🎵｜stable-audio

[coarse lantern]

@fiery ruinYou're the only staff online atm, so have at this compromised account

[fiery ruin]

I got a perfect hammer for all of that 🙂

[stone rose]

Why the hell is this posted here?

[river latch]

Still got that hammer handy? #🎵｜stable-audio message

[fiery ruin]

yessssssssssssssssssssssssssssssssssssssss

[river latch]

Thankies

[fiery ruin]

Np! KaBLAM!

Ty for letting me know!

[coarse lantern]

So do i for the server i mod on xD Though, need to reprint hammer head as the damn thing got a few bad layers from 3d print lol

[serene solar]

In the heart of the American Southwest, nestled between towering mesas and endless expanses of desert, there existed a town that time seemed to have forgotten. Eldridge, as it was called, was a place where the past whispered through the creaking timbers of weather-beaten buildings, and shadows clung to the edges of every sunset.

[still crypt]

I didn't get it done in time

[shut saddle]

Did you get it started though?

[still crypt]

I got through a chorus and a bit of a verse

[pulsar bridge]

would be amazing if you could post up a few "best practice" prompts to use in stableaudio, just got pro and would love to get a better idea of the best prompting template/stylke to use, cheers

[limpid salmon]

a future thought, some type of clipvision detecting video motion changing music intensity or something

[grizzled imp]

" Genre: Cinematic, Suspense, Instruments: Tense Strings, Ominous Brass, Pulsating Percussion, Piano Accents, Style: Classic Thriller, Mood: Intense, Foreboding, BPM: Variable, often Slow to Moderate, Features: Evoking Alfred Hitchcock's Iconic Thriller Atmosphere, Sharp Melodic Turns, Crescendos and Diminuendos, Unsettling Harmonies, Designed to Build Tension and Suspense, Perfect for a Scene of Psychological Thriller or Mystery."

[supple lava]

/audio

[open kiln]

Hi, how do you create videos like this?

[grizzled imp]

with stable diffusion (animatediff)

few others stable audio

[junior notch]

hi all, this is my first post here and i've never used discord before. i wanted to share my favourite audio tracks that i discovered using stable audio. i'm going for a tropical house/chill/uplifting vibe similar to kygo's early remixes/avicii vibe. if anyone is interested, please check out these quick 45 second tracks made by stable audio--all credit was given to stable audio in the video description

https://www.youtube.com/watch?v=6j6wG2zDVC4

https://www.youtube.com/watch?v=qLkZZyrEzY8

Let me know what you think or if there's a better place to share these audio tracks. Thanks all<3

[dense falcon]

hi! would appreciate any pointers for generating audio that is not full tracks, but various isolated single note tones and timbres that can be sampled and turned into playable sampler instruments. i've had some great results, hoping to optimize my prompts and take it further. thank you!

[junior notch]

https://www.youtube.com/watch?v=X42bnPHLr5s

javolk - luminara
generated by stable audio

[limber citrus]

Is there an "AI upscale" for audio out there? Like, something that uses a model to take a low-quality audio clip and interpolate a higher res? e.g. take an 8-bit sample and create a 16-bit version that sounds similar to the original?

[wraith patrol]

Generate a 3:45 mins song with prompt Pop, Pop-Electronic, Ballad, Billboard, Drum Machine, Bass, Lush Synthersizer Pads, Synthersizer Arp, Synth Bass, Vocal Sample Chops, Percussion, Honest, Heart-Felt, Melancholic, Vibe, Cool, Modern, Atmospheric, 115 BPM

[junior notch]

Hi all,
I wanted to share a track created 100% by stable audio using the stable audio model: stable-audio-audiosparx-v1-1
please let me know what you think...sound quality appears to be better than what suno ai can do.

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

[worn wagon]

That's awesome

[junior notch]

thank you

[mental drift]

Hello! I'm wondering if there is a plan to add a feature to mass/bulk download generations from the web app. I usually generate multiple results per prompt and it can become a workflow snag having to click and download one file at a time. Thanks!

[junior notch]

Hi all,

here is a song generated by stable audio which i slowed and reverbed. I hope you like it:

https://www.youtube.com/watch?v=_vyXk1XdYpc

[spring summit]

https://youtube.com/shorts/SiStx-rXGiA

[drowsy reef]

Can "Stable Audio" generate an embedding with audio to find similarity within a vector db?

[hollow dune]

@shut saddle did amli leave SAI?

[shut saddle]

That's not something I know the answer to. 🤷‍♂️

[sly jetty]

she's no longer at Stability. I believe she's working now with SplashMusic, she's very active at twitter.

[uncut sable]

how to use stable-audio to make music with discord? could anyone give me example prompt, thanks

[shut saddle]

Unless SAI changed the discord bot recently that I am not aware of, you can't use discord to create music though stable audio.

[sly jetty]

hi, as @shut saddle said, you can't use stable audio at the Discord server, you can use it at https://stableaudio.com/ Also there you have examples on how to prompt and a complete user guide that explains prompting, the interface and the model.

[nimble narwhal]

https://tenor.com/p2xtJRbGtB7.gif

[pallid sapphire]

The scene depicts a family of three – mother, son, and daughter-in-law – devoutly practicing their Buddhist beliefs, engaged in a prayer for blessings.

[thorny coyote]

idk man I think it's just a fella dancing

[junior notch]

https://www.youtube.com/watch?v=BswTEqxF3u8
generated with stable audio - all credit to stable audio love the work you guys are doing

[pastel totem]

anyone know what the purpose for .index files in RVC is?

[rare wind]

Hey guys - which is the currently best Prompt2Sound Model with a GUI?

[floral comet]

its the accent control of the cloned voice

[pastel totem]

Is there a particular repo you recommend for training voices that generates index files?

Im using audio-webui but it doesnt generate index files

[floral comet]

i always use rvc disconnected to make rvc models, works on free colab too: https://colab.research.google.com/drive/1XIPCP9ken63S7M6b5ui1b36Cs17sP-NS

but if you train local try applio

https://github.com/IAHispano/Applio

[brazen sandal]

Is stable audio available through API?

[shut saddle]

If you are referring to the audiosparx models that you can find at https://www.stableaudio.com/, then no. Those models are not getting released. Fauno made a stable audio GitHub repo that you can use to run stable audio models though.
https://github.com/Stability-AI/stable-audio-tools

The main setback for that program is you need a model to run stable audio tools. SAI hasn't released a universal audio model just yet

[shy mulch]

Is it planned anytime? Even if only for memberships? 😉

If I remember correctly, there were some annonces about a possible open source version, I've read it some where (possibly reddit).

Thanks!

[shut saddle]

The Harmonai team has weekly office hours where they discuss the current progress of the open source project (and sometimes offer sneak peeks on how the models sound) Office hours are scheduled for most Thursdays and the next one is scheduled for <t:1706814000:F>

[vocal helm]

Where/how can one tune in to that?

[shut saddle]

You can join the Harmonai discord and be on the lookout for when they start office hours. I can't link the discord (this discord prevents posting direct discord join links), but go to the Twitter/x post for the discord link.

#1072229057707659404 message

Their current posted discord link is expired so you'll have to use the Twitter one instead.

[wintry stag]

@still crypt I dont know how to get to you, but dont answer my old account, it was hacked

[slender valley]

Is there any ways to disable NSFW Filters on SD?

[vocal helm]

found it! thank you!

[shut saddle]

Awesome! Sorry I just couldn't link it here.

[vocal helm]

@shut saddle I'm doing some research on generative AI for music and Ai tools for musicicans in general... What are some other tools you use/find interesting?

[shut saddle]

There's a few projects I find interesting. Vocal conversion programs like so-vits are pretty neat.
https://github.com/svc-develop-team/so-vits-svc?tab=readme-ov-file
Suno and MusicGen are probably the more prominent audio generation tools at the moment with Suno even having text to song capabilities.

[vocal helm]

I'll check it out!

[low wasp]

VStudio

[vocal helm]

This is kinda crazy! I didn't know we were so close to realtime voice conversion!!!

[sullen pond]

can i feed in a video and it creates audio synced with the video? like sounds for karate kicks

[shut saddle]

This doesn't have anything to do with audio diffusion.

[modern idol]

Have nice continuation on this one ♥

[hushed quarry]

Hi everyone, how can I compulsorily let stable-audio NOT doing something by using the prompts?

[random verge]

Can anyone recommend a best practice/pro guide for stable audio?, all the youtube videos I see are a bit old and meh

[grave breach]

meh

[random verge]

xd

[gaunt mica]

So far the best guide is on stableaudio.com

[vocal helm]

Truly, just use the guidelines on the website, but my recommendation is to experiment a lot and learn that way, sometimes you can get surprisingly good results with pretty strange prompting with this model

[tacit breach]

这个软件还能用吗

[pliant elk]

this is for stable audio, but I bet y'all never heard what stable diffusion sounds like under Coil whine >:3

kinda sounds like an old printer, makes sense given it's making images xD

[next gyro]

Thats the sound of the future happening

[pliant elk]

XD yes

[bleak lodge]

https://stableaudio.com/1/share/2d99bcac-a768-4c19-8118-1f527ee2234d

[mental merlin]

any status on an api being available for stable-audio?

[shut saddle]

There is an API available for stable audio, what is missing is the weights.

[mental merlin]

on the docs I only see text -> video, text -> text and image -> image endpoints

[woeful torrent]

good afternoon, I'm trying to figure out which model is better for generating sound effects (for a game to be precise)
met audiocraft, stable audio and dance audio
In your experience, which one is better?

[coarse lantern]

At the time of posting, there was only one model available. So absolutely no clue if there's more today

[acoustic steppe]

I can't brush up well on the sample, but I wonder if the chemistry of the prompt combination is the deciding factor.

[shut saddle]

I'd say the main factor comes down to our lack of ability to convey what we want to get out of our audio outputs. It just seems like it's much easier to describe a picture with words than it is to describe music using words.

[worn galleon]

Hey everyone. I have a question maybe the community can answer! I've been using the stable audio website for a while and it's come out with some pretty good sounds! But the audio quality isn't there. Is this because I'm on a free tier or is it a limitation of AI generated audio? Sounds like a bad mp3, or worse!

[still crypt]

Afaik paid tier doesn't sound any better. Better prompting can often provide cleaner results. In my time tinkering, I found that creating short, repetitive solo-instrument loops yielded the highest quality outputs. I lean towards obtaining clean stems, then combining, editing and processing them later. If you're generating full-fat compositions, it will be hard to get uber-clean results. That's the current state. No difference between paid and unpaid output except for max duration.

[worn galleon]

Thank you so much for the detailed reply. great idea to try to get clean stems and then work on them. I think this is the limit of generative ai audio as it stands. I remember reading an article which said it takes A LOT of processing power to make it. I wonder how long until it starts making better quality audio?

[still crypt]

My guess, not long. Video is already at the point we thought it'd take 5 years to get to. So ultra-clean high-quality audio might not be far off. In my uneducated speculation, the biggest issue is the availability of good training data.

[deft gazelle]

Haven't logged into Stable Audio in a while and got this, anyone know whats up?

[junior aurora]

I'm seeing the same thing. I've never used it before and was stoked to check it out. Looks like something is bunged up with the DB in the backend so it will likely have to be resolved and redeployed before it will work for anyone.

[deft gazelle]

Unfortunate. It's a lot of fun to play with when you do get a chance.

[shut saddle]

@tight anchor who on staff takes care of the stable audio website? Looks like there's a System Error.

[deft gazelle]

Figures, the first time I try to use it in like a couple months lol. Sorry fellas, blame my luck

[shut saddle]

You could try refreshing and / or trying a different web browser?

Might fix the problem, might get the same error.

Yeah so Firefox is giving me that error too.

[junior aurora]

tried in multiple browsers with cache clearing page refreshes

no dice 😦

[shut saddle]

Looks like the website is back up guys. I got a generation that went though.

[tight anchor]

yeah sorry, it was down for a bit. Should be all good now.

[shut saddle]

Thanks Fauno.

[deft gazelle]

its time

I thought adding reference sound file was a thing last time I used stable audio, perhaps only a paid feature?

