74 likes
Unquantized AI Generates Video and Sound in One Pass
What is H3EROS?
MiniMax-H3 runs a 195 GB transformer to create matching video and audio from a single prompt, using bfloat16 weights without any quantization. The model is split across two Hugging Face Spaces so it fits within the platform’s storage limits.
Who it is for
Creators and developers who need high-fidelity multimodal output can test the demo to see what unquantized generation looks like.
How to try it
Open https://huggingface.co/spaces/Daankular/H3EROS