About
The person who built HeiChitty-AI
96HH · Bassist · Composer · Software Engineer
HeiChitty-AI is a personal project built by one person. From planning and design through the frontend, the backend API, the AI worker and the audio pipeline, every piece was built and wired together by hand. The models used for composing and separation are only open source that allows commercial use.
It started simply. I wanted to hear a song that came to mind without having to explain it to anyone — but doing that first demanded arranging and instrument knowledge and a DAW. Closing that distance down to “one line of text” is what this service is aiming at.
So HeiChitty-AI does not synthesise vocals; it focuses on instrumental music. The lead melody is played by an instrument into a finished song, and the same melody also comes as a vocal guide MIDI. Letting you download the MIDI, the per-track audio and the master mix as they are, rather than locking the result in a black box, comes from the same thinking. A song you made should be yours.
If something seems off while you are using it, or a feature you wish existed comes to mind, please feel free to tell me.
What was built here
This is not a shell that forwards requests to a paid external AI API. The whole path from composing to mastering runs on its own pipeline.
Composing engine
Reads key, tempo and structure from the prompt to build the melody and accompaniment, then assigns tracks to match the instrumentation you asked for.
Sound rendering
Performs the generated MIDI through our own SoundFonts into per-track audio. Each track comes out as its own file, ready to take into a DAW.
Automatic mastering
Sums the tracks and runs EQ, loudness normalisation and limiting to finish a mix you can play right away.
Stem separation · transcription
Splits an uploaded track into vocals, drums, bass, guitar and piano, and writes each part out as a MIDI score.