[junior aurora]

I was able to use one of the clips I generated as an input but didn't immediately see the option to add your own

I was pleased with the first generation: https://stableaudio.com/1/share/445b0c36-7725-4bc6-aabe-221a988beb15

I was going for a Fantastic Planet soundtrack vibe (https://www.youtube.com/watch?v=RHyP3tUt3V4&t=23s) and it was pretty spot on with the vibe of the instrumentation. Nothing mindblowing but I'd throw down some bars if I was a bar thrower downer.

I fed it back in and liked the second one too. Was looking for more of a lead sound but hey, first two tries and I wasn't completely disappointed. https://stableaudio.com/1/share/1dbfce3c-b2da-416b-9f7e-3a7c7e62f213

[deft gazelle]

I'd throw down some bars if I was a bar thrower downer.
I'm saving quote lol

I forgot how dangerous mentioning "synths" was. They just tend to overrule everything. No matter the wording.

Anyone have any input on how much the 1.1 beta improves? Considering buying back in, if it helps a lot - and because I wanna queue multiple tracks.

[vocal helm]

Sounds better in general, less artifacts, but the max length is 45s. I still use 1.0 for some specific things but 1.1 is overall much better

[hard steeple]

https://soundcloud.com/at-odds-69/ruby-1?si=9300f9d76f1f4a8d9828e6bb0aa7bf13&utm_source=clipboard&utm_medium=text&utm_campaign=social_sharing Loads of stable audio based sound design in this hope you enjoy!

Made during my process of QA on the stable audio models =]

[hard steeple]

I really forgot to reply to these eeeeeeek! there is a user guide that's based on my notes in the stable audio website I think the button to find it is in the prompt box =] if you have any specific questions on how my opinion has deviated from that since I'd love to answer. Hopefully abit more timely hahahahahahahahahaha

beeen very busy

[shut saddle]

Poor guy waited 2 months for a reply 💀

[hard steeple]

This is abit of hacky way to do it based on how the CLAP model was trained.... but you can use Format: Solo...|Instruments: Violin...|.... Just change the parts for what you want it to do ie....Duo, Trio etc

Still trying to figure/get round to sensible way to incorporate that UI wise

[hard steeple]

I know i'm a bloody idiot

[shut saddle]

Maybe give them boxes to fill in? Idk

[hard steeple]

I so forgot I posted that in there

well yeah of course

that's not for me to do hahahaha they have been told though hehehehehehe

[shut saddle]

I mean it's up to sai how they want to do their stable audio website ultimately. You don't want to fully rely on tokens, but they sure do make things easier to hammer out prompts with.

[hard steeple]

We're working on deeper level stuff to help with prompt fidelity and accuracy. =]

[shut saddle]

Except if you do that it's trickier to make up gibberish and call them instruments.

[hard steeple]

but it be nice to streamline this with what's already there toooooo

[hard steeple]

lmao

[shut saddle]

I'll have you know I'm an expert Blamash player.

[hard steeple]

Oh yes the blamash beautiful bellows that makes hahahahahaha

[dense falcon]

thanks for the reply. so “Format: Solo” and “Instrument: [description]” become part of the prompt? separated by commas along with other stuff as usual? (“breathy, irregular, B major”)?

a couple of examples of prompts that resulted in usable audio:

"bell-like instrument, solo, no accompaniment, single notes, unison, C note, high quality, 44.1kHz"
"unison, wind, woodwinds, forest fire, soaring, bird-like, high quality"

would love examples that incorporate your tip. thanks again!

[hard steeple]

you see how i've formated it with | that's the important part when you seperate what is essentially pointing towards metadata fields

no spaces when you put one in as well

I know it's wierd

it's abit of a backdoor

so "bell-like instrument, solo, no accompaniment, single notes, unison, C note, high quality, 44.1kHz"

this would look like

Format: Solo|Instrument: Bells

you can add a | after bells and add more if you want

[rare wind]

Is there a good Sound AI with a GUI yet?

[sacred compass]

Just noticed this typo in the Interface guide

[shut saddle]

I think that might be one of you problems.

[lunar matrix]

Hello , can we use music generated with the ia to generate money with spotify for example? I talk of course with the subscription adequat , but once the subscription is finished, it still works?

[bitter galleon]

What does Stable Audio do?
What does it take as input, and what does it produce as output?

[lunar matrix]

what?

[bitter galleon]

To use Stable Audio, what do you supply as input? And what output does it produce? What does it generate?

[lunar matrix]

I would like to generate electro music for music platforms

[still crypt]

[ruby bronze]

Probably, but just not from here

I saw some YouTube and Spotify videos on the media page of this thread

[tiny bison]

just wrote this song, lemme know what you think!
https://youtu.be/YZYCehncT0c?si=7hiSEBCqW7592HH6

[lunar matrix]

thank you but if my music remains several years on my spotify page , but that the" stable "subscription lasts up to 1 month it works how?

[tiny bison]

just wrote this song any producer up to make a track for it:)?
https://youtu.be/uSG_wPaQbOk?si=b2xAD0pAVwZsleGL

[still crypt]

what do you have in mind?

[tiny bison]

NEW SONG ABOUT BEING LED ON!
https://youtu.be/g5MN14GZrHE?si=SZ8bbguJCrc5hya1

[tiny bison]

base it on my guitar

[burnt wing]

any tips on making a 5 second catchy jingle as a youtube opener? if i set the duration to 5 seconds it just sounds like the start of a song that's not complete.

[still crypt]

Make it 60 seconds, use the keyword 'loop'. Trim to the duration you want. My best guess.

[minor glen]

is the model available to run locally?

[shut saddle]

Not yet. Coming soon ™️

[tidal lava]

Soon™

[strong anchor]

any word on API access?

[shut saddle]

https://www.stableaudio.com/ exists. There's an unofficial stable audio tools gradio created by DionTimmer. The problem has been that there is no officially released model to run it right now.

[sweet wharf]

Hello everyone, my name is Angelo, and I'm the co-founder of BitSong. I would love to discuss a potential collaboration with one of the developers. Who can I contact regarding this?

[shut saddle]

Yesterday, the Harmonai team were showing off the early version of a 47 second freesound open source model that's currently in training. It sounded pretty good, so some good news there!

[copper tundra]

nice

[tidal lava]

so , on the licensing for Stable Audio, says the generations cant be used to train AI models. im really curious how well that policy would hold up legally speaking if the generated audio were edited /remixed /whatever and then used for training

fair use laws are kind of fucky... of course, i guess stable could ban from the platform and leave it at that

[clever geyser]

HI

[warped oasis]

Where do I find the Volume Slider?

[warped oasis]

🧐

[rigid dagger]

good question 🙂

[shut saddle]

I guess I get to post this first then...

https://twitter.com/StabilityAI/status/1775501906321793266

Guys, the newest stable audio 2.0 model is available right meow. Go check it out! Each prompt is 2 credits, but you can generate the full 3 minute tracks now!

Stability AI (@StabilityAI) on X

Introducing Stable Audio 2.0 – a new model capable of producing high-quality, full tracks with coherent musical structure up to three minutes long at 44.1 kHz stereo from a single prompt.

Explore the model and start creating for free at: https://t.co/E9ZIGagmPf

Read the…

Twitter · Apr 3, 2024, 12:33 PM

https://www.stableaudio.com/

Not only that...the radio is back!

https://www.youtube.com/watch?v=yvOXZ6SV2Rk

[glad cargo]

wow, i love closed source model cash grabs

[warped oasis]

I will have to wait until we get the volume slider update though

[copper glen]

lets try this

[long creek]

Very curious how the model does on advanced beatboxing techniques

Such as humming and beatboxing at the same time, or stuff like https://www.youtube.com/watch?v=5EADJGrNK3o

[copper glen]

i think it's haunted

oh vocals (sort of), neat

[copper glen]

the like button on the generator is occasionally not doing anything

[tropic peak]

it seems like the autoencoder has a "snake" block in it.

Anyone want to confirm-not-so-confirm that it is a mamba SSM?

[frozen siren]

If I want to create a 24/7 channel on youtube and monetize it is it possible?

Can I monetize my music on Spotify?

[copper glen]

[limber citrus]

Whether you can monetize AI-generated music on Spotify at all is still up in the air. https://www.bbc.com/news/technology-66882414

[harsh steeple]

on the reddit post about stable audio 2, Emad mentioned among other things, the word "comfy", so does it mean it will eventually be available offline?

[tight anchor]

it's not a Mamba SSM, Snake activations are their own thing

[rose wharf]

Is there any plans to release the weights of Stable Audio 2?

[shut saddle]

We need a pinned message since this gets asked a lot.

[limpid ice]

According to that particular Reddit thread. I think Emad did mentioned that they will release another model trained on another dataset

Stable Audio 2 is pretty good though

[shut saddle]

Yep. The freesound open source model is still cooking. They used audiosparx to train a really good autoencoder though.

[limpid ice]

It is crazy how many of them wanted the model to be released, but I think that make sense.

Several people said that it is not as good as Suno, which I agree. One of them said that Audiospark stock music are not usually good so it turned to be not very good.

[novel frigate]

does stable audio has an api?

[warped oasis]

I just wish the license wasn't so restrictive.

[tight anchor]

the autoencoder is also trained on freesound

[shut saddle]

Oh, I didn't know. Interesting.

[thorn roost]

Hi all! Is it possible to generate voice for the songs?

[vocal helm]

Got this message while generating:

Error - ClientError: Received client error (400) from model. See the SageMaker Endpoint logs in your account for more information.

[sweet torrent]

Is it possible to upload public domain songs? i.e. classical music that is over 200 years old and definitely in the public domain

[still crypt]

Depends.. are you generating the song? If someone records a version of the moonlight sonata, they don't own the composition, but they do own the audio/recording. Are you reuploading someone elses recording of a classical piece, or uploading your own version of it

[sweet torrent]

Thanks for the reply! Beethoven's recording of Fur Elise is in the public domain for instance. Can that not be uploaded? Composed in 1810

[warm belfry]

Metallica

[still crypt]

It wasn't recorded in 1810. You need to find out whether the actual audio recording is in the public domain or not

[sweet torrent]

see here: https://imslp.org/wiki/Für_Elise,_WoO_59_(Beethoven,_Ludwig_van)

[still crypt]

The only recording you can legally use/for money is the last one, where the copyright is in the public domain

[indigo badger]

🕺

[solemn salmon]

love it!!!!

is there an api for stable audio 1.0 and if not when is the api for 2.0 coming? Thank you!

[light sable]

Attempted some t e l e p a t h style slushwave manipulating a sample I generated 😛

[indigo badger]

Some of the music stable audio can create is actually kind of interesting but I honestly much more enjoy just seeing what kind of hellspawn I can create

[crimson pollen]

woow this is huge. produced using v2

It may not sound interesting or even boring to you, but from a musical point of view, this thing is terrifying in its quality of playing, sustain, creativity and instrumentation, even better than Suno in that respect, and it just needs a little humanisation.
Stability AI is sweeping the field.

We'll see in a year's time where it will go, and this is a company that is bigger than Suno, rivalling Open AI and has a lot of experience in AI!

[light sable]

Oh my 😳😂

[light sable]

Final boss music

https://www.stableaudio.com/1/share/cdfe367b-b2c2-4b86-9c67-2c2cf3fc84f2

For such a silly prompt there is so much sound design potential here

[crimson pollen]

Seems like a hunting vibe of music of scary movie for me but It fits also as a boss theme.

[crimson pollen]

2nd try for the terrifying-core hunting piece of music

[crimson pollen]

Seems enhanced

[crimson pollen]

Image testing this on v3 lol

waiting for it

[glass gorge]

/a cinematic lyric opera vocal melody, powerful woman's voice

[fiery mortar]

https://stableaudio.com/1/share/398ce56b-a5ee-4d7f-886c-01c02529a996

[mint shale]

Here is the image you requested.

[fallow parrot]

Hello community, is anyone having problems using stable audio? because from the first stable audio model I want to start using it and it gives me a sign saying that I was blocked and it won't let me enter

[patent hemlock]

Can you add lyrics here like riffusion?

[noble gull]

Stable Audio with ElevenLabs SFX!
A perfect match, took a few hours to concoct this one! Pretty happy with the results.
https://www.youtube.com/watch?v=gpRMX07VNIU

[vast narwhal]

Here is my demo via Stable Audio.

https://vm.tiktok.com/ZGemgnea5/

[empty lava]

Hey all! having so much fun with the new model. can anyone tell me what the licensing implications are for downloading the music, reworking it into a derivative version (like a remix you would make with Ableton etc) and then releasing that? I know this is a question for the team so point me in the right direction if you can. Thanks!

