How To Install Audiocraft Locally - Meta's FREE And Open AI Music Gen
Вставка
- Опубліковано 7 сер 2024
- In this video, we'll learn how to install Audiocraft, the new open-source and free AI music and sound effect generator from Meta AI. I go over both Musicgen and Audiogen.
Enjoy!
Join My Newsletter for Regular AI Updates 👇🏼
www.matthewberman.com
Need AI Consulting? ✅
forwardfuture.ai/
Rent a GPU (MassedCompute) 🚀
bit.ly/matthew-berman-youtube
USE CODE "MatthewBerman" for 50% discount
My Links 🔗
👉🏻 Subscribe: / @matthew_berman
👉🏻 Twitter: / matthewberman
👉🏻 Discord: / discord
👉🏻 Patreon: / matthewberman
Media/Sponsorship Inquiries 📈
bit.ly/44TC45V
Chapters:
0:00 - Intro
0:11 - Musicgen Installation
3:13 - Audiogen Installation
Links:
Install Instructions: gist.github.com/mberman84/afd...
Audiocraft Github - github.com/facebookresearch/a...
Audiogen PR - github.com/facebookresearch/a...
Audiocraft Announcement - about. news/2023/08/aud... - Наука та технологія
Awesome man. Thank you. I'm trying to learn this stuff and these step by step tutorials help let me know what's possible.
Ask gpt to write a very long descriptive prompt (150 words and +) and use the large model. Results is absolutly breathtaking. If your music sample is long, ask gpt to add description for the overall timeline as well.
This is debatable.
@@cristian15154 I'll have to agree haha. Yesterday was giving excellent résults, but it was the opposite with my series of test tonight. It tend to get stuck on a note for a while... You have the same problem ?
can you share the prompt that generated a great output? @@ultimategolfarchives4746
Great video as always!
Wow... This seems really interesting. When you say "any" audio sound effect, I would have really liked a few more examples!
This could have a reverberating effect in the soundFX industry...
I've been dying for an audio gen that does >30 second clips. Can't wait to install this, thanks for the update and another great tutorial.
You got it! This one does up to 2 minutes.
audio webui does longer than 2 minutes
@@matthew_berman audio webui does unlimited time
Audiogen+ have been out for weeks now and can generate 5 minute audio clips.
@@digidope Is this still recommended? I'm looking for one that will generate music from files as references.
Absolutely love your videos very easy to follow. I’m curious how you setup your terminal
Astonishing Matthew, as always 🚀
Thank you!
Thank you for helping us with the mouse cursor. I always have a hard time following a regular cursor.
to use gpu, 'conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia' // after install pytorch with cuda and then 'python -m pip install -r requirements.txt'
Very cool!
Hi Sir, i've got the following error "RuntimeError: ffmpeg not found." at the end of the 500 steps (even if i had installed the ffmpeg) Do you have any idea?thanks
God damn! Thanks for the instructions how to install it, really useful
You're welcome!
this is great - I removed the --share part of the python app command as it looked like it was creating a public url on my machine 🙂
Cool. GJ!
Great video. It took me a while to get this all working with musicGen, I had to manually install a few of the requirements and try different versions of Nvidia Cuda. Ultimately got it working, and it's really fun to play with. Do you know if anyone has cracked the 2 minute max? Also, what's with the public URL that spins up when we start the local instance? Thanks again for the great vid.
thank you so much for the guide! I have to admit the results are pretty much useless, at least with 5-60 seconds of generation. I had completely inconsistent results, that had close to none connection to the original prompt.
Hi Matt!
So that we can install with No hiccups?
Can you share your Mac model? memory & hard drive space?
Lastly, so we can block it out...how long did it take for everthing including downloading the Models?
Thanks in advance! For doing so!
Hey, great video man! Do you think you could develop the tool, and give the possibility to enter a number of file to upload more than 1. That way for example, we could upload 10 songs, and let the AI creating a new song based on these 10. Would be pretty amazing I think!
Install failed, I got about 1/2 way through then it says .... "git is not recognized as an internal or external command"
Got it to work one time, and the for some reason when I try again it fails. Tried reinstalling from scratch and keep getting the error: " ERROR: Failed building wheel for xformers" went online: others experiencing the same problem and so far, can't find an answer. Any clues?
