MiniMax H3 collapses per-task video, image and audio models into one
Source
x.com
Author
MiniMax (official)
Date
Context
MiniMax designed H3 to interpret a creative brief containing text, images, video and audio together, then generate video with stereo sound. Its launch article describes combining a camera movement from one reference, a character from another and an audio performance through written instructions. The intended benefit is that creators can specify how references relate within one request. MiniMax says H3 produces clips up to 15 seconds long at up to 2K resolution. These are the company's stated capabilities; its broader claims about production quality remain its own assessment. The launch article explains the design.
Terms in this piece · Glossary
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
pretraining — The first, biggest phase of building a model: training it on enormous amounts of text so it learns language, facts, and reasoning in general.