[shut saddle]

According to the website: "Use the music you create with Stable Audio in your commercial projects."

[woven canyon]

lmao that reminds me of some of those overdramatic bollywood movie scenes where everything is in slowmo and shown from 10 different perspectives

[drowsy silo]

Hi, I've received the following error on a prompt: ClientError: Received client error (400) from model. See the SageMaker Endpoint logs in your account for more information. Any idea as to what wrong and where can I find these logs ?

[vocal coyote]

"Sorry, you have been blocked
You are unable to access stableaudio.com"

[woven canyon]

made a little track with only stable audio samples and my new favorite plugin (amigo sampler) 🙂 https://www.youtube.com/watch?v=c3NivDcgAA8

[hazy oar]

stable-audio 2.0 is really good at synthwave

almost perfect

I love that AI

[lone rose]

how can I extend a song I made using the stable audio website? is there any way to set the model up for audio completion

[patent hemlock]

can stable audio be installed locally too?

[clever geyser]

[hoary linden]

hi guys! any tips for consistently generating new sounds in the same chord/scale as the input i provide? i can't seem to find the sweetspot... i find it easier to generate clean percussive material from an input audio, but harmonic stuff seems a lot harder to control

[tight anchor]

Yeah this is totally allowed, and one of the main intended use cases for the model

[nocturne agate]

is there any way i can run stable audio locally?

[remote patio]

The stuff my momma watches

[clever geyser]

[radiant swift]

Animation for the uploaded music, animation is an ancient Chinese style of animation

[still crypt]

Here's the animation you requested

[radiant swift]

A handsome man in ancient costume drinks in a pavilion by the lake, and a beautiful woman in ancient costume lies on a small boat on the lake and bows her head to play in the water

[mint shale]

Here is an image you requested.

[remote patio]

nuh uh

[sage pecan]

doubt it since they offer subscription based stuff? 🙂

tried out the stable audia and played on my paino a small piece with close to 3 minutes duration. unfortunately the output gets cut off. but the monthly fee is too much as I just want to try out a single song ( in general all those AI subscriptions are unfortunately too expensive since i would use them not often enough)... Is there a different pricing model planned in which i can buy individual tokens to produce output? or another option i might have overlooked?

[fresh pawn]

i have something in mind, wanna build an AI music producing tool where people can just sing to create the tracks, that would be awesome

when will stable audio release its api?

[warped oasis]

still no volume slider?

[stuck kite]

Great question

[kindred horizon]

I could use some help. I'm trying out SA. Probably using it wrong. I thought I'd have fun screwing around with a minimalist time-signature phase piece... like the 60s and 70s Steve Reich and friends. But apparently, you can't do accelerandos & decelerandos? I tried using "slow down" and "speed up" as well. I think I can cut and paste in a sound editor to do what I want. But it seems a doable thing? Here's part of the percussion stem I tried to execute.... "Solo symphonic percussion anvil only, no change in pitch, just a simple beat with gradual accelerandos, decelerandos as indicated. Anvil sets heavy 4/4 beat strong downbeat, diminishing echoes on beats 2, 3, 4. Start at 106 BPM, anvil only, 8 measures. Then 8 more measures Accelerando to 112 BPM, anvil only. 114 measures 112 BPM, anvil only. Then decelerando to 100 BPM over 8 measures, finally 100 BPM [ast 4 measures. 1. 8 measures, - solo Anvil sets a heavy 4/4 beat 106 BPM, strong downbeat, diminishing echoes on beats 2, 3, 4. 2. 8 measures 106 BPM gradual accelerando to 112 BPM, solo anvil. 3. 14 measures 112 BPM, solo anvil. 4. 8 measures 112 BPM decelerando to 100 BPM, solo anvil. 5. 4 measures 100 BPM, Solo anvil continues then ends. "

[pearl verge]

hi, I'm just wondering if it's possible to install stable audio locally?

[shut saddle]

This is a GUI developed by DionTimmer, but until a model drops this isn't that useful.

https://github.com/diontimmer/audio-diffusion-gradio

[coarse lantern]

Hoi, is there a A.I generator similar to UDIO , but locally ran?

[still crypt]

https://youtu.be/MRpMV0qD2Q0?feature=sharedhttps://youtu.be/MRpMV0qD2Q0?feature=shared

[shut saddle]

Ok I laughed.

https://suno.com/song/c15f0251-fbac-4a30-a3e1-002dbc78cb79

[remote patio]

embed fail

[wraith vessel]

Will stable audio 2 be open sourced?

[limpid pendant]

not yet, but i eagerly subscribe to Google scholar alerts for text-to-music papers in hopes of finding a solution once any emerges. Suno and Udio seem to be a few months to a few years ahead of SOTA text-to-music open source (which languishes in MIDI and non-lyrical content)

https://paperswithcode.com/dataset/magnatagatune this is probably a critical dataset for any future model

one could probably fine tune a TTS model to music mostly easily with a few hundred $$$ (my best guess) https://github.com/jasonppy/VoiceCraft new model and code drop for audio

[shut saddle]

Stable Audio 2's paper is out.

https://arxiv.org/abs/2404.10301

[limpid pendant]

please open source it!

