MiniMax H3 R2V

• 🆕 Reference-to-video variant of MiniMax's H3 video family, 2026
• 🏢 MiniMax describes H3 as a new-generation open multimodal video model
• 👁️ Generation conditioned on supplied reference material rather than prompt alone
• 🎯 Aimed at holding subject identity and look across shots
• 🔧 Complements H3 text-to-video and image-to-video endpoints
• 🌐 Served through MiniMax's platform API and partner catalogs
• 📚 Sits in a lineup spanning text, music, speech and video
• 📏 No public parameter count, license or architecture details released

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 R2V is the reference-to-video entry point of MiniMax's H3 video generation family, released in 2026. Where a text-to-video system works from a written prompt alone, a reference-to-video mode takes user-supplied material — such as images or clips of a character, product or scene — and conditions the generated footage on it, so that the subject's appearance and styling carry through the output rather than being reinvented on each run.

MiniMax positions H3 as a new-generation, open, general-purpose multimodal video model in its own platform documentation, treating multiple input modalities as one shared context instead of separate single-purpose tasks. That framing is what separates the H3 generation from MiniMax's earlier single-input video endpoints: rather than one prompt type per model, the family exposes several conditioning routes over a common backbone.

Those routes are the siblings listed here: MiniMax H3 for prompt-only generation and MiniMax H3 for animating a still, both dated to the same 2026 release. R2V is the variant to reach for when consistency with existing assets matters more than open-ended invention.

Beyond video, MiniMax ships models across other modalities, including MiniMax M3 Preview and MiniMax M2.7 for text and reasoning, MiniMax Music 2.6 for music, and MiniMax Speech-02 HD for speech. MiniMax has not published parameter counts, licensing terms or architectural details for H3 R2V, so this description stays limited to what the provider states directly.