[emperorcj]

(context: i make ai death metal & also i worked on stable audio). 100% it was a dataset problem. Diffusion models still work well when you train them on death metal: https://www.youtube.com/watch?v=rlsRMQzD_6Q