music with vocals. Our focus is on the generation of instrumental music, so we do not provide any conditioning based
on lyrics. As a result, when the model is prompted for vocals, the model’s generations contains vocal-like melodies
without intelligible words. Whilst not a substitute for intelligible vocals, these sounds have an artistic and textural
value of their own. Examples are given on our demo page.
``` this makes me sad

maybe it could be fine tuned with an emphasis on text-to-music with vocal lines

SOTA text-to-music vocals is still proprietary et al

[coarse lantern]

Indeed. The thing if not intelligible, could possibly for now be used to hum along the tune. Like those dance/trance songs with women who just hums after the song. Or like orchestral fantasy where the woman just says "aaaa, ooooo" along with the fantasy music

[limpid pendant]

i guess one could:

1. produce text-to-speech of the lyrics via the tool of your choice in file1.wav
2. produce a song in the style one wants "anti lullaby, dubstep" whatever in file2.wav from stable LM
3. using stable LM maybe transfer style file 2 to file 1
   4)???

use text-to-speech and then use some style transfer

[patent hemlock]

i wish i could run this locally

Fooocus Muzak

[remote patio]

Must be similar to how people get game characters to sing

Transferring style file (in this case, the original song) to file 1 (text to speech of the character from preferred tool) and do whatever’s next

[patent hemlock]

so how do i run this locally. lol, worth a try....

[harsh steeple]

https://tenor.com/view/gimme-what-about-bob-bill-murry-i-need-gif-19552065

[patent hemlock]

Yes why isn't this one available locally anywayz...

[shut saddle]

It's been explained a few times if you read back through the thread but the short answer is "revenue split with Audiosparx".

[harsh steeple]

im just gonna continue crying ok?

[shut saddle]

https://tenor.com/view/tenset-10set-cat-clock-gif-11918755868618435360

Just trying something. Might work.

[harsh steeple]

[patent hemlock]

so SD basically used these sparx clown's audio library to train and this didn't work out like with Getty hmm?

Doesnt surprise my music industry more copyright anal retentive than visual arts

whatever

theyll go in the dust like hollywood

[shut saddle]

This is an output from the FreeSound model. Sounds pretty good imo.

[signal frigate]

Do you have a source for that?

[shut saddle]

The source is the freesound audio model weights that are coming out really soon.

[signal frigate]

Is there a github page for that?

I have no idea how inference for that is done. Tools etc...

[shut saddle]

This is the official s.ai repo, but without the weights this doesn't do anything.

https://github.com/Stability-AI/stable-audio-tools

[patent hemlock]

i don't care if it's not trainable and I have to use the "base audio model

but is it possible to run locally and is it any good?

and can it do lyrics?

Like is there a webui interface for this like A1111

I can;t do this without an interface

my brain isn't wired like that

[harsh steeple]

ha wired.... i get it... :3

[deep idol]

https://youtu.be/vFE0p8NWLeE?si=S9PtJt-oKOBnvYip

is this ai?

[deep idol]

wait anime name?

[autumn forge]

has anyone else subscribed to the 'studio' tier of Stable Audio but not gotten the promised 60 minutes of upload capacity? Mine capped out at 30 minutes even though I am subscribed to the tier 'Studio'

i sent a support ticket in through the website and never got a response

[still crypt]

No clue, sorry xd

[remote patio]

I believe it’s Lycoris Recoil.

[light sable]

I'm just here tryna feel sorry for myself, then Stable Audio comes up with some crazy trumpet thing at the end XD: https://stableaudio.com/1/share/53e29eb2-15f5-433b-bee2-5e2432efb864

[light sable]

OH MY GOD this is incredible XD please post this on youtube or some meme page

[shut saddle]

It's not mine. Probably should have made that clearer.

[wide frost]

Hello! This is very exciting to hear, since I commonly use Freesound to get SFX, samples and sound design, but almost never seem to find exactly what I'm looking for. Having an audio model based on that website would be revolutionary! Anyway, I haven't heard about that model variant anywhere else. I've heard there is a local version in the works, but this is the first time I've seen results and the dataset mentioned. Would you mind telling me where I can inform myself or where you got that information? Thanks!

[vocal helm]

Join the Harmonai discord 😉

[graceful quest]

Once upon a time, there was a little turtle named Xiaoming who lived in a beautiful lake. Xiao Ming enjoys exploring and making new friends very much. One day, Xiaoming heard that there was a mysterious garden in the forest near the lake, filled with various delicious fruits. He decided to search for this garden.

[still crypt]

[merry snow]

[shut saddle]

[patent hemlock]

[lavish shore]

Are there any devs here interested in working on an audio project with stable audio?

[patent hemlock]

This is like the saddest, loneliest SD room.

[still crypt]

Just how I like it

This is my SD server

[shut saddle]

Tends to happen when no weights 🤷‍♂️

It's coming soon ™️ though.

[harsh steeple]

i hope they release some version of stable audio 😦

i mean the community can then improve it perhaps

[shut saddle]

I was told late May. That's the most up to date number I have.

[harsh steeple]

told by who? just curious of the source :3

[shut saddle]

Stable Audio lead: Fauno15.

[harsh steeple]

kk, well let's hope we have something

[shut saddle]

That model is a 47 second model trained on Freesound's audio library so the idea is that the community could train off of that model when it releases.

[harsh steeple]

agent 47 would be proud :3

[alpine tulip]

how can i run this thing? i have magnet and audiocraft but this might be better?

[midnight heron]

no ai

[hardy turtle]

the stable audio website is crashed or something. the live audio stream isn't running and nothing else works.

[crystal valve]

Why isn't Stable Audio offering the ability to extend a song?

[tight anchor]

because that's something that LLMs do natively, but is trickier with diffusion models

[shut saddle]

The Freesound 47 second open source stable audio model is still on track to be released soon ™️.

[storm briar]

Does uploading audio work for anyone here? It's been a few days and it seems to still be broken and the website and web app is currently barely even functioning.

I suggest making it open source so we could all help making it that much better. The tech is there, just the app is bonkers 🫥

[shut saddle]

@storm briar #🎵｜stable-audio message

Should be coming really soon ™️

[storm briar]

what should? fixing the web app or..?

[shut saddle]

There's going to be a 47 second open source model trained off of FreeSound's audio library released really soon. You will be able to use stable audio tools with those released weights to run open source.

As for the web app, I can pass that up the chain of command and see if there's some bug to fix.

Are you trying to upload copyrighted music? They block copyrighted music btw.

[storm briar]

Got it, thanks, no, I'm literally just trying to upload and record my own sound, whether it's beatboxing, guitar, drums, and nothing is working right now.

It even says the contact support

Since a few days ago, uploads stopped working completely, I've also seen various users report this on Twitter/x

This is the main differentiation by the way from stable audio to any of the other current tools is that you can upload your own audio and modify it.

Without this feature it's simply not as good as audo, suno, elevenlabs, etc (And it's a shame because I'm rooting for stability and want them to do good)

[storm briar]

its same for when uploading audio btw (not just recording) - Looks like the processing fails

[tight anchor]

Notified the service team about this, hoping it gets cleared up soon

[shut saddle]

It's coming soon guys, get your audio datasets ready for fine tuning. 🙂

[jagged dust]

Sweet! Any timeline? Everyone's clamoring over SD3, but this is just as exciting for me as someone who lacks talent to make music from scratch 🤣

[shut saddle]

Harmonai just had their office hours and they are aiming for the release of the freesound model next week, provided there are no delays.

Fingers crossed

That goes in line with their previous goals of having it out by the end of May.

[jagged dust]

That is fantastic news, thanks for the update! I really appreciate it

[shut saddle]

You can keep up with the news on the Harmonai discord as well. That's one of the best places to keep up to date on progress.

[jagged dust]

Just found it and joined. Thanks for the heads-up!

[storm briar]

@tight anchor any updates about the audio ulpoading? still not working...

[tight anchor]

Not sure what's making it time out for you, probably requires more information about your account to debug, have you filled out a support ticket for it?

[storm briar]

worked now.. checking if it works twice in a row!

also, does stablaudio have an API by any chance?

[tight anchor]

oh good! It was working on my side, wasn't sure what was different on yours.

[tight anchor]

It's something we're working on, yeah

[limpid pendant]

soon i have two new music hobby choices - new guitar amp, or more vram 🤔

[storm briar]

@tight anchor Please add ability to download the uploaded audio - its super critical IMO, and should be easy to implement

right now you can only use itas "input", and for history outputs you can download (but not uploads)

specifically relevant whe im using the stableaudio UI to record (not upload files)

[still crypt]

Ran into an issue - when launching the file my antivirus told me it was dangerous. Should I allow it anyway

[astral quest]

No it's a virus

[still crater]

Can anyone from stable audio reply my email? I have sent emails to get stable audio's API and start API integration~~

wish to get feedback and start cooperation asap

[patent hemlock]

is stable audioe better than udio suno or riffusion?

[tired pecan]

Different

[tired pecan]

Say, I have a generation that's stuck generating. Can it be canceled somehow to allow me to generate again?

[shut saddle]

Udio and Suno can create whole songs with lyrics, whereas stable audio only creates audio snippets. They just do completely different things currently.

[small leaf]

anyone else having issues with StableAudio timing out constantly when using input audio on generations?

looking at all of your generations, you're all having the same issue @storm briar

did the issues clear up for you yet?

been paying for stableaudio for about a week

was working fine for me up til yesterday

🤔

[tired pecan]

Yeah, that's what I was asking about above actually

[tired pecan]

No, still having issues...well since last night

[tired pecan]

Actually, no problems so far this morning

[trail fox]

New system to accompany your sonic creations https://www.youtube.com/watch?v=9DbRJDitVhA

[coarse lantern]

Hoi, do you guys know if there's a A.I audio upscaler/sampler to read from the song, and add the missing higher quality that flac would otherwise give?

[still crypt]

I highly doubt this. If it existed, it wouldn't exactly 'give it the quality it would otherwise have'. I think of the example of Samsung phones using AI upscaling on moon photos, and adding details that aren't actually present on the moon. If an AI upscales audio, even if using a relevant and accurate dataset, the result will not be 1:1 with what the audio would have been had it been recorded/processed/kept uncompressed. That said, I'm sure AI audio upscaling will becoming a thing and, at its best, be sonically imperceivable against original uncompressed files. Probably be a while before we have that, because the use case is very niche

[coarse lantern]

The thing is,with the moon photo, it's because the moon is always showing the same face, so samsung could easily just store a bunch of photos from different sades of "sun", or yellow/redness, thus it tricks the user into thinking they took that photo.

But with A.I upscale/upsampling of audio, it'd be trained on 100's, or 1000's of songs in flac quality for that upper range that mp3's doesn't have, and add the missing higher range.

[still crypt]

You make a good point about the moon showing the same face. Those thousands of flac songs don't show the same upper-frequency face though, and their individual quality will vary widely. You will need a tremendously large dataset of each genre of music (to name one of the many desirable variables), in lossless quality for training. Most music is streamed or otherwise available in mp3/lossy formats.

side note, the average audience would rather listen to an mp3 than a flac, because they're used to hearing mp3 quality, and the additional information presented in the flac is perceived as artifacts. At least, that's what I've found with my students and peers. This isn't a recommendation to pursue lowres content by any means - just an interesting note. I prefer wav files any and all day.

I'm an audio engineer, but no expert on AI. My understanding is quite limited in that respect. I think the combination of audio upscaling being a rare use case (right now, at least) and the datasets for such training being so limited, would severely prolong the wait for that iteration of AI technology.

@coarse lantern

[coarse lantern]

Aye Cause like with images, with fingers missing/melted into eacother, it will of course most likely not be "great", but first one need to train a model on the higher ranges :P

All i'm waiting for now though, is a open source version of UDIO, which can actually make quite damn good songs

[lost ravine]

Hello, Version 2 is not working. Support did not reply to my mail. Is there a bug?

[shut saddle]

Weights are here guys!

https://huggingface.co/stabilityai/stable-audio-open-1.0

#😊｜co-creators message

[quick tapir]

Ooh nice

I haven't been following this, are there sample generations somewhere?

Oh here https://stability.ai/news/introducing-stable-audio-open

[plush glen]

hey guys whats the automatic1111 of stable audio open?

[shut saddle]

DionTimmer's gradio GUI probably, but I'm not sure it's updated to handle SAO just yet.

Stable Audio Tools (Main Repo):
https://github.com/Stability-AI/stable-audio-tools

[harsh steeple]

oh nice they finally released something for audio 😮 now we need this in comfy somehow :3

[shut saddle]

Dion's on tour right now so he probably won't update his repo for a little bit.

[small leaf]

@me when a comfyui implementation for stableaudio is ready

Ive spent 4 months working on an animation thats comprised of entirely open source ai tools
and this final hurdle is killing my spirit

sometimes it works; most times it doesnt
this is SPECIFICALLY an issue with Input Audio

[harsh steeple]

can this technically be finetuned as well?

[plush glen]

controlnet for audio

[shut saddle]

Yes, you can fine-tune your own models now!

[harsh steeple]

nice

[small leaf]

do u know what's happening with the timeouts?

[shut saddle]

