MiniMax H3
An omni-modal generative system that understands mixed text, image, video and audio input and generates video with native stereo audio, distributed with nine bundled prompt-writing and video-generation skills for agent harnesses.
MiniMax H3 is a general-purpose, omni-modal generative system. It understands multimodal context made of text, images, video and audio together, and generates video with native stereo audio, at resolutions up to 2K, in durations from 4 to 15 seconds and a range of aspect ratios.
The repository bundles nine skills. One, h3-prompt-writing, is a plain Markdown and reference-file skill with no external API calls, so it works in Claude Code, the Claude Agent SDK, Cursor, Windsurf, OpenAI-based agents such as Codex, LangChain, or any other harness that can read a skill file and local files; it ships two prompt guides, one for text and keyframe modes and one for full-reference mode. The other eight are style-specific video-generation skills built for MiniMax Hub's own canvas workflow and are not portable to other agent harnesses.
The model can also be used directly through an online API or through web and desktop apps. It is written in Python.