A voice AI model already knows its own vocal range. It’s in the weights
Diff the pitch embedding against the base, see which slots moved, read the range out in microseconds. No audio played. Auto-key then just counts which octave shift fits.
So while times have been tough, and times have been tough, you have been put down, washed up, erased out.
But honestly, and be honest with yourself:
How much time do you waste??
How much time do you blow every day??
A voice AI model already knows its own vocal range. It’s in the weights
Diff the pitch embedding against the base, see which slots moved, read the range out in microseconds. No audio played. Auto-key then just counts which octave shift fits.
an ai agent edited this music video. it's driving a video editor i built from scratch in c++
7mb binary, native linux, no python, torchless onnx inference. the editor and all the ml run locally on your machine
gave it a pile of random clips and a song. it built the cut, i directed from there
agentic video editing, fully open source
Built Pop Maker Studio data-first: one AppState struct is the entire app. UI redraws from it every frame.
Result: the MCP server cost almost nothing. Claude’s tools are just another consumer of the same struct the UI is. No binding, no events, no negotiation.