I don't unfortunately. I'm not as familiar with the website.

Might be a good question for #🤝｜tech-support

[stable goblet]

Thanks for the link. I wish stability AI would not call a model with -nc license restriction "open source". There is - per se - nothing wrong with a -nc license model, it is simply not "open source" in the traditional software sense.

[robust pivot]

bummers, audio-diffusion doesn't run

Does it run on GPU? Because I have AMD

[tight anchor]

There's a gradio interface built in to stable-audio-tools, if you've accepted the model terms on Hugging Face, you can launch it with python3 ./run_gradio.py --pretrained-name stabilityai/stable-audio-open-1.0

[shut saddle]

Congrats on the successful model release!

[tight anchor]

Thanks!

[shut saddle]

So here's the golden question: who's gonna be the first to make a repo for running this on a bot in discord?

[hardy rain]

I have created a Colab Notebook where you can try Stable Audio Open 1.0 immediately. Please feel free to use it.
https://x.com/xqdior/status/1798431457096114345

Since this is an announcement account with low traffic, if you find it useful, we would appreciate likes & reposts.

D̷ELL (@xqdior) on X

本日、#StabilityAI から「Stable Audio Open 1.0」が公開されました。 #StableAudioOpen をすぐ試せるColab Notebookを作成しました。 お気軽にご利用ください。
※モデルのご利用にはHuggingfaceからの申請が必要です。ご留意ください。

https://t.co/HYOgQWfTj7

Twitter · Jun 5, 2024, 7:07 PM

[shut saddle]

Nice work! Thanks!

[small leaf]

are stable audio devs in the tech support channel?
because if not, it wont be solved in there

[wary sky]

i mean this stable audio is not open source lol

[wary sky]

i have hered that nodes will be added to comfyui? and will we be getting code for finetuning? :3

[shut saddle]

DionTimmer was messing around with comfyui node implementation a long while back. I wonder if anything came of that little experiment of his.

[lean tangle]

https://github.com/lks-ai/ComfyUI-StableAudioSampler

[sweet swift]

oh there needs to be a civitai website but for audios

[lean tangle]

RIAA:

[sweet swift]

whats tha

link?

[empty galleon]

Always considered giving ComfyUI a try but now I might as well do it just for Stable Audio 🤣

[still crypt]

[willow fractal]

Dropped a node: https://github.com/lks-ai/ComfyUI-StableAudioSampler .. community is iterating from the model weights

[willow fractal]

gonna look at the gradio tomorrow and see if I can do this initaudio stuff in the Node

[shut saddle]

You gonna join us for office hours on the Harmonai discord tomorrow?

[sweet swift]

it kinda not recognize some music genres

i want to see if anyone finetunes one

[shut saddle]

I'm planning on fine-tuning soon when I get my library in order.

[sweet swift]

remind me if its finished

i would want to try some hyperpop

i also wish stable audio has lyrics input...

the prompt adherence prob aint good cuz its like kinda undertrained (i tried burger mukbang once)

tbh it stucks at 0:47 even though i set the end seconds to 20 and etc

[still crypt]

Perhaps. What time?

[shut saddle]

<t:1717696800:f> on the Harmonai Discord server.

That's 11 AM PST.

[sweet swift]

gibberish amen break

[lime adder]

hey i cant find the node inside comfy ui\

i installed t

ikt

it

and it not apears in my search

.............................

[still crypt]

I'm booked @ the studio. Though I am interested. Another time perhaps.

[shut saddle]

Another time ig.

[patent hemlock]

udio keeps being updated

what are we going to do

how are we going to kick suno and udio's asses?

[wary sky]

i think we shuld start by finetuning the model we have on acually good data.

[rare wind]

Is there a way to use it in ComfyUI without an HF token yet?

[lean tangle]

yeah that's the only thing keeping me from using it offline

[fervent plaza]

https://huggingface.co/audo/stable-audio-open-1.0 use this repository

[rare wind]

Thanks, but this time around I'm patient enough to wait for a GUI. It's pretty neat - I simply didn't check on Audio AI as much... and here we are. 😄

[fervent plaza]

yeah I'm using it with python directly, just thought the extension might allow you to change the repo

[wary sky]

but ther are guis alredy

[rare wind]

There is a Comfy node - but that needs a HF token currently.

[wary sky]

hmm why?

[wary sky]

ther are gradio guis also

[rare wind]

Because it dl's the Model instead of looking for a local one. I guess there will be an update in a day or two.

[rare wind]

I know there is Audiocraft, but there was no update for a month.

[ancient parcel]

Is Stable Audio broken? These are my recent generations...

Almost never works and I'm paying...

[wary sky]

no i think the stable audio github has a gui for ther model

[ancient parcel]

constantly getting charged credits and then getting this error

using the 2.0 model

[plush raven]

https://tenor.com/view/crazy-dance-rave-gif-11332166

[robust pivot]

Hi, I don't understand this part

I downloaded model.ckpt and model_config.json

device cpu? Also I don't see how to select the config file

[signal frigate]

I had the same problem. I put the checkpoint with the config withing a separate folder within the models folder.

For cuda you need to do this here:
Open cmd shell in the folder
activate venv with venv\Scripts\activate

pip uninstall torch pip cache purge pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu121

[robust pivot]

I have AMD! I guess I need to install torch nightly

ty!

[signal frigate]

Oooooh. Ok.

If the model thing still doesn't work. Try renaming the model_config.json to model.json

[shut saddle]

Heads up. That diontimmer repo was made before the release of SAO, so if you are having problems, you might want to try getting the main repo running first.

[signal frigate]

And maybe restart the whole thing. Took a while until i got it recognizing the model.

[robust pivot]

you mean this one? https://github.com/Stability-AI/stable-audio-tools

[signal frigate]

Ooofff, sorry, this is out of scope for me. I am happy that i got my thing barely running. But it feels like breaking every second. I am hoping for a usable comfyui implementation soon.

[robust pivot]

oh no problem

[shut saddle]

Yep. That's the official repo

[robust pivot]

it is only for training or you can generate?

[shut saddle]

I believe it does both but training requires a little more setup.

[signal frigate]

The diontimmer repo is using stable-audio-tools

[robust pivot]

ah but it doesn't have a gui

[robust pivot]

oh I download the .json wrongly

how can I do this?

[pulsar needle]

are there any prompting docs?

[pulsar needle]

yea will need to finetune

[edgy heath]

So I purchased a subscription to Stability Audio and it says Im still on the free plan. Is this a glitch?

[hardy burrow]

Does it work with 8gb of VRAM?

[tulip horizon]

is there some ok gui for stable audio?

[primal hatch]

is there a way to make audio that is for example 4 seconds long without having 43 seconds of silence?

[wary sky]

yes

i just read this post and they say sd auio open is also diffuion trasformer architecture?! is that true? https://wandb.ai/byyoung3/ml-news/reports/Stability-AI-Releases-Stable-Audio-Open--Vmlldzo4MjQyNTQ3

"The model details reveal that Stable Audio Open 1.0 is a latent diffusion model based on a transformer architecture. It leverages a pre-trained T5 model (t5-base) for text conditioning, converting text prompts into numerical embeddings that guide the audio generation process. The model was trained on a dataset consisting of 486,492 audio recordings, including 472,618 from Freesound and 13,874 from the Free Music Archive (FMA). All audio files are licensed under CC0, CC BY, or CC Sampling+, ensuring respect for creator rights while providing a robust dataset for training."

[shut saddle]

Yes SAO is diffusion based architecture, unlike Suno which is LLM based.

[wary sky]

the trasformer thing suprised me not the diffusion thing

i thoght only stable aduio 2.0 was diffuion trasformer

[robust pivot]

ohh it works 🎧

[bleak ocean]

Can anybody point me to a step by step tutorial about setting up Stable Audio Open as a standalone package preferrably with a Gradio or ComfyUi interface? Ameerazam08 has a nice free interface here. https://huggingface.co/spaces/ameerazam08/stableaudio-open-1.0

[bleak ocean]

Adding a link to a ComfyUi workflow - Thanxx Nate https://github.com/lks-ai/ComfyUI-StableAudioSampler

[latent jacinth]

any chance to use stable-audio-open-1.0 with cpu only?

i mean how long does a prompt generation take with a average 4 core cpu?

[quick flare]

I've used SAO on my M1 macbook CPU only and it takes around 12 minutes to generate a sample from prompt

[bleak ocean]

Saganaki22 has a one click installer - which worked for me. https://github.com/Saganaki22/StableAudioWebUI

[shut saddle]

Tried it out and it works for me. If you need a simplistic solution for installing, this one is a good choice.

[wary sky]

banger

"j-pop"

"Electronic, Electronic dance music, Synth-pop, Electro, "

[wary sky]

daim its good xD

[hardy burrow]

no its not

even version2 doesn't give me anything good

the old google AI music lm was actually good but low quality but they nerfed it

[patent hemlock]

So this is out? 😮

Channel will be active? 😄

let's go anywhere i can try online?

[patent hemlock]

lol wtf

sounds like it just gave up and made some random cartoon noises

[wary sky]

Tgis is a different open source Model

[wary sky]

Funny that u say that my prmt said woky cartoon soun effects

[wary sky]

On huging face spaces

[still crypt]

I got a chuckle out of the furry on a train track pfp. Nice.

[south sage]

Hi everyone
I used stable audio(free-tier) in website to mix some music, but it doesn't work.
All of my trials of generation with uploaded input audio always be failed by time out 😦

is anyone happens similar issue?

problems cause only with uploaded input audio!

[wary sky]

this was the promt"clown, Glitch , Goofy, Wacky, cartoon, sound effects, . clown. random sfx sounds, peenis , experimental"

[wary sky]

https://tenor.com/view/neiguex-based-gif-21892118

[stone rose]

This model compared to Udio is like comparing a starting guitarist to a rockstar

It's cool, but I just wish there was more genuine coherency

[wary sky]

Izs more for sound effects

[patent hemlock]

In that case it is extra accurate at prompt following.

[patent hemlock]

literarily... starting to practice guitar.

[silver atlas]

ring ring ring

[steel harness]

is there any tutorial on creating Loras for the new model?

a drummer could fine-tune on samples of their own drum recordings to generate new beats

[pine leaf]

ERROR: No matching distribution found for pedalboard==0.7.4 who knows how to solve this problem?

Defaulting to user installation because normal site-packages is not writeable

ERROR: Could not find a version that satisfies the requirement pedalboard==0.7.4 (from versions: 0.8.2, 0.8.3, 0.8.4, 0.8.5, 0.8.6, 0.8.7, 0.8.8, 0.8.9, 0.9.0, 0.9.1, 0.9.2, 0.9.3, 0.9.4, 0.9.5, 0.9.6)
ERROR: No matching distribution found for pedalboard==0.7.4
but Stable Audio requires 0.7.4.

[thick zephyr]

😁

[thick zephyr]

Im not sure if this will help, but I tried this repo and it worked on my PC https://github.com/Saganaki22/StableAudioWebUI

[shut saddle]

Nobody I'm aware of has made a walkthrough except for @spring girder

https://www.youtube.com/live/ex4OBD_lrds

She was on the ball day one. Actually kinda impressive!

[wary sky]

I woder if they want to qd comfi ui suport

[pine leaf]

no(

[thick zephyr]

What gpu do you have?

[pine leaf]

4070 super

[thick zephyr]

Have you used this .bat to install it?

Its on the same github repo on releases

[pine leaf]

problem with ModuleNotFoundError: No module named 'packaging' but im already install packaging

[pine leaf]

Gpu bat

[thick zephyr]

That´s so strange

[pine leaf]

new issue

but i have cuda

[scenic rune]

There are a million tedious things I want an AI to do in my DAW. Not a single one of them ever involves generating low quality audio samples. We have tons of fantastic audio samples floating around Splice already, and also quite advanced tools which can synthesize mathematically perfect waveforms for anything else conceivable. Why would we want worse audio samples from an AI? Why are developers not instead building AI enabled DAW plug-ins to help producers? Instead of this pointless fart-generator, I want a DAW co-pilot like what's available for coding developers. I want to prompt my AI-enabled DAW to do tedious things that cost me time and money. Why isn't AI doing tedious things for creative people, rather than doing creative things for tedious people?

Example prompts:
Phase-align this group of multi-mic tracks, take this mono track convert to mid/side and generate side audio content to make stereo, take these three existing tracks and build an 8-bar riser out of them ending at this timestamp, take this Amen drum break and chop it up along transient onsets, dynamically suppress resonant frequencies between 200-800hz on this piano track, generate three harmonies to this lead vocal using 4ths and 6ths, create 8 background vocal harmonies to support this chorus and phase-align them all, group everything but the vocals and kickdrum into a bus and enable sidechain compression using the kick as input, replace the transient on this kick drum 10 times and let me pick the best one, take this audio sample and use it to create an Andalusian cadence in D minor, create a reverse reverb fade-in for this lead vocal for 1/2 measure, and on and on and on....
As a music producer, this is what I want from an audio AI, not deriving low-quality samples from high quality samples.

also, hi.

[steel harness]

What you want is a large action model (LAM).

[still crypt]

I like you. Welcome.

AI is absolutely incapable of that level of music understanding due to a tremendous lack of training data. If you fine-tuned ChatGPT 4 in conjunction with Suno on musical (mostly just tedious processing) tasks such as those, you might be able to get close. But obviously that isn't a possibility rn.

Workstation-wise, it'd definitely have to be aax plugin format. One of my producer buddies uses gpt api to have is reaper instance do stuff based on voice command, which is sick af and saves him a lot of time, but even then, it isn't capable of listening to the session / knowing it to that degree.

I too want that kind of functionality from an AI. I also want an AI that cleans my bathroom and does my taxes. But those don't replace the workforce in bulk, so they won't be as profitable for the big companies. Probably. So in the meantime, I'll just keep making music the long way. I'd lightly argue the fun way.

[scenic rune]

Hard disagree, in fact one of these examples is already being worked on. The problem is developers are focused on the completely wrong issues when it comes to AI as a tool for music production. If producers were involved in the development of these tools there would be more than enough training data. There is a disconnect between Silicon Valley and the rest of the industry which caused this, and that needs to be fixed. So far generative AI has been a solution for problems which don’t exist. Stable Audio is basically the “Not Hot Dog” app from the TV show Silicon Valley, only less useful.

