Generate a short video with its own synchronized soundtrack for free, without an account: MiniMax-H3 is the open-weights video model released by MiniMax, the company behind Hailuo, and produces picture and sound together from a text prompt, or from a first and a last frame that it animates in between. This demo runs a 4-step Turbo version for faster results.
MiniMax-H3 is an open-weights video generation model from MiniMax, the Chinese company behind the Hailuo video service. Unlike most open video models it generates the audio track together with the frames, so speech, ambience and effects are synchronized with the picture. It runs at 24 frames per second.
Wan 2.2 produces silent clips from an image or a text; MiniMax-H3 produces sound and picture at once and can be guided by a first and a last frame. Both are open weights, both need a large GPU: the H3 weights alone are about 77 GB, which is why the demo is hosted on shared hardware.
Video with audio is the heaviest generation on this site: each clip needs a data-centre GPU for a minute or more, and the queue is shared. This demo is hosted by a third party (Hugging Face Space) and embedded here; it can be slow, queued or temporarily unavailable when the host is busy. Nothing to install, no account needed.
Read the guide: How to write a video prompt (Wan 2.2 guide, same rules)