MiniMax H3-Image: 16-megapixel open weights, but not for the US or EU

MiniMax's 33B H3-Image makes 16-megapixel pictures in six steps and edits with up to 16 references. Its license excludes the US, EU, UK and South Korea and caps commercial use at US$1M.

  • New release
  • 2 min read
  • Published

MiniMax released MiniMax H3-Image as open weights on 10 November 2026. It is a 33.1-billion-parameter image model from the MiniMax H3 family that writes pictures of about 16 megapixels in six sampling steps.

What is new

One checkpoint handles both text-to-image generation and editing with reference images, up to 16 of them, and it accepts prompts of up to 4,096 tokens. It writes images of about 16 megapixels directly, from 4096 x 4096 at 1:1 to 6272 x 2688 at 21:9, with no upscaler, and MiniMax gives a working range of about 4 to 20 megapixels. The checkpoint is distilled to six sampling steps and uses no classifier-free guidance.

The diffusion transformer is the H3 architecture without the audio layers, and the text encoder is the Qwen3-VL-based encoder from H3, unchanged. MiniMax says a 4K image takes under 10 seconds on its own service, up to 5 times faster than mainstream alternatives, and calls the model’s prompt-following the strongest among open-weight image models. These claims are MiniMax’s own, and it names no hardware.

Running it

The repository holds about 68.65 GB of BF16 transformer and autoencoder files, and the 66.7 GB text encoder loads from the MiniMax-H3 repository, so the full set is roughly 135 GB. MiniMax publishes no minimum GPU memory. SGLang, vLLM-Omni and ComfyUI can run the model on NVIDIA and AMD accelerators, and stable-diffusion.cpp added day-0 support on 11 November.

The model card expects a detailed English description of the whole image. For short prompts, MiniMax offers a hosted prompt-rewriting API and a local alternative, MiniMax-H3-Image-PE, a prompt expander of about 17.9 GB in NVFP4 that is derived from Qwen3.8-27B.

License

This is the part to read before you download. The MiniMax-H3-Image Community License covers only the world outside the European Union, the United Kingdom, South Korea and the United States. You need MiniMax’s written authorization if you and your affiliates earn more than US$1 million a year, or if you offer the model as a hosted service to anyone else, whatever your revenue. Commercial products must show “MiniMax-H3-Image” in their interface, and you may not use the model or its outputs to train other AI models.

Why it matters

The territory clause is the same as on the H3 video model, but the commercial threshold is 20 times lower: US$1 million a year against US$20 million. Only the license of Drama 3, at US$500,000, sets a lower threshold in our catalog, while FLUX 3 Image Klein sits at US$5 million. For a hobbyist outside the excluded regions the model is free to run, and for a company it needs a conversation with MiniMax early.

Our revenue cap and territory restriction entries explain these terms, and FLUX.3 Image Klein vs Z-Image compares two other open image options. The image and video list shows the whole category.

Sources

Models in this story

Model Released Parameters Context License Smallest variant needs
MiniMax H3-Image MiniMax 33.1B – MiniMax H3-Image Community (Conditional) 155 GB Big-memory desktop
MiniMax H3 MiniMax 33.1B – MiniMax H3 Community (Conditional) 166 GB Big-memory desktop
FLUX 3 Image Klein Black Forest Labs 5.6B – FLUX Kommunity (Conditional) 8 GB Laptop or phone