https://www.instagram.com/reel/C3-sMkDoCq5/

mrbillstunes

A little trick for widening a track using Combobulator by @datamindaudio.

Watch the full tutorial in the HCA Feed (if you aren't a Hardcore Abletoneer yet, subscribe today—link in bio).

#musicproduction #datamindaudio #combobulator #stereoimage abletonlive #ableton #abletontips #beatmaker #beats #electronicmusic #mrbill

Likes

890

Instagram

[hardy burrow]

Amen

[patent hemlock]

this channel on fire now

[still crypt]

Hard agree

I don't keep up with this stuff so I speak ill-informed

[shut saddle]

Not sure what I want but stitching songs together with outpainting seems like a fun idea.

[summer scroll]

what are you guys using as a ui?

[compact plume]

Hi , I would like to ask you if we buy a "Professional" license, how does it work if we cancel it after 5 months ? Will the music that was generated still be able to be in the project or will we have to delete it?

[novel frigate]

Don't speak for all creatives and music producers. I couldn't wait for a open source sample gen. I'm training currently on my own samples.

[scenic rune]

If your goal is to turn high quality samples into a trash generator, sure. Knock yourself out. My point is there are fantastic uses for AI audio, which the developers aren’t doing because they lack insight into the needs of most people in the industry.

[deft eagle]

👋 curious, you switched to DiT in the stable audio 2.0, just like in SD3, but still use latent diffusion, while SD3 started using flow matching (FM). Is there a particular reason to not use the potentially more efficient FM in the stable audio 2.0 and onwards ?

[plush glen]

stable audio is amazing for noise/experimental stuff, you can generate all kinds of crazy things, guess you think it's all garbage but hey theres more to samples than shiny commercial packs

[scenic rune]

Correct these days I need something that works quickly. I don't have time to roll a dookie log in glitter to see whether anyone can tell if it still smells or not

[plush glen]

come on man show us your duck, post your music 😁

[shut saddle]

I'm just using the default Gradio included in stable-audio-tools right now. Tech too new to have a lot of shiny toys at the moment.

[novel frigate]

If you can't flip a sample into something great, it's not a problem with the generator, it's more likely a skill issue. But if it's such a need for "most" industry people, get a dev, some vc money and make it happen.

[scenic rune]

It's not a matter of flipping a sample, the point is we already have a squillion ways to get good samples already without inferior AI generation. What we need is AI tools which are time-savers to most producers. Sample generation is not.

[shut saddle]

Yes it's a sample generator, but if the community eventually develops an "auto1111" tool or creates extensions for audacity I could see a sample generator like this become helpful especially from the audio2audio aspect.

[plush glen]

if you mean the AI should be doing all the hard work, hell yeah it should, but that's a vastly different problem than training a model for samples, you are talking about deep integration with DAW processes, VSTs, etc... not something we have today

[shut saddle]

Really early audio2audio example:

Like September 2022 early.

Organo audio source: https://youtu.be/ydUSm7BoMwk

[novel frigate]

You should stop acting as we. It's just your personal standpoint. AI Sample Gen is a big time saver. Instead of wasting hours searching for the perfect sample. I just describe it to the AI and get it.

[plush glen]

banger

[scenic rune]

But it IS something we have today. Check the combobulator link I posted above, it solves a very specific problem using AI quite effectively. This is an example where the tools are applied to a problem and not the other way around.

[plush glen]

post your music right now

[summer scroll]

i couldnt get stable audio tools installed. it kept giving me an isue with the packaging module

[tidal lava]

Made a visualizer for my track.
Images were generated using Stable Diffusion.
Overlay was made with Canva.
Visualizer built with After Effects.

https://youtu.be/esUZOunT2dw?feature=shared

[summer scroll]

anyone know how to get ComfyUI-StableAudioSampler running? i am getting the error: ModuleNotFoundError: No module named 'packaging'

i'll be your bestie for the restie

[pine leaf]

Same idk how to fix it

[summer scroll]

i tried a bunch of stuff last night its a pain

[tidal lava]

did you try navigating to the root directory of your local clone of the stable-audio repository and installing the packaging module?

You might try running pip install packaging in the root directory of your repo folder.

It's a dependency issue, that's all. You're missing a module that is required to run whatever it is you're doing.

You may have to trace the error back to a deeper problem if installing the module doesn't prove successful.

[summer scroll]

i installed it to the env site-pacakges

it looks like its a bug with flash-attn>=2.5.0

it might just be incompatible with 3.12

[tidal lava]

Gotcha.

I'm going to go through the process myself and check it out.

[summer scroll]

(.venv) PS P:\ComfyUI-ZLUDA> pip install ../GitRepos/flash-attention produces: ModuleNotFoundError: No module named 'packaging' [end of output] thats directly from github

[scenic rune]

pip install -r requirements.txt

You may need to upgrade pip to latest version

[summer scroll]

i have

[summer scroll]

"Might work for Windows starting v2.3.2 (we've seen a few positive reports) but Windows compilation still requires more testing. If you have ideas on how to set up prebuilt CUDA wheels for Windows, please reach out via Github issue."

[sweet swift]

i noticed lumina has a music model, anyone tried it?

[summer scroll]

comfui plugin?

[sweet swift]

theres no such thing

[summer scroll]

https://github.com/bdashore3/flash-attention/releases/tag/v2.5.9.post1

https://i.imgur.com/MzGbVkT.png

https://github.com/Stability-AI/stable-audio-tools
cd stable-audio-tools
pip install build
python -m build
https://i.imgur.com/qeo724S.png

[pine leaf]

how to install it

[pine leaf]

[summer scroll]

pip install <*.whl>

i didnt get it to work yet

if i do ill do a video on it

[toxic jackal]

You need pip some

[pine leaf]

already

but there's still this error

[thick zephyr]

@summer scroll have you tried with this one? https://github.com/Saganaki22/StableAudioWebUI

[wary sky]

yes i tried it its under trained for now

[toxic jackal]

@pine leafbro,I have already run Gradio, and I bet you haven't configured the environment yet. Try using pip install more. ..

[opaque wren]

has anyone had the flash-attn install stuck here?

im on python 3.8, cuda toolkit 12.4, and torch 2.2.2+cu121

[stuck hamlet]

how can i start stable audio open interface again, it opens automatically when downloading it for the first time but now I can't find the .bat file to open it 😦

[toxic jackal]

@opaque wrenpro,I successfully installed flash_attn.If you are a windows system like me, execute the instructions first.
(pip install torch==2.1.0 torchvision==0.16.0 torchaudio==2.1.0 --index-url https://download.pytorch.org/whl/cu121 )
And check the website (https://github.com/bdashore3/flash-attention/releases) to download the flash_attn version of the corresponding environment variable.

[opaque wren]

Uff thanks for that, ill give it a shot

[sweet swift]

oh

[wary sky]

but you can try the web demo they have to see for yourself. but i think stabel audio open is better

[sweet swift]

i cannot find it

[wary sky]

http://139.196.83.164:8000/

on the github are also links to ther other demos https://github.com/Alpha-VLLM/Lumina-T2X?tab=readme-ov-file

[sweet swift]

oh

[wary sky]

but i still like the lumina project. its open souce and they even plan to make a model that has multible modalatys. like it can make videos images audio and so on. all in one model

[sweet swift]

he image rendering arent that good but

are they the same ppl that made LUMA?

[wary sky]

no they are chinise. but some people that worked at stabilety made the luma video model

[sweet swift]

i see

lumina need to improve the rendering quality for now

[wary sky]

yes they say its still under development

they will also relese a text to speach model and other stuff

[sweet swift]

thats a big passion

[sweet swift]

16 color vae lumina t2i when

[wary sky]

the proplem with chaging to a diffrent vae is that you probably have to start traing the model from scratch. that could be to expensive or time consuming

[sweet swift]

sad

couldve been 16 color channel at the start, i would want to see how would they do with text in images like sd3

[summer scroll]

https://stackoverflow.com/questions/78604018/importerror-cannot-import-name-packaging-from-pkg-resources-when-trying-to pip install setuptools==69.5.1 https://i.imgur.com/BuDDE3T.png

im noded up

[harsh steeple]

those are rookie numbers :3

[polar nexus]

Why are stable audio servers always pending recently?？？

[raven stirrup]

I just realize why don't they call the model "Stable Audio Diffusion". That would be SAD.

[wary sky]

i asked and they said they are experementing with 16 channels now xD

[wary sky]

its called stable audio open. so its sao

[sweet swift]

YEAAH

im hoping they would fix the quality rendering, hands etc

id imagine what would it be with text generation in images

[wary sky]

yes but a lot of these models are undertrained and then finetuned on a small subset of astetic images. its not realy optimal

but it saves cost

[sweet swift]

damn

someone should pick that up rn

[vital ravine]

i just installed the webui for stable audio but for some reason it's using my CPU for generating. does anyone know how i can change it to my GPU? i've installed all the dependencies. also stable diffusion runs on my gpu just fine so i'm confused lol

[vernal rampart]

this is the best technology to tap out a beat and turn it into drums right ?

[toxic jackal]

You can enter (nvcc --version) at the command prompt to check whether you have installed cuda. Most ai programs are based on CUDA programming of NVIDIA graphics card.

1. Go to official website (https://developer.NVIDIA.com/cuda-toolkit-archive) to download and install CUDA and configure computer environment variables.
2. Go to official website (https://pytorch.org/get-started/previous-versions/) and install the torch corresponding to cuda with pip instruction. The reason why your program runs on the cpu is probably because you installed the torch that only supports the cpu version.
   I hope the above suggestions can help you.

[pine leaf]

still

[thick zephyr]

Yooo cool, do you have the prompt of it?

[fair delta]

is there a more active discord for stable audio

[prisma reef]

https://discord.gg/kPamgUnHm6

[shut saddle]

They already have joined.

[wary sky]

i am not sure something like this "reverp base bosted"

[coarse lantern]

What's the recommended text to music these days? Preferably also with vocalists actually singing :P

[shut saddle]

If you want vocalists, you probably aren't going to have luck here. You'll need an LLM model like Suno or Udio for that.

Vocals in SAO or Stable Audio 2.0 sound like The Sims talking

[coarse lantern]

I can work with a llm if there's a gradio of sorts that can combine both music and vocal :P

[wary sky]

i think you dont need a llm model for that. its just not implemented in non llm models

[wary sky]

ther is no open source llm that can do that currently. the closest thing we got is bark. but we dont have traing code and it cant realy do music rn.

[coarse lantern]

Hmm indeed.

So for these, what git's will be needed for these? As sony's for instance doesn't have a config.json, so it can't be loaded into diontimmer's audio diffusion

[wary sky]

use stable audio its better then lumina for now

[coarse lantern]

I can't even use lumina as there's no model config json included, and stable audio i mentioned can't load it

[wary sky]

i dont think that lumina can loade with stable audios code

lumina is a complyltyl indeependent model with its own code

[coarse lantern]

Gotcha. How do I find out what gradio/git one uses for each model on hugging? For instance for lumina's, same with Sony's.

Also, for stable audio webui one, how do I make it only generate 5 or 10 actual seconds and not imaginary ones? As less samples gives less duration, but also utter ear blasting nonsense

[wary sky]

you need to fine the github of the models. with the code. and they often deskibe how you intall them

but the lumina has a web demo also so you dont need to intall it

here is the lumina web demo http://139.196.83.164:8000/

and ther github repo https://github.com/Alpha-VLLM/Lumina-T2X/tree/main?tab=readme-ov-file

[coarse lantern]

Aye, i know how to install the gits that provides the how's. What code do you speak of that i can trace to a git?

And i prefer to load them locally :)

[wary sky]

maby u shuld use the pinokio installer i have hered that its easy to use.

i only istalled stable auido open so far

[coarse lantern]

I just asked what code you were speaking of. Once i understand that part, then i can start looking/waiting for gits with said "code" for a git that corresponds to the model i want to try out

[wary sky]

thats the code for stable audio open diffuion https://github.com/Stability-AI/stable-audio-tools

that has to be installed and then you need a folder where you put the model and the model config

[coarse lantern]

Code? You mean link?

[wary sky]

and i have not installed a lumina model jet so you have to look into ther github and see how to install https://github.com/Alpha-VLLM/Lumina-T2X/tree/main?tab=readme-ov-file

[wary sky]

the program

[coarse lantern]

Ah, now i get you Sorry for the mass confusion from my end lol

[wary sky]

xD

[plush glen]

kinda weird how stable audio seems a little under the radar compared to image generation models

[summer scroll]

been trying to get my img2vid, stable audio, sd3, and sd-3d working.. almost there

[wary sky]

yes

i guess you cant make corn with it lol

[wary sky]

asmr

[foggy turtle]

It would explode if we could not just make audio, but songs like Suno locally.

[wary sky]

yes i think the model architecture could do it. just traing data is the problem

[patent hemlock]

All these other channels feel so relaxing and serene compared to the raging fire that is the SD3 channel hides

[wary sky]

xD

[summer scroll]

look... im not going to take stability serious until there is the shitty flute lora for audio models: https://www.youtube.com/watch?v=nF7lv1gfP1Q

[mellow torrent]

been trying to instlal for 30 minutes 😭 with pip install stable-audio-tools --no-cache-dir on my MBP

❯ pip list
Package Version
------- -------
pip     24.0

[summer scroll]

its not simple

[summer scroll]

for that error, pip install setuptools<70

[mellow torrent]

how do u even figure this stuff out

I am from Rust land so all this stuff is like IQ 9999 for me

[summer scroll]

before there was ai there was me

[mystic obsidian]

