Built a web editor that converts any pose photo into semantic animation code using computer vision. Demo: https://popo.love/playground, https://mmd-mpl.vercel.app/ How it works:

Upload pose photo → MediaPipe extracts landmarks My solver generates readable MPL code like leg_l bend forward 120; arm_r bend forward 45; Edit the code, export to animation formats

Tech stack: Rust compiler (WASM), MediaPipe, Babylon.js 3D preview. Runs entirely in browser.

The language (MPL) bridges human-readable syntax with quaternion math. I've used it to fine-tune LLMs for text-to-animation generation.

Use cases: Game dev prototyping, VTuber expressions, animation learning, AI motion research.

GitHub: https://github.com/AmyangXYZ/MMD-MPL