Summary
Fahd Mirza goes hands-on with MiniMax HiLo H3 (“HiLo” meaning sea snail in Chinese), a newly released open-weights video generation model from MiniMax that supports text-to-video, image-to-video, reference-to-video, and first/last frame generation. The model outputs up to 2000-resolution clips with native audio-visual generation, accepts up to nine reference images, three reference videos, and three audio clips as input, and generates clips from 4 to 15 seconds across all common aspect ratios including cinematic 21:9 and vertical 9:16.
Mirza generates two test outputs live using the API: a romantic beach scene from a text prompt, and an image-to-video animation of a man dancing in a nightclub from a single still frame. The beach scene earns praise for correct golden-hour backlighting, natural anatomy, and clean water detail — with minor complaints about plasticky water-sand interaction and the sun’s position deviating slightly from the prompt. The identity-preservation test on the dancing clip is assessed as a standout: the model faithfully held face, hair, jacket, and accessories throughout the animation while constructing a convincing lit environment and animated crowd.
The creator notes the model is aimed at commercial workflows including advertising, e-commerce, gaming, and interface design, with particular strength in instruction-guided edits, brand-accurate text rendering, and video-to-video motion transfer. Self-hosting via Hugging Face is planned imminently, and Mirza intends to follow up with a local installation video once weights are available.
📺 Source: Fahd Mirza · Published July 31, 2026
🏷️ Format: Hands On Build
