Hey everyone, we’ve been working on BachGround, a video-to-music / AI music generation project, and we recently prepared a cleaner public-facing repository for one part of our research: a text-to-piano model. The idea is pretty simple: text prompt → symbolic piano tokens → duration/velocity enrichment → MIDI Instead of trying to generate everything in one model, the pipeline is split into two stages: A fine-tuned Llama-based model generates the musical structure / piano token sequen...