About this model
MiniMax H3 Image Edit is the editing side of MiniMax's H3 Image release: you supply one or more source pictures together with a text instruction, and the model re-renders the result. According to this catalog's API description, it accepts up to 16 input images in a single request and can render at up to 4K. Its direct counterpart for generation from text alone is MiniMax H3 Image, released the same day.
The H3 name places it inside MiniMax's broader 2026 H3 generation, which began with the video models MiniMax H3 and later MiniMax H3 Max, alongside text models such as MiniMax M3 Preview. Where the video line produces clips from prompts or reference frames, the H3 Image models target still frames and persistent edits to them, and MiniMax publishes the image weights on Hugging Face under the MiniMaxAI organization.
Relative to doing edits by extracting single frames from the H3 video models, a dedicated image checkpoint is the more direct route: edits are expressed as instructions over supplied references rather than as motion prompts, and multi-image inputs let a subject, product or style reference be carried into a new scene. MiniMax additionally publishes H3-Image-PE, a prompt-expander model that rewrites short instructions into longer, more detailed descriptions for the image model to follow. Common uses include inserting logos or signage, swapping backgrounds or clothing while preserving a subject, and localized retouching.
This About section is AI-generated from public sources (Claude Opus 5), with no human editing. It may contain inaccuracies — verify critical details against the sources listed above.
Data sources: Venice API · HuggingFace · Wikipedia — enrichment updated 5d ago