search

Audio-AGI / AudioSep

Official implementation of "Separate Anything You Describe"
https://audio-agi.github.io/Separate-Anything-You-Describe/
MIT License
1.63k stars 118 forks source link

Support for negative prompts #19

Open crlandsc opened 1 year ago

crlandsc commented 1 year ago

Does AudioSep have the potential to support negative prompts?

liuxubo717 commented 1 year ago

What does the negative prompts mean? :)

crlandsc commented 1 year ago

For example: Prompt: "voice and whistle, no wind or rumble"

From this: Positive prompt: "voice and whistle" Negative prompt: "no wind or rumble"

I have found the model has a hard time understanding sounds sometimes, so I was wondering if you could add a negative prompt to further guide it toward what you wanted to separate.