A fal-post-trained version of MiniMax H3 generates video with inference optimized for low latency.
H3 Max is fal's post-trained version of MiniMax H3, developed to improve how generated video follows a prompt and realizes a visual style. The team worked on the model alongside its own inference system, connecting output quality with the practical cost of serving video generation. Released in August 2026, the text-to-video endpoint makes that work available through fal's playground and API. Its story is one of adapting an existing foundation model and the infrastructure around it into a production tool for other builders.
Be the first Cosign
Add it to your Products I use list. You can also share why you use it.
