The Model
MiniMax H3 is the video model from MiniMax that this page runs. It takes a written prompt, up to 9 reference images with clips and audio, or a first and last frame, and returns an MP4 of 4 to 15 seconds with sound already in it. The MiniMax H3 AI video generator above is the browser front end for it: no API key, no install.
What sets it apart from a silent clip model is that speech, effects and music come out of the same pass as the picture. That changes how you write for it: a good prompt describes what is heard as well as what is seen, and a good result needs no dubbing step afterwards.
We run H3 beside Wan 3.0 in one workspace and bill both from the same credit balance, so you can try the same idea on each and keep the better take. Failed renders return their credits automatically.