show clip creation date. I find myself trying to navigate to a specific clip based on the date already even though it's not visible on default layout on "My Clips"

[supple lake]

On the webpage it seems it only gives you 10 pages of generations and then they drop off, is this true? or is there a way to get to older ones?

[feral aspen]

That used to be a but it should go past 10 now. Are you see seeing that?

[supple lake]

Yes,

I didn't check today

THANKS

I was worried for a moment. LOL

This software is absolutely amazing first of all. But I will echo some of the others and say that what I would love to have (beyond the controls of the Discord on the web which I heard is already happening) would be:

Something like a seed if I like the way something sounds. I know we can do continuation but I wonder if there's anything else to grab onto if you don't want it THAT similar just near similar.

A bit more clarity on genres. I'm sure it's at the mercy of the input audio but would be great to figure out what it "knows," so like if I want 1960s surf rock if it has any idea of that genre. It seems like when it doesn't know the genre it just starts putting it in the lyrics. I'm not even sure if putting in a year is something it understands, it seems not.

A search for files on the web. I used to be able to search through lyrics on Discord.

[zenith gorge]

Goes to 100 now

[delicate cedar]

I would dearly love to have something like this for the vocals on this one song...the voice was amazing and I'd love to do dozens of songs around it but it is as likely as winning the lottery to get it again consistently.

[frozen saffron]

nice will be good if them made full track not only 120 secs.. and option to download "Acapellas" and "midis" and "stems"... and keep ID of track

[rose seal]

Thanks for this feedback!

[molten orchid]

I know that it may have been asked before, but this is a critical (and not even a complex one in my opinion, unless it consumes tokens I guess) feature and it has to get noticed. Why can't I iterate through different lyrics in the web version if I want a different one. I'm not even asking for manual editing but just iterating the lyrics the AI gives. Since the AI output is heavily dependant on the lyrics and it just wastes 10 credits for nothing getting something I didn't want.

[feral aspen]

It's not a design goal, it's simply that the web version/interface is a quick limited preview to test the new Chirp model. The goal is to allow for full manual editing ASAP, devs working on it now.

[supple lake]

I wonder if perhaps we could do a reverse search like the /describe feature on Midjourney.

If I knew how Chirp hears music I would know how to prompt for it.

[molten orchid]

I made a guess about the reason why you can't manually edit the lyrics. It's the same bull reason OpenAI didn't open source gpt3 and such, because it's "too powerful and dangerous to society" and is risky. But what I'm hearing from you is that v2 was a bit rushed as the reason for this feature absense?

[feral aspen]

What I'm saying is just that that there are many messages in this Discord from the devs that they are working on adding it into the web app maybe even possibly this week

[rose seal]

We working on it!

[cold hound]

longer prompts my AI made lyrics to a full song. but ur prompt window is too small

[autumn pilot]

It would be great if when you downloaded a song it auto-generated a name for it using GPT, since now it's just a random bunch of numbers.... I've been having to run the lyrics through ChatGPT to name them accordingly to distinguish between songs I like and download.

[graceful tendon]

Can you do one where it shows the song with the backround music and the other video without just the voice singing?

[ripe mango]

Music Visualization option. It would be cool..

[supple lake]

Or it could just do the first line of lyrics

If that's easier

Or the prompt

Dall-E 3 saves using the prompt

[timid wasp]

I wonder if something like a custom GPT bot that helped with lyrics creation is applicable.

Though, I certainly don't know how the lyrics interact with the model to create the vocal melodies, phrasing, etc, it seems that perhaps right now it's just feeding the lyrics prompt straight into ChatGPT?

You should at least be able to select the lyrics' genre style (Black metal lyrics iver pop music?) - but Writing a custom GPT that takes into account things like the lyrical methods, different rhyme schemes depending on genre, etc, would be super fun!

Is the model like, oh X genre has Y style vocal delivery so that's what the model uses to take the lyrical content from words to vocals that fit the song? 🤔

[glossy junco]

can't see full lyrics on the "my clips page" - trying to find previous song, however I only remember the last line which I can't see so I'm having to go through playing every single one

suggestion - address this

should be a download button on the player bad at the bottom in the event user has scrolled and decided they want to download,but current playing song isn't visible in the playlist (can just scroll back up, but shouldn't need to do this)

[spiral bough]

Not sure if this has been asked before, but a possible future idea: the ability to feed an existing track to the model as a prompt.

