How do i make it generate only instrumental

[GeeveGeorge]

It currently generates whole songs with vocals.

But sometimes I only need instrumental. How can I do that?

[GeeveGeorge]

Also what are the inference vram requirment for the different sized models

[nitinmukesh]

@GeeveGeorge

What steps you followed to install, A quick guide plz. Also how much VRAM it needs.

[platform-kit]

@GeeveGeorge I'd love to know how you got it running.

[CyberTimon]

VRAM usage is "not optimized" but with T5 in 4 bit it's around 20GB for the giant model. I'm very optimistic that this can be improved significantly.
Keep in mind that it doesn't generate full songs with samples. Instead I only got it to generate 10 second melody recordings which are low quality and definitely not anywhere near as good as what is shown on Melodio.
I'm still unable to get Clap-L working. <- this might lower the quality of the samples by a good amount so no final conclusion on the audio quality.

[nitinmukesh]

@CyberTimon

Thank you for sharing your experience with this tool. Please could you share quick installation steps, plz.

[painebenjamin]

I believe this post is referencing https://www.melodio.ai/ (which is linked from the ReadMe,) not FluxMusic.

[feizc]

Hi, you can try negative prompt in the sample.py.

[daiw10]

请问下现有 infer 能产生人声的原理是啥？ 输入 prompts 并没有带任何有意义的歌词，输出的人声的内容应该没有什么逻辑？