Live Music Diffusion Models: Efficient Fine-Tuning and Post-Training of Interactive Diffusion Music Generators

A method for converting offline audio diffusion models into real-time interactive streaming generators via block-wise outpainting fine-tuning targets consumer hardware deployment. Existing SOTA music generators use discrete autoregressive architectures requiring industrial compute, making live performance use cases impractical. This approach repurposes open-source diffusion models, potentially enabling low-latency co-creation tools without retraining from scratch.