Hey guys, we are interested in a Stable Audio Enterprise License, who should I talk to?

[still crypt]

https://stability.ai/enterprise

[limpid pendant]

https://arxiv.org/abs/2406.08761

[wary sky]

I would just crete a python envirment git clone it and install requierments.txt

[signal sage]

StableAudio constantly times out or reports errors from AWS Sagemaker. Is there any support channel at Stability AI that can help with that? My emails and support tickets have not received any responses so far.

[wary sky]

Stability cant pay amazone bills probably or its something else

But they have a open source audio model now

[signal sage]

that's the old model and it doesn't support the mps backend for metal acceleration 😦

[wary sky]

its relativly new acually

what is mps?

[signal sage]

Metal Performance Shaders - it's the acceleration framework for the new Apple CPUs. You set mps instead of cuda as your torch device. Unfortunately, the stable-audio-tools have cuda hardcoded all over the place.

[wary sky]

daim

[tidal lava]

guys anyone experiencing pending issue on generations?

my other two has been timed out and the third one is... still pending for like 3-4 days

[signal sage]

It doesn't work at the moment. I have been getting 1 successful generation out of 30 - 40 attempts.

[plush glen]

I would like to hire a dev to make a nice multi platform client for stable audio. If you are the right person please DM me

[wary sky]

https://www.reuters.com/technology/artificial-intelligence/music-labels-sue-ai-companies-suno-udio-us-copyright-infringement-2024-06-24/

[floral pollen]

when running local, is there a way to use its api, at /generate or whatever?

[primal merlin]

same question

[still crypt]

^the music that plays when you have three friends in common with comfy

[shut saddle]

https://tenor.com/view/cat-head-shaking-head-shaking-cat-gif-20123810

[slim briar]

I'm running into walls trying to resume training on a Stable Audio Open model I made... Neither the wrapped or unwrapped models will resume, I'm getting a few variations of the same error every time:

https://github.com/Stability-AI/stable-audio-tools/issues/114

"The size of tensor a (14) must match the size of tensor b (12) at non-singleton dimension 0"

I only changed the LR, warmup and amount and content of sample sounds in the model config, I'm not sure what's going on here.

Hmm, actually I now realized what this was: changing the sample prompts caused this error, which is not how this would work I think...

Anyway, been getting superb results already by finetuning on my own music, as the base model sounds kind of neutral and boring.

Some sounds with the base model.

The same after 1200 steps with batch size 38, training on 158 tracks made by me, with the default sample length and random crop on.

Loss went down really slow, I mean, that's already almost 50K updates, and I only got down to around 0.64 or something, which is kind of high for a diffusion model. On the other hand, the outcomes dictate the usefulness, not the analytics... And I can already see this as a very cool tool. Next I think I'll train on my stems, but they need a lot of work removing the silence, I'll need to find some batch tool, not sure if Audacity is up for that.

A clip from training, same model, I can definitely hear my influence.

[wary sky]

maby the people in the harmoniai discord can help you

[slim briar]

Oh right, that was the Discord for the audio model creators. I reported the weird stuff on Github already.

[sick laurel]

is anyone else having issues generating with stable audio when imputing audio file, I thought I try audio to audio, a simple 20 sec long drum beat, but I get stuck on pending forever and eventually times out trying to generate, I tried some other clips and still not working , I even bought pro to maybe not get stuck in a queue for generate, but still have not gotten anything to generate using audio as input, can generate text to audio, so clearly servers are up

[granite acorn]

For Stable Audio, is the upload limit a one time thing? What if I'm uploading the same track because I want to edit multiple times, or is it upload once - So I don't have to upload again if it's the same

[light shoal]

I was having trouble when sending it audio in FP 32 wav format. The same file rendered to an int 24 format seems to work ok.

[sick laurel]

@light shoal ok i might try it got a a few generation working but sometimes still fails.

[light shoal]

TBH, I've now got a couple that are just sat there spinning based on an int24 file, so I think I was premature thinking I'd solved it

Of my last 10 generations, only 3 have worked with the remaining 7 either timed out or looking like they are going to. I've put a support ticket in and I'll see what they say.

[tidal lava]

Hello. everyone.
Is this discussing room?

Could I make a question?

[tidal lava]

???

[tidal lava]

???

[patent hemlock]

Go for it Kan.

[raw zealot]

hey y'all i just started getting into stable audio last week and while doing my research i got a bit confused. i keep seeing people ask about running it locally, but isn't that what stable-audio-tools lets you do?

i wrote a script using that and have no issues generating audio on my laptop (6gb vram). the one thing that i find mildly annoying is i can't seem to figure out how to make a clip that is under 47 seconds. like i can make one that is 20 seconds, and then there will be 27 seconds of silence. i'm guessing that is just how the model is trained, and the best results would be to leave it at 47 seconds?

[limpid salmon]

you can also run it on their website where they offer a newer 2.0, ive only poked at it a tiny bit but using ComfyUI where i can set latent length, peeking at their repo its this sample_size: int = 2097152,

[mystic geode]

Every time I try to use the 2.0 model I get these errors- does anyone know a way to fix?

[mystic geode]

This error only occurs when I try to use an input track

[mystic geode]

did you ever get this fixed?

[mystic geode]

did you ever get this fixed?

[signal sage]

No. It’s consistently broken.

[mystic geode]

have you managed to get in touch with support? I just messaged them twice this morning but not sure if I should actually expect a response lol

[signal sage]

Never heard back from them.

[sick laurel]

@mystic geode tried a few days later and it worked but it still failed like every now and again , and then the failed ones got stuck pending for ever until it said I could not make any more request because had to many pending, so gave up on it and the results that I did get , was not really good enough for what I wanted to use it for anyways.

[mystic geode]

Is it possible to run stable audio 2.0 locally? Or only version 1.0 ?

[shut saddle]

You can't run stable audio 2.0 locally only because the weights are not available. You can run stable audio open locally though. Different weights.

[mystic geode]

are you aware of any other ways to use stable audio 2 other than the stableaudio website? soopa frustrating not being able to reliably use the audio2audio feature, haven’t been able to get it to work for days

[shut saddle]

Not that I am aware of. Stable Audio 2 is through the website only as far as I know.

If you want more flexibility, Stable Audio Open is best for running locally.

[stone smelt]

anyone know why the result player is grayed out? its generating the audio flac

comfy up to date

[raw zealot]

this is probably useless for most ppl but i made a colab notebook for the open source model. i've never made a notebook before so it might have a lot of bugs. i've been using it to generate a big sample library. https://colab.research.google.com/github/xxristoskk/stable-audio-sample-generator/blob/main/Stable_Audio_Sample_Generator.ipynb

[vital ravine]

does anyone have a comfy workflow for stable audio? I'm trying to figure it out on my own but Comfy crashes when it tries to save the generated audio

[stone smelt]

just drop a flac file from this channel into comfy and it should load the workflow unless the creator removed it

[vital ravine]

Thanks!

[vital ravine]

Alright now I'm generating a 10 second audio clip @ 50 steps, getting 15it/s but it's taking over 10 minutes lol. How long does it normally take to generate a clip?

[stone smelt]

the full 47 seconds takes about 2-3 seconds on a 4090

[raw zealot]

on a 6gb 3050 it takes about 30-45 seconds

[wind granite]

What's the sota for voice cloning?

[lavish shore]

hi everyone, I am new to SA and recently installed. Can someone please point me to a tutorial for getting started creating a custom model?

[wicked heath]

Good app for this sorta thing: https://github.com/Nerogar/OneTrainer

Check their discord for help (linked on the repo)

[storm holly]

A clip from training, same model, I can definitely hear my influence

[lavish shore]

thank you!

Has anyone installed Stable Audio Open on a server and found a way to run parallel generations?

[forest swallow]

We have implemented stable audio on Replicate here : https://replicate.com/stackadoc/stable-audio-open-1.0
And the cog source is here : https://github.com/stackadoc/cog-stable-audio
If you need any help to implement the solution on your own backend, feel free to ask 🙂

[onyx fiber]

sorry I don't know how things work here. I just made this beat and love to share. I am new here https://www.stableaudio.com/1/share/dba9c3a6-14ab-4ec9-96cd-1dc333065ed2

[lavish shore]

Does Stable Audio Open support generations in key?

[forest swallow]

With the base stable-audio-open-1.0, no, since the key is not present in the training dataset. But you may be able to support it if you fine-tune the model with the key in the training prompts. For the stable-audio 2.0, I don't know

[wanton ocean]

is there any ai music/song generator that can generate a project file for daw software like fl studio? or a way to generate a project file for daw from an audio track? i'd like to make edits to my ai music generations

[wanton sundial]

https://klang.io/blog/mp3-to-midi-daw/ This might not be what you want, but it looks like it converts your suno (or whatever) mp3 to midi to use in your daw project…

[lavish shore]

is there any guide on the UI options for SA explaining what things like cfg_scale and #steps affect?

[forest swallow]

https://huggingface.co/docs/diffusers/main/en/api/pipelines/stable_audio
See "guidance_scale" & "num_inference_steps"

[lavish shore]

Thank you! Looking at the inference steps I understand the relationship between steps and quality but not quite sure what slower inference translates to? Does that just mean it takes longer?

[forest swallow]

Yes, it increase proportionally

[patent hemlock]

How come that new 3D generator thingie doesn't get a channel? 😦

[flat shore]

😂

[west hawk]

hey, do we know when the api will be available? or if it will be available at all?

[crimson pollen]

I posted a new song called "Sands of Time" (Sands of Time), a metallic metal song with Arabic melodies without vocals (Instrumental) 🎸🌟 which was previously produced with Stable Audio artificial intelligence 🤖🎧. Those who follow me may know it, but it is now available in an optimised version in high quality on SoundCloud! 🎵🚀

Song link: Sands of Time 🎶🔗:
https://soundcloud.com/tber-mohammed-mehdi/sands-of-time-ft-stable-audio

Please sign up for the site and support it with a heart button ❤️ - it really helps spread the word ✨

#Metal_Arabic 🎸 #Music_without_Vocals 🎶 #Sands_of_Time ⏳ #AI 🤖 #High_Quality 🎧 #SoundCloud 📻 #Mohammed_Mahdi_Altabar 🎤 #Stable_Audio

[proven scaffold]

Hi,
Is there an API available for StableAudio that can be used for custom projects, or is it only accessible through Hugging Face/stable_audio_tools?

[grim harness]

https://youtu.be/W53LvOq_dBM?si=v7me7oHnWDM5C5Il

[cosmic lantern]

Hello, You all seem to you what you doing when using this AI music gen. I just started today. Usually I write and play my own music, but due to neighbors not like drums, guitars and the like, and not having access to a studio, I'm trying this.

I trying to put together, a track, for my youtube channel, it's a gaming channel. An intro song for my vids, with my vocals and lyrics, Is there someone willing to take some time to help out and show me how I would accomplish this...

using this stable audio system..

[hollow ravine]

How does one actually get in touch with StabilityAI? I have tickets in 'pending' now for over a month. I've emailed, I've updated my tickets with details / screenshots of the CONSTANT errors that I'm encountering (error 400 sagemaker) and when trying to use my own audio, it has failed (timed out) EVERY TIME that I've tried over the last couple of weeks. Things seem to be getting worse and worse.

Are there any 'official' StabilityAI / StableAudio mods / support people on this discord server ? Do any of you have ANY information on how to mitigate/resolve these issues using the 2.0 web interface? I'm so frustrated - not able to use what I'm paying for - and unable to reach anyone to find information about these issues. Any advice would be VERY APPRECIATED. Thank you!

[hollow ravine]

A month later - still a constant problem and absolute silence from 'support' --- it's maddening as we're unable to actually use the service that we're paying for. Such a shame as StableAudio2.0 is so useful when it works. Please let me know if you ever hear back from them or find any workarounds to the perpetual 'Timed Out' (+filled queue that appears to be limited to 3, not 5 like the error states) --- or if you've any information on the Sagemaker Error 400 issues that come up with audio2audio prompting. I'll be sure to report back if I ever hear/learn anything.

[signal sage]

I think that Stability AI is barely surviving as a business. This may be a team that is no longer staffed.

[patent hemlock]

No one there. the models went self aware and will train themselves and lament until the electricity is shut off.

[waxen swift]

https://youtu.be/UhIWnr8vWDg?si=FEWGKdU5O9x5HMR4

Are these voice generated in Eleven labs?

[plucky trail]

Hi there, i'm trying to get Stable Audio Open to work locally with 0 success so far. Since there is about 0 information about it anywhere, I'm not even sure it can make something else than music at this point.
Is it supposed to be possible to run it on a 3060 TI GPU, i'm using Pinokyo for testing but so far cannot make anything, it just run forever.
I have no idea what generation time i'm supposed to hope for or how to prompt or absolutely anything really.
Any idea ? Thanx

[inland loom]

You have to provide your background logs so that others can help you.

[plucky trail]

Do I even have access to it on Pinokyo ? Also there is no error message, it's just running forever, but since I never made a gen and there are no info nowhere I have no idea what to expect for speed or parameters

[inland loom]

你是否装了cuda？

Sorry! I accidentally spoke Chinese. Did you install the matching cuda to pt?

[plucky trail]

Supposedly pinokyo is supposed to do that automatically, did work fine with forge for flux/sdxl

[inland loom]