I was thinking about this in relation to a song I'm listening to right now and how I wished there was an ability to make a song just like it in fact, and the thought occurred to me: since "custom music generation LoRA-esque training" is probably months away at earliest, at least one possible early shortcut would be the ability to let the model listen to a track and deconstruct its parts, mood, instrumentation, etc. and construct a new song from it

[spice tartan]

It's just a matter of time, I can't wait to be able to generate music based off my favorite songs/artists/genres

[spiral bough]

So for example if I prompt the Suno model to create a "melancholic indie rock song about the horrors of war," it could take that nearly infinite number of ways

But if I give it that prompt and attach, say, Radiohead's "Codex" to it, the output could be manually steered towards something that sounds just like it. Similar mood and composition
And if I attach Brand New's "137" to it, any output I request from the prompt sounds more like that one instead.
So one is extremely atmospheric and downer piano and horns every generation; the other is more of a dark and moody grungy track with every generation.

I have no idea how feasible that is. However I relate it to something like img2img

Especially way back in August/September 2022 when ControlNet, LoRAs, textual inversion, and Dreambooth were not yet things that even existed for Stable Diffusion, img2img was the first taste test of "focused" generation
This would be similar to that. Not really training the model on a song or artist, just giving it a framework for a sequence of outputs.

[fringe knoll]

...and then Radiohead sues the pants off you, as they do.

https://www.rollingstone.com/music/music-news/lana-del-rey-faces-radiohead-lawsuit-over-songs-creep-similarities-201487/

[glossy junco]

+1 for more radiohead sound though I imagine that'll require more training and who knows what their stance is on copywritten trainnig 👀

[spiral bough]

Probably should clarify again, I'm not saying they're training on Radiohead
I'm saying the model first listens to a Radiohead song and tries making something that sounds like it, without having been trained on it previously

Think img2img instead of LoRA

Like it listens to the song, breaks down the basic gist of it (slow melody, pianos, certain chord progressions, etc.) and then applies that to the prompt.

Kind of like style transfer

I was thinking about this in terms of a basic MIDI progression I had that I wondered if it was possible to feed that into a music generation AI to turn that progression into something with instruments

Then I remembered that is possible because researchers have done it years ago
But I have no idea if the Suno team has the ability to do it in a more "applying style to a new output" sort of way

Case in point, from 2018:
https://www.youtube.com/watch?v=YQAupr7JxNY

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

Though in this case, less "copying style from one artist to another" and more "listening to one's style and applying it to a new generation without training the model" (or at least it's only "trained" on that one song when you upload it)

I've no idea if I'm even making sense with this, and since chatGPT's down, I can't even feed my rambling into it to whip it into coherency

[autumn pilot]

It would be cool if we could regenerate while keeping the lyrics, so like on the discord bot you can regenerate , but on the WebUI

[feral aspen]

#soon

[frigid timber]

https://sonoteller.ai/

you can try this out, I use this if im going for a specific style from a song and cant quite describe it

[tired yarrow]

Would be amazing if there was a way to use an audio file to train a character/person's voice and use it inside suno AI. Of course, would be even more amazing if you are able to create a voice with "standard" or "low" quality, which will be free, and one "high quality", which will be paid.

[spice tartan]

I think that is the #1 request

[zenith gorge]

Continue and Cut button Sprite Ideas [Cut is essentially meant to be a Way of deleting a clip from a continue chain or just delete a bad chirp from your library]

[knotty moth]

There is no suggested Suggestion format?

SUGGESTION: Name the downloadable file after the prompt. Would save us a ton of trouble. This idea got enthusiastic support in #💬┃general-chat

[fluid crag]

Hello there 👀 I would love to share 2 suggestions, maybe they happen to be there already but i was told to come here.

**1. More payment options. **
Self explanatory xd Some people - me for instance - rarely use master/credit card. 👀 personally im more into quick payment methods like BLIK for example. [ about blik just in case: https://blik.com/en ]
I'm aware it's probably not gonna work for some countries but still more payment offers and general accessibility with diversification of the options would turn positive. 👀

2. One Month Sub/ Gift Sub
Ability to buy subscription specificaly for 1 month (without recurring), or one month as a form of gift to someone, both options would be great 😊

[gentle wasp]

Hello excellent product, first of all. I don’t know if this is possible but I unfortunately waste a lot of credits because I can’t seem to control the gender of the output song (website). A firm way to tell the AI “male singer” or “female singer” would be much appreciated. I haven’t gotten any reliable way to do this via prompting, maybe there is one.

[topaz steppe]

Can we get the continue feature in the website also. I have a few songs i'd love to make bigger.

