language-control

H3-World transforms the MiniMax-H3 video generator into an interactive world model by enabling language-based temporal control, achieving effective control with minimal fine-tuning.

OpenVoice is an open-source instant voice cloning model with style and language control, now available on GitHub.

A 3D avatar that can be controlled using natural language descriptions instead of buttons, built on the programasweights system that compiles neural programs from plain English. It runs locally in the browser and can generate action sequences like 'wave while walking, then jump a couple times.'