MiniMaxMiniMax H3

MiniMax H3

H3 generates video with native stereo audio from text, images, video or audio references. Its system separates reference understanding, base generation and a 2K regeneration stage.

Restricted access

Reference mode has a shared input limit

The reference pathway accepts up to nine images, three video clips and three audio clips, within a combined limit of 12 files. Video and audio totals are each limited to 15 seconds. The frame pathway accepts zero, one or two images.MiniMax H3 system and model card

Commercial terms apply in addition to geography

The community licence requires visible MiniMax H3 attribution in commercial products. Commercial products and services above US$20 million yearly revenue require prior written authorization. Its territory rules also apply to model outputs.MiniMax H3 Community License

Compare equivalent workflows

Practical guidance. A base-weight result and a hosted 2K result may use different processing stages. Hold reference material, output size and audio requirements constant when comparing quality. Confirm the licence for the location where the model and outputs will be used.

Specifications & access

MiniMax H3 specifications
DetailValue and conditions
StatusRestricted access
AccessAPI, Downloadable, Provider product
Model identifiersMiniMax-H3MiniMax-H3-HighspeedMiniMax-H3-LiteMiniMax-H3-MaxH3-Base-FL2VAH3-Base-Ref2VA
Standard output4–15 seconds at 24 frames per second, with 32 kHz stereo audio. The base output has a 768-pixel short side; 2K uses regeneration.MiniMax H3 system and model card
API variantsStandard H3 supports 768P and 2K. Highspeed, Lite and Max offer 480P or 768P; Max starts at five seconds. These options are grouped here.MiniMax H3 V2 generation API and variants
Download boundaryThe published base weights do not include H3-Context-IR or the complete 2K workflow.MiniMax local deployment overview
Territory restrictionThe weight licence excludes the EU, UK, South Korea and USA unless separate permission is obtained.MiniMax H3 Community License

Related models

  • A text-to-video and image-to-video model with a separate Fast option for image input.

    Launch date API
  • An earlier hosted video model with text and image input and 512P, 768P and 1080P output options.

    Launch date API