MiniMax H3 builds a clip around the reference images you give it — up to nine people or objects it has to keep — and makes sound with the picture. It runs on Reelfo as a managed API call, with nothing to install, at 5, 10 or 15 seconds.
ModelsH3 is a reference-to-video model: you hand it up to nine images of the people and objects the clip has to keep, and it builds the shot around them. That is the input a cast needs, so the same person can appear in shot one and shot six.
It is built for scenes with more than one subject, each held to its own reference, rather than one hero and a blurred crowd.
The clip comes with its own audio track, made together with the picture. Reelfo asks for it by default and the model does not offer a silent mode.
Sign in and your first 100 credits are already there — no card needed.