Bro you vids are so goood Plz make one about Gorilla!
Noted! It's on my list.
judging from the thumbnail, this looks priceless! ears should be really clean!!
Mine is also generation on the CPU only. I would imagine this is something that the GPU would do better (although it might be too big for vram) but I've no idea if/how you'd instruct it to use the GPU. It works, though.
I knew this would come but my mind is blown away. Sad thing is that it's harder to describe a music than an image.
Hello! I'm new to installing AI environments on my computer and not super well versed in working in the terminal in general. When I open up terminal it just has "my username@the computer ~ %" and then the solid box to type in a command. I see the one here that says "(base) -> ai projects." Do I need that? And if so, how do I get it? Am I even in the right application? lol Also, when I typed in "conda" I got a message that says "command not found: conda." So I downloaded Anaconda as opposed to Miniconda, just to be sure, reran the command and got the same message. I would appreciate any help I can get. I'm guessing I'm way over my head, and so if trying to quickly explain what I'm missing is too much, I completely understand. But even if you could just point me in the right direction I would sincerely appreciate it. Thanks!
Any idea I got this error: "PermissionError: [WinError 32] The process cannot access the file because it is being used by another process"??
Amazing Matthew; If I could, I would subscribe again :)
Thank you :)
What's the configuration of your computer?
Thank you for your video! Is it possible to create an Acapella with this AI, I am looking for a model that allows me to create a similar Acapella, of music that already has a singer
The default model directory is to C disk, is there a way to change it? 🤔
tried to run audiogen after installing musicgen and changing branches, but this happened. AttributeError: module 'gradio' has no attribute 'Progress'
Thanks Matthew! I got it working but ran into one issue (and solution) on my setup (M1 MacStudio 128GB). Following along exactly my installation errored when installing Xformers. The solution for me was to delete Xformers from the requirements.txt file then try installing again. It finished the install perfectly. Then when I ran "python -m demos.musicgen_app --share" it failed to load saying it was missing Xformers. Then I put Xformers back in the requirements.txt and ran "python -m pip install -r requirements.txt" again. This time Xformers installed correctly and now it's all working.
(audiocraft) F:\Music\audiocraft>python -m demos.musicgen_app --share
WARNING[XFORMERS]: xFormers can't load C++/CUDA extensions.
I am getting this error message. Can you help?
I got the same error. Say's its a problem with the package module.
Same error, tried your solution, but the second time trying to install Xformers Ive got the same problem again.. Any idea?
@@dgodiex you need PyTorch which isn’t in the steps they assume you have all of the stuff already
I got an error on the "PYTHON -M PIP INSTALL --share" step - not sure what to do next, it just says theres a problem with the module and it quits. It was going really well up until that step.
manually installed PYTORCH and it works i think
im in - i was typing a . instead of a _ in musicgen_app - once i fixed that it went - downloading modules now i hope its going!
How do I make all this work on my graphics card?
Do you have any idea if Apple Silicon graphic cores are or can be used? I looked in documentation but don't see any clues. thx
It runs fine on Mac studio M2, takes around 30 GB of memory.
Lol I guess my 8GB Macbook Air M1 can't handle this.@@carloshanneman
Can this fine tune the existing audio file in terms of sound engineering like reverb, denoising, stereoing etc?
There is finetune code out there if you want to try to train your own models for stereo etc. Have not tried it yet myself.
@@digidope Okay
how do you run it in gpu?
It works, thanks!
But why does it use cpu instead of gpu? Is it normal or I should look for a solution?
There’s probably a solution, I just wasn’t able to find it by the time I wanted to post the video.
@@matthew_berman It runs on GPU by default. Not sure how you managed to run it on CPU.
audio webui has all the voice audio and music all in one with some extras
Noice
Try it out, let me know what you think!
how to make it use GPU instead CPU?
is this the same thing from your past video "How to Install Meta's FREE Text-to-Music AI Generator Locally (AudioGen)" ?
Nice. how about a google colab version?
How to install meta musicGen Model ? If possible please make a video How to install Model in musicGen .
unable to install xformer on my mac book pro 2017 model.
afaik xformers works with nvidia gpu.
