MAJEPPA — Conditional Piano Performance Generation

Score MIDI in → expressive performance MIDI out, conditioned on performer skill and recording context.

⏳ First call after the Space wakes takes ~90–120 s. Subsequent generations within the session are ~40 s each (Since we are using ZeroGPU on huggingface...).

Pick a curated example and hit Generate. Each preset is a hand-tuned (score, style) pair.

Preset

Download generated MIDI

MAJEPPA — anonymized review demo. Generation is autoregressive over MIDI tokens with a LoRA-adapted MIDI language model and condition-tagged prompts.