[barren radish]

its coming.. maybe even this week

[formal bear]

maybe?! 😄

[delicate cedar]

https://tenor.com/view/soon-gif-19121692

[supple lake]

I hope we can continue some of our old website generations...please! There are some I've made I just love.

[oak ravine]

instrumental and acapella options?

specific chords?

[delicate cedar]

Instrumentals happen sometimes when you don't want them...

But you can try putting instrumental in the style section if using the Discord version.

A capella, you can also put in the style section.

[teal timber]

less censorship, can't even use the words deceiver

[teal timber]

make the a.i less biased

[teal timber]

good to know, I won't be able to support a business that censors and edits, under the pretense of "for our safety" is just a lie, to control the narrative.

[rose seal]

lol nope, we just wanted to get the model out so people could try it

[molten orchid]

Well, we will soon see the feature in v2 to be able to edit the lyrics manually, and that's as free as you can get

[teal timber]

Hopefully last thing the world needs is more censorship and bias.

[molten orchid]

I'm curious how it will look like in the browser. Because right now you prompt it and it shows you the generated lyrics and then proceeds to generate the song without caring if I liked the lyrics or not. So I guess there will be a "submit" button after the lyrics are generated with the ability to edit them before submitting it? But Ideally I would expect to be able to also iterate between different generated lyrics, by just clicking "next lyrics" before deciding on the best one

[molten orchid]

why would an English word be banned in the first place lol. Even banning the n word would be ignorant because you would not be able to make authentic "black people's rap music".

[fringe knoll]

Authentic "black people's rap music" is made by authentic Black people.

[molten orchid]

Well you know what I mean

[fringe knoll]

so do you.

[molten orchid]

So you think there should be censorship?

[fringe knoll]

yep.

[teal timber]

disgusting.

[molten orchid]

I disagree

[fringe knoll]

because you want to say n word?
gosh. so liberal.

[teal timber]

no I don't want to say the n word. I want to say the word deciever.

[fringe knoll]

and this isn't the channel for this.

[teal timber]

but if you have censorship you will have it across the board.

[molten orchid]

No, because censoring English words is ignorant. Censoring sentences I can get, but words?

[fringe knoll]

what are you going to do about it? Write a letter to Sam Altman

[teal timber]

Not promote it or spend my money simple.

Open source will catch up.

[fringe knoll]

we're waiting.

[fringe knoll]

🤷‍♀️

[teal timber]

then the word must be abrahamic

[fringe knoll]

Someone is a 'disgusting 'deceiver... that's for sure.

I'm sure it was the torture references.

[teal timber]

typical character assassination type of person you are. how liberal.

[fringe knoll]

you were saying you could use a word.
Do you need any more help with your prompt?

[teal timber]

I don't need help from individuals like yourself.

[fringe knoll]

I love this song. It's catchy.

[molten orchid]

you can actually use the n-word I just tested it, so that's not even censored

[fringe knoll]

so what are you complaining for?

seems like we proved this is not an issue.

[teal timber]

It's even worse they censor religion.

[fringe knoll]

you are funny.

[molten orchid]

I aint complaining, he said an English word was censored so I expressed my concern by an example, and then you came out of nowhere being a karen

[teal timber]

I keep getting error prompts for my lyrics about my Gods, so what is funny to you is a spit in the face to me.

a literal karen.

[fringe knoll]

you can PM you lyrics. Maybe I can help.

[teal timber]

The bridge is already burned.

[molten orchid]

Sometimes you just need to refresh the page for it to work again without errors (if you are on the web app)

[teal timber]

I did, and tried same prompt, just wouldn't do it.

[fringe knoll]

you two understand you are the Karens complaining about imaginary things, right?

You do understand you are not employees here, right?

Karen?

[molten orchid]

I said: "why would an English word be banned in the first place lol", I didn't directly complain, you started replying to me with disagreement about my general point of view, when was I ever complaining about something I didn't know?

[fringe knoll]

I'm going to delete this whole thread, because I don't want to punch down or embarrass you any further.

[molten orchid]

"embarass yourself" you mean. But I get it, delete this mess

[fringe knoll]

just letting you two have your tantrum about made up censorship.

which isn't even true.

[molten orchid]

I ain't checking if the word deceiver is censored by wasting my credits, even though I did waste on the n-word right now

[fringe knoll]

I just proved it. Do you need me to post the clip again?

[molten orchid]

Just letting you know, @teal timber raised his issue, I never assumed it was indeed the situation, but I've heard about it before so I expressed my point of view on the subject raised. That's it, never assumed it was 100% accurate because I don't know, but I am against censorship and you aren't, end of story