YuE-UI gradio interface featuring incremental song generation, low vram support

[joeljuvel]

Thank you for the great project!
I've made a gradio interface YuE-UI focusing on the creative experience, centered around a batch generate, select & continue workflow. Play with settings as you go. Like you would do with a synth. For example you could generate a cool start using an audio prompt with the ICL model. Then you could continue the rest of the song with another model and other settings. If you make a mistake you can rewind!
You can visualize what you've generated so far with a timeline. Another feature is stage 1 post process audio output, so you can decide if you really want to use a clip before refining it with stage 2, speeding up the creation process! There's full save & load support (json).
It works great with only 8 gb vram when using quantized models.

For the inference I'm using YuE-exllamav2 and the interface was started from YuE-Interface

[a43992899]

Fantastic work! I'm including this right away in the news update of the README!