It's great, but I'm really waiting and searching for audio restoration Ai tools, esp podcasts, which still made really low quality, there's almost no choice. Adobe provides a demo tool, but it's very bad with non-English podcasts. Other tools very hard to install and they work only with raw wave audio, anyone can calculate how hour podcast in wave format will be many gigabytes only for one file, and I need to restore a list of podcasts.
iZotope RX is pretty good for audio restoration. AI tools are stil quite far from that due the very low quality they use.
@@digidope thanks for advice, I will check.
Thanks for the video. Can anyone verify if the process is the same for PC?
It is. Check out the description for my installation instructions, I modified it for PC as well.
@matthew_berman Thank you sir. Always appreciated
'brew' is not recognized as an internal or external command,
operable program or batch file.
I used choco. It's good now.
Thank you for making this video: installs like these can have so many hurdles is you don't do them all the time. I do think you forgot step 1, though: install anaconda haha.
Anaconda required for most of my tutorial videos :)
No GPU support? Why?
what about adding lyrics? Will it sing the song with lyrics?
I don't think so...but you can sing along :)
Oh okay. I saw another AI application that adds the lyrics into the song. The user will write the lyrics and the AI will sing it into the audio generated file
@@emmanuelkolawole6720which?
conda install -c conda-forge ffmpeg
If you are on Windows though Anaconda
Thanks!
What's the difference between musicgen and audiocraft?
Audiocraft is like the base of musicgen and audiogen, musicgen is for generating music, and ofc the audiogen is for sound effects
Qwen-7B: Alibaba's NEW Opensource LLM
Why would you want it locally?
Did anybody tried to make any swosh, clicks and these kinds of sound effects, it would save a lot of money buying and searching for these files for video transitions or web effects.
'ai' is not recognized as an internal or external command,
operable program or batch file.
"ai projects" is part of his prompt. i.e., you only type what comes after it, for the first command it would be "conda create -n ac python=3.9" (without the quotes).
WHAT IF YOU DONT KNOW WTF A CONDA ENVIRONMENT IS?
Dude, you have some crazy people in your comments 😵💫
It’s UA-cam, bound to happen
God. Developers who use a Mac should have their own internet so they stop suckling off ours built on machines that let us do what we want.
Only the code is open-source.
For 30 second melodies works well, and actually the melody model is the best, but any model after 30 seconds still sucks.
Which ones did u compare?
conda create -n ac python=3.9
conda activate ac
My conda just doesn't work anymore... i am about fully factory reset my garbage PC there is so much shit that just doesn't work like i see in tutorials. Windows 10 sucks!
As always as all facebook can create is a low quality shht
Played around a bit and the results are really shit...especially for melody conditioned generation.
My man, I think mainstream people use PCs You should make tutorials on windows only.
I use a PC, but not Windows. It's the same process for Windows, Linux, etc... it's just a conda virtual environment.
I do both PC and Mac
@@matthew_berman That's what she said 🙂
There goes the junior musician/sound engineer job!
Amazing, this technology is going to create such a void in this world that will be filled with blood and destruction. Unless UBI is implemented
There will still be need of artists to be able to create good music/sound, as of now think of these tools as assistants as the quality of the generated content isn't as good
@@just_one23
👍Agreed!
For anyone thinking this is the End of your careers?
(& UBI is meeded? 👎)
This is our glorious beginnings!
Your glass is now! Half-Full!
Not for a while. The quality is really low and you cant really control the output.
Every time I see comments like yours it seems clearer to me that all this discussion about "we will be replaced at work by AI" is simply cheap sensationalism and an ideological excuse to lead to something that does not make the slightest economic sense like UBI.
@@digidopewho cares about control? 😂 its over
dude this is not enough, it will be without cuda
Just don't ask it to make chiptunes - absolutely devastated my ears.
crap this is exactly what I wanted to do with it
If Pence doesn't think Trump should be indicted Trump might be right about him being delusional.
Not sure what this has to do with my video...
@@matthew_berman In a LLM it's all connected somehow. But I think I might hove been watching a different video before yours.
ModuleNotFoundError: No module named 'julius' Why???
did you do the pip -r requirements.txt ?
@@jonathaningram8157Yes