MiniMax H3
MiniMax H3 is an open-weight, omni-modal video model released by MiniMax on July 31, 2026. It generates video with native audio at up to 2K resolution and 15 seconds.
What MiniMax H3 can do
Key facts from MiniMax and the public model pages on fal.ai.
Omni-modal input
Understands text, images, video and audio together as context for a generation.
Native audio
Generates sound in the same pass as the picture.
Up to 2K, 5–15 seconds
On fal: 480P and 768P native, 2K and 4K upscaled from a 768P base.
Open weights
Weights are published on GitHub and model hubs, so you can run H3 on your own hardware.
MiniMax H3 vs H3 Max
H3 Max is a post-trained version of H3 released by fal.ai in August 2026.
Pick H3 for flexibility
Multimodal references, seven aspect ratios on fal, and local deployment.
Pick H3 Max for speed
Tuned for prompt adherence, visual quality and fast generation.
Pick H3 for 2K/4K output
H3 on fal offers 2K and 4K upscaling; H3 Max goes up to 1080P.
No GPU? Use H3 Max here
Running H3 locally needs a large GPU setup. This site runs H3 Max in the cloud for you.