If you have multiple pythons on your computer, you may get an error when performing this operation. Enter your terminal to check whether the environment you are adapting to is compatible. If there is no problem, try changing to a smaller model or smaller parameter requirements. It may be that the GPU provided by 3070 is not enough.

If it is the first startup, it will automatically install the missing startup environment and check whether the network is normal.

[plucky trail]

Yeah that's why I was asking, since I had no idea what to expect, like how long would it take normally to make let's say 5 second on 30 steps or something. I think any bad python release would just throw an error, since GPU is running full during the generation

[inland loom]

The 4090 I use is fully loaded with GPU. If you want to make high-quality works quickly with low video memory, you can only manually modify the core algorithm and make a set of loops that are more suitable for your device.

You can use Visual Studio to write new algorithms you need.

[fading matrix]

he 4090 I use is fully loaded with GPU. If you want to make high-quality works quickly with low video memory, you can only manually modify the core algorithm and make a set of loops that are more suitable for your device.
You can use Visual Studio to write new algorithms you need.

[deep idol]

which one do i download using AMD

[lean steppe]

Hello! I consider your audio service as one of the best in the game. 🙂 Therefore, I have a short question. Can I use the music I generate as a main content of my YouTube video? E.g. with static background or short animation for a video with music to study with? (something similar to lofi girl)

[restive coyote]

Like this music!🎶

[vocal helm]

Hello! Just wanted to share this with the community, if any of you are in or near Barcelona on October 4th you can't miss this IRL get together!!!

https://www.eventbrite.com/e/entradas-barcelona-ai-music-meetup-v10-1006681229657?aff=oddtdtcreator

[empty basin]

anyone else getting issues using Input Audio on https://www.stableaudio.com/generate ?
if i generate with an audio input, the generation is permanently stuck on pending and i can't generate new songs

[empty basin]

now shows
error - ClientError: Received client error (400) from model. See the SageMaker Endpoint logs in your account for more information.

[lean steppe]

I have exactly the same issue and I reported it a few days ago.

Last time I got the answer after one month. Let's see.

[steep sorrel]

Hey there
Hope you are doing well
I am a senior full stack developer who has full experience in web and AI development
So if you have some projects, please let me know
thank you
https://github.com/typhon0130
https://figma.com/@typhon0130

[tulip orbit]

/subscribe

[steep star]

/subscribe

[empty basin]

now getting Error: Timed Out
@willow roost are there any devs that can take a look?

[lean steppe]

I also still have this issue

[vocal helm]

Hey everyone! My album made with AI tools (lots of stable audio in there) is finally out!!!!

https://open.spotify.com/intl-es/album/70V3ApjtDqTmXyKBFN2PVM?si=H83jLIl1SyOxldetuvMGKA

I really would love any feedback or thoughts or questions or whatever!!

[near pond]

Why can't I make an audio-to-audio connection? Whether it's an uploaded file or an uploaded record, the generated composition remains "pending" for a long time, before finally displaying an error message.

[crimson pollen]

https://soundcloud.com/tber-mohammed-mehdi/sands-of-time-ft-stable-audio?si=d039888645214b24a6d4c239353f2af9&utm_source=clipboard&utm_medium=text&utm_campaign=social_sharing

[crimson pollen]

Never seen such long solo in an Ai service other stable audio. can't believe what I generated

[full sundial]

Hi, somebody know sources for custom audio safetensors? I found only the offcial 1.0 model.

[mint quest]

kinda neat. what was the production process in terms of editing? I assume this wasn't generated start to finish in a single click

this is what an unedited single click gen from my model sounds like:

[crimson pollen]

No I generated it from one take and the version in SoundCloud is just remastered

[stiff ferry]

Hi!
I want to generate one-shot samples for music making. It's important to get precise notes, like C3. How to do?

[ruby wraith]

i have two gpu's one with vram 12 gb on with 16 gb. is there any possibility to run stable diffusion video using these two. its would be a great help. i am new learner .

[half hill]

Check msg @ruby wraith

[wide breach]

https://open.spotify.com/intl-pt/album/3zJcmthj0np6kgS5Rk41sL

[crystal blaze]

Line-break appears to be ignored in the prompt field. Would suggest interpreting line-break as whitespace.
I.e.

first term
second term

seems to result in first termsecond term

[oak talon]

https://youtu.be/UFH5W82Bb_I?si=PTDDWQMJDU2Lm5KM

[coarse lantern]

Did the stable audio diffusion just get abandoned? As the audio tools hasn't been updated for half a year, and haven't found any new programs that can do the same/better with the models

Can't even get it to work, and can't find the darn requirements.txt as my setup is just all broken for it.

[candid edge]

I released my debut single last thursday :)) i used ai to generate the bulgarian choir effect throughout the song. Let know what you think :))

https://tr.ee/o4fRq4KMvR

[gentle zodiac]

Even stableaudio.com is busted.

[thorn roost]

Is there an api for stable audio?

[fallow loom]

Is there any french people here or ... ?😌

[vocal sparrow]

anyone know what ai was used to make these music extensions?
https://www.youtube.com/watch?v=CTdMM2EtEOM

[crimson pollen]

Listen to Sands of Time (ft. Stable Audio) by Mohammed Mehdi TBER on #SoundCloud
https://on.soundcloud.com/MSGuhxTaybbg7shw9

Thanks for the support guys and listens

[woven crypt]

Its a old scam. Probably his account was stolen

[short raptor]

Hi everyone, does somebody know the meaning of this thing mentioned on the subscription page?
Monthly upload amount
30 minutes
Cropped at 3 minutes

[brazen tusk]

you get 30mins of audio/video at 3mins at a time

[short raptor]

is this the limit of download of the audio we generate?
I mean I just got the pro subscription and I can generate 500 audio but download 30 mins?

[viral quartz]

you can upload a total of 30 minutes to stable audio, to use it as a source for prompting

[vale shadow]

https://www.youtube.com/watch?v=6hd-FOFYsuQ

[sharp wedge]

Aight, new here. Been doing image gen with Auto111's SD for a bit.
Any faq or noob-friendly guide I can follow to set up a local audio gen?

[cursive pilot]

The most current model might be YuEGP on their github you can see the installation guide. The git lfs install might be the only unusual step the rest is straight forward. The gradio app works great.

[austere wasp]

An astronaut on a white horse on Mars

[final salmon]

Anyone else unable to download tracks atm? It only lets u download most recent. Cant download any other tracks. Please fix

Just gets stuck on download circle with .wav

[vale shadow]

https://www.youtube.com/watch?v=_8NNIjvz7sM

[wooden quiver]

https://youtu.be/b3YBYZ4sig8?si=fqVBo66b9oiOZQLB

[gusty hamlet]

https://www.youtube.com/watch?v=QQU6TX78b8s

[primal beacon]

any good front ends? I'd like to experiment

[gusty hamlet]

I'm mostly familiar with Suno

[winter agate]

wait

sd has audio model？

Is it?

[cobalt surge]

yup

https://stability.ai/news/stable-audio-2-0

[winter agate]

thanks

[primal beacon]

so, any viable local front ends?

[tender island]

There's at least two different Github repos containing nodes for ComfyUI & Stable Audio

[hot shoal]

/linkwallet

[tardy lynx]

Has anyone been able to get an API for stable audio?

[shut saddle]

@low fjord can we get any info on the future for stable audio? Last update was 9 months ago when stable audio open released and I was wondering if stability.ai have moved on.

[low fjord]

@tight anchor anything you want to share?

or more accurately are able to share?

[shut saddle]

Didn't mean to put you on the spot Fauno sorry! I was just curious.

[low fjord]

suffice to say, yes SA is still being worked on, I have a sync with Fauno later this week and I'll find out what we're allowed to talk about 👍

[shut saddle]

I'm one of the mods for the Harmonai (stable audio) discord and people are talking about other open source audio model releases lately, so I was wondering what S.ai has been up to.

[low fjord]

oh cool, that's a fun server

[tight anchor]

We've got a good model improvements ready to go, as always our biggest blocker is an improved dataset

[shut saddle]

Do you think audiosparx would be willing to negotiate an update their terms to allow for a public release or is that likely a dead end?

[tight anchor]

That’s one of many conversations we’re having

[knotty finch]

Hey, I am interested in an AI model or a nueral network that shouldn't be modulating over the note duration. Shouldn't sound like a melody. This there any?

[ebon geyser]

It's almost a year since Stable-Audio was announced, is there any date for the release of the API?

[mint yew]

does anyone know how that could be done ?
https://www.youtube.com/watch?v=8Np2JQjz5WE

[patent hemlock]

https://tenor.com/view/palla-deserto-desert-hot-gif-6014273

[austere lodge]

create santa claude

[full burrow]

Hi!

I'm working on a binary classifier to detect whether music is AI or human.

Do you guys happen to have a folder of Sparx 2.0 audio samples that I could train my classifier on? A few hundred would be super helpful

[full burrow]

following up on this curious whether anyone has any tips

[oak igloo]

Can we run this locally?

[crimson pollen]

Stable Audio 2 just got released
must try it

I once made this using Stable Audio 1.5:
https://soundcloud.com/tber-mohammed-mehdi/sands-of-time-ft-stable-audio

Which was epic enough and impossible to recreate using another AI.

Btw, Stable Audio spectral quality is better than Udio but It doesn't support lyrics and a lot of things like Udio.

[crimson pollen]

Something is wrong with 2 in terms of notes, they are not working together as before. I remixed the same song and look at this.

[loud robin]

Not sure if this is the place but is there recommendations for removing the music of a video while keeping the rest of the audio, like talking, footsteps, etc?

[viral quartz]

when is the Stable Audio page gonna be improved/workable? beyond just a simple generation thing

[stray vessel]

I love the Stable Audio UI, can we get some of that UI for Image generation

[robust pivot]

Hi, I don't know about something like SD for that, but I know you can work it that with something as Izotope RX, or look for similar software. I think it is ai trained for the task.

But at the same time I would like to know if there is some stable app for working on audio with AI, like edit or mastering, stem separation, etc

Sometime ago I tried SD audio and I was able to generate but I couldn't do anything else, and in some gradio app there was things like mastering

[high briar]

@loud robin UVR UVR5 yes there is. Oh maybe not depends on filters i guess. I used it to lift the voice only. once. Not quite the same.

[robust pivot]

Oh I didn't know that one, great

[misty mountain]

Hey fam! I am uploading original compositions and the system is flagging them. These compositions range from unreleased music that I have created, to music I have released (both originals, and remixes for which I have compilation rights. \I;d like the system to key off of my work -- why is this not functioning as I expected? Thanks in advance!

[vale shadow]

https://www.youtube.com/watch?v=D6kqVYlBmn8&ab_channel=OvrllBgzy

[nova wasp]

Does anyone happen to know: if I have an AI voice model, is there a way to use it as a VST, or use it in the same way as one, to convert spoken audio in real-time?

[queen mantle]

How’s your health?

[wheat plank]

Vocaloid is a VST / VSTi that may work for this. If used as an FX plugin/Vocoder.

[dapper crypt]

Does anyone know a really good text to speech voice AI? like a storytelling voice

[hasty fox]

hello

[opal umbra]

Yo! Trying to find out if Riffusion straight up stole a song! This sounds wildly familiar to me! Is this a released song and does anyone know the name of it?

[coarse lantern]

Download shazam and soundhound, feed it to both, and see if it's recognized by any of them

Also, fun fact, all A.I models are trained on existing media lol. Nothing A.I makes is remotely original

---

Do you guys know of a voice model processor akin to stable audio tools, but a docker like whisper/piper, so i can make my home assistant make different non-verbal noises? As i got a voice model i use for my HA, but i need it to make a sneeze for instance using that voice model

[brisk wind]

to me, ai just mix existing style to make new style

[brisk wind]

you mean tts

https://youtu.be/TCHXzX6vUcA that's why I feel like my song is a copy of some songs that I did not know

[turbid wing]

i'm currently trying to use stable audio-to-audio (api) to generate some samples that can be paired with the original audio.
So far, the results haven’t been great, they don’t really capture the style or feel of the reference clip.
Has anyone had better luck with this? I'd love to hear any tips, prompt structures, or API configurations that helped improve your outputs. Appreciate any help!

[swift thicket]

Hello, I have a problem!
I have lost my wav download in stable audio.
Who can tell me how to solve this?

[queen dagger]

What do you mean you "lost" it?

[shut saddle]

Some important questions: when you said you were using stable audio are you talking that you are running it locally or are you referring to the website?

[pliant vapor]

does anyone have a working space or guide to run the audio-to-audio style transfer showed in this demo ?

https://arc-text2audio.github.io/web/?utm_source=catalyzex.com

[dim burrow]

theres an apk for stable audio small on android? why they promote the app for arm devices and dont release an apk?

[manic epoch]

Here’s some 👂🍬 for you 🌕🦉

WRG🐰Neuro🌶️Spicy🌶️Sampler🥟:
14 of the hottest unreleased tracks from every genre.
https://soundcloud.com/whiterabbitgeometry/sets/a-kid

[brisk wind]

My song is now up https://youtu.be/0icOFbnn32U https://rumble.com/v6tkhv1--secret-lilies.html

[vale shadow]

https://www.youtube.com/watch?v=pM9RTGPaFzo

[winter berry]

https://youtu.be/_dkLhhwh6As

[vale shadow]

https://www.youtube.com/watch?v=INSUydt4Pcc

[swift thicket]

I cannot download wave files for songs generated since the end of April. mp3 and video files can be downloaded.
You can download mp3 and video files.

[swift thicket]

I cannot download wave files for songs generated since the end of April. mp3 and video files can be downloaded.
You can download mp3 and video files.

[somber cradle]

Is there a way to fix this error in Stable audio???

It happens when I put non-original music on the site.

This didn't happen before, it started today

[somber cradle]

Hello??

[cobalt surge]

invalid or unsupported file type seems explicit, you probably changed the input format.

[somber cradle]

I'm using YouTube music from games

This error was not happening it started these days

It is not Copyright because the site would warn about it