Krea adds MiniMax H3 video model
By Steven Van ·
The clips run 5-15 seconds at 2K with native stereo audio, and Krea says it matches Seedance for less.
Krea has added MiniMax H3, the new video model from MiniMax (also known as Hailuo 3.0), unveiled at WAIC 2026 and now rolling out fast across the AI video ecosystem. Krea's framing is blunt about the value proposition: H3 matches Seedance's features at a fraction of the cost, with open weights coming soon.
What H3 actually does
MiniMax H3 generates 5 to 15 second clips at 2K resolution and 24fps with native stereo audio baked in, not added afterward. Its signature capability is Omni-Reference control, feeding the model up to 9 reference images, 3 video clips, and 3 audio clips at once to lock in style, character, motion, and voice across a generation. Characters stay consistent scene to scene, which is the piece most video models still struggle with.
Open weights change the calculus
Krea's promise of open weights is the detail worth watching. Most frontier video models, Seedance included, stay closed and API-gated. If MiniMax follows through, H3 becomes something builders can self-host, fine-tune, and integrate directly rather than call through a metered endpoint, a meaningfully different position in the market than 'another expensive video API.'
Why it's rolling out everywhere at once
H3's simultaneous arrival on Krea alongside Magnific, Leonardo, Lovart, Pika, and Runway reflects how quickly a strong new open or near-open model gets absorbed into the multi-model platforms that now define the category. For Krea users, it means one more high-quality, cost-efficient video option sitting next to the image and video models already on the canvas.
Sources: Krea on X.