MiniMax H3, the third generation of the 'Hailuo', series that MiniMax has developed as a video generation tool, was officially released as open source on the 3rd August 2026, with their next generation, general purpose video model including several advanced AI video generation features.
Native Multimodal Generation
Unlike traditional pipelines that separate video and audio creation, MiniMax H3 generates text, images, video and native stereo audio simultaneously in one pass. The audio includes background music, sound effects and stable dialogue across 11 languages.
Omni Reference Workflow
The model solves workflow fragmentation by accepting a mix of up to 12 reference files which includes up to a maximum of 9 images 3 video clips and 3 audio clips in a single request. Users can dictate instructions in plain natural language, such as telling the model to extract camera movement from one video, a character profile from an image or a voice style from an audio file.
Video Output & Resolution
It natively generates clips between 5 to 15 seconds long at 24 frames per second. Using its 'In-Context Regeneration', system, the model processes generations at a lower resolution of 480p or 768p and upscales them up to a crisp 2K or 1440p resolution, adding context-aware detail rather than guessing pixels.
Multi-Shot Prompting
MiniMax H3 natively handles time coded cuts directly from a single text prompt, allowing creators to generate sequential scenes, essentially creating several seamless shots inside one clip, with consistent characters, objects and environments.
Language & Instruction Precision
It stands out for its text-rendering capabilities, keeping letters and brand logos stable and sharp during motion sequences. It also features plain language editing, allowing users to tweak generated videos via text commands while keeping the rest of the scene stable.
