search

riffusion / riffusion-hobby

Stable diffusion for real-time music generation
MIT License
3.36k stars 386 forks source link

Best prompts so far - Best settings - Please share and build our knowledge base #120

Open shiloh92 opened 1 year ago

shiloh92 commented 1 year ago

I'm surprised so few prompts have been shared in this space. It seems we are just getting started in AI music! Mubert can't compete with the flexibility of SD used for music... and I'm really grateful for the work by the riffusion guys/gals.

Please reply with your best settings to produce higher clarity sound so the community can experiment.

-Wider images 512x1024 seemed to yield more variety , ie, less repetition -More references to musicians or albums seems to yield slightly better quality -Negative prompts I've tried: (male vocals:1.2)(female vocals:1.2)(speech:1.2),(simple:1.2), (looping:1.3),(((repetitive))), (((missing harmony))), (((poor quality))), (((high noise))), (((poor mastering))) -- but I can't tell yet if it makes a difference

Anyone else have tips, tricks, please share ?

rbong commented 1 year ago

I can't tell yet if it makes a difference

Best way to tell in general if your prompts are making a difference, assuming you don't want to generate 100s of images, is to try again with the same seed and different prompts.