MiniMax launches H3 multimodal video model with open weights

Chinese AI startup MiniMax announced the release of its H3 multimodal video‑generation model. H3 can accept text, images, audio and video as inputs and produce up to 15‑second 2K videos with native stereo sound. It also supports editing existing clips and transferring motion between videos via V2V Motion Transfer.

The company said the model’s commercial‑grade output is aimed at advertising, e‑commerce, product design and gaming, and that generating a 2K video will cost less than one‑third of competing products. MiniMax plans to make the H3 model weights publicly available within days, extending the open‑weight approach spreading among Chinese AI developers. The move follows intensified competition from ByteDance’s Seedance 2.0 and Kuaishou’s Kling 3.0, and reflects China’s push to reduce reliance on U.S. semiconductors by designing H3 to run on domestic chips. MiniMax, founded in 2022 and listed in Hong Kong, is also developing a 2.7‑trillion‑parameter language model.