Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Max generation time 30 seconds #13

Open
gitfabianmeyer opened this issue May 28, 2024 · 8 comments
Open

Max generation time 30 seconds #13

gitfabianmeyer opened this issue May 28, 2024 · 8 comments

Comments

@gitfabianmeyer
Copy link

Hi, first of all, great work! Saw this on reddit. Is the 30 seconds limitation due to the model behind it or what is the reason? Is it planned to increase this? Or even possible?

@gabotechs
Copy link
Owner

It's a bit involved, the current models only generate up to 30 seconds, this is a hard limitation of the models used (https://audiocraft.metademolab.com/musicgen.html)

There's a way of overcoming this limitation using a variation of the model that can be conditioned not only by text prompts but also by music samples (https://huggingface.co/facebook/musicgen-melody), my plan is to implement this.

@freetoad
Copy link

freetoad commented Jun 7, 2024

does musicgen-melody allow users to generate vocals?

@gabotechs
Copy link
Owner

I don't think so, musicgen-melody just allows conditioning the music generating based on other music samples, but it's still not trained to generate vocals

@Nurb4000
Copy link

Looking forward to this, as 30 seconds is not overly useful beyond "hey, it works" ( and i realize its not your fault, appreciate what you have done.. ). Its too bad they 'hard limit' to such a small time.

It's a bit involved, the current models only generate up to 30 seconds, this is a hard limitation of the models used (https://audiocraft.metademolab.com/musicgen.html)

There's a way of overcoming this limitation using a variation of the model that can be conditioned not only by text prompts but also by music samples (https://huggingface.co/facebook/musicgen-melody), my plan is to implement this.

@adamperez
Copy link

There's a section on the MusicGen site titled "Long generation", where they reference having a sliding window to make a longer track. Is that something that is surfaced with this tool?

@George3d6
Copy link

Also curious about what @adamperez mentioned? And, if not surfaced, how easy would it be to add? I'd be interested in contributing it

@rubeniskov
Copy link

To support these features, the ONNX export models need to be available for melody models.

huggingface/optimum#2095

@designspin
Copy link

This is super impressive and lightning quick on my m1 Mac studio... Overcoming the 30 second limit would be awesome, this would be a super time saver for my indie game music. It takes me weeks to come up with something myself, this is churning out better results in minutes!

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

8 participants