Text2midi

Text2midi is an end-to-end model that generates MIDI files from textual descriptions, using pretrained large language models and an autoregressive transformer decoder to produce symbolic music that follows prompts for chords, tempo, and style. It provides a quickstart guide and a live HuggingFace Spaces demo, and its approach has been accepted at AAAI. The project is MIT-licensed and open-source, with code, models, and examples to generate music from captions.

AI named Text2midi from December 2025 to August 2026.