MiniMax H3

• 🆕 MiniMax's newest-generation video model line, released July 2026.
• 👁️ Image-to-video variant: animates a supplied still into motion.
• 🏢 Described by MiniMax as an open general-purpose multimodal video model.
• 🎯 Shares generation with sibling text-to-video and reference-to-video endpoints.
• 🔧 Routed as a separate catalog entry per input modality.
• 📚 Part of MiniMax's wider stack spanning text, speech, music, video.
• 🌐 Documented on MiniMax's own platform API docs.

MiniMax

MiniMax is an AI company building generative models across multiple modalities, with a focus that spans both language understanding and audio creation. Their rapid release cadence in early 2026—delivering several new models within just a few months—reflects…

MiniMax H3 (image-to-video) is the picture-conditioned entry point to MiniMax's H3 video generation line, released in July 2026. You supply a still image, optionally with a prompt, and the model produces a moving clip from it. MiniMax's own API documentation announces H3 as a new-generation open general-purpose multimodal video model, rather than a narrowly task-specific generator.

It ships alongside two same-generation siblings that share the H3 name and release date but differ in what they consume: MiniMax H3 builds a clip from a written prompt alone, while MiniMax H3 R2V conditions on reference material to carry a subject, character, or style across into the output. In this catalog the three are listed as separate families purely because their input routing differs.

Compared with MiniMax's earlier video generations, which were published as individual, purpose-built video endpoints, H3 is positioned by the provider as one general-purpose multimodal video system covering the generation surface as a whole. Detailed clip-length, resolution and benchmark figures are not covered here, as no primary MiniMax specification or independent evaluation of H3's video quality is cited.

H3 sits within a broad MiniMax model family: reasoning and coding models such as MiniMax M3 Preview and MiniMax M2.7, speech synthesis in MiniMax Speech-02 HD, and music generation in MiniMax Music 2.6.