MiniMax H3

MiniMax H3 is an omni-modal AI video generator that uses text, images, video, and audio as one combined context—then produces a finished clip where the sound is generated inside the same pass as the visuals. It’s designed for short, cinematic shots with tight timing, delivering 4–15 second outputs at 24 FPS.

The workflow supports multiple ways to start a scene: text-to-video, image-to-video (using a first/last-frame style), and multimodal Reference-to-Video. Reference workflows can use images, MP4/MOV video, and MP3/WAV audio together, letting you reuse identity, motion, and dialogue direction rather than starting from scratch. A Prompt Reproduction Lab helps you copy “video recipes” by pairing real motion examples with the full prompt behind each one.

A standout capability is native audio generation: voice, effects, and ambience are created alongside the picture as 32 kHz stereo audio, rather than added later. Outputs can reach up to 2K, with the base output described as 768P and 2K achieved through in-context regeneration.

MiniMax H3 is credit-metered on this listing. Pricing shown includes annual plans with monthly credit allocations and image limits, plus descriptions of commercial-use license coverage and excluded territories.

What is MiniMax H3 best used for?

Use MiniMax H3 for short, cinematic video clips where you want the audio generated in the same pass as the visuals—typically 4–15 seconds at 24 FPS.

Does MiniMax H3 generate audio, or do I add it separately?

MiniMax H3 generates voice, effects, and music inside the same generation pass as the video, producing native 32 kHz stereo audio.

What video length, frame rate, and resolution can I generate?

It generates 4–15 second clips at 24 FPS, reaching up to 2K output. The base output is described as 768P, with 2K achieved through an in-context regeneration stage.

How does multimodal Reference-to-Video work with inputs like images and audio?

Reference mode can use images, MP4/MOV video, and MP3/WAV audio. Ref2VA caps references at nine images, three video clips, and three audio files, with a separate combined limit of twelve files; audio references also require an accompanying image or video.

What input types are supported in the H3 Video Workspace?

The Hero workspace supports text-to-video, first/last-frame image-to-video, and multimodal Reference-to-Video. Reference mode accepts images, MP4/MOV video, and MP3/WAV audio.

Is MiniMax H3 available everywhere?

Hosted access and the community license exclude the US, EU, UK, and Republic of Korea. In those excluded regions, treat it as a model to understand rather than for deployment.