Photo: Merlin Lightpainting / Pexels
The Chinese AI firm's 33.1 billion parameter model combines image generation and editing, but its license excludes users in the US, EU, UK, and South Korea
MiniMax has released H3-Image, an open-weight image model that generates and edits pictures at 4K resolution.
The model went live on November 10, 2026. It is available on Hugging Face under the MiniMaxAI organization, through MiniMax Code, and via the company’s APIs. Developers can download the weights and run the model themselves, or pay per image through the API.
There is a catch for a large share of the world’s developers. The community license excludes use in the US, the EU, the UK, and South Korea. The weights are public, but the welcome mat has a few countries crossed off.
What H3-Image actually does #
Most image tools split the work in two: one model creates images from text and another edits them. H3-Image handles both jobs in a single framework, covering text-to-image generation and advanced editing.
Under the hood, it runs on a diffusion transformer, or DiT, architecture. The model has 33.1 billion parameters, the internal settings it learned during training.
Output resolution is the headline feature. H3-Image produces native images of approximately 16 megapixels, with supported sizes including 4096×4096. It can also go wider, up to 6272×2688, which is a shape better suited to panoramas and banners than profile pictures.
“Native” matters here. Many image models generate at a lower resolution and then upscale, a bit like stretching a photo and hoping nobody zooms in. Native output means the model builds the image at full size from the start.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Speed, references, and the fine print #
Generation is efficient. H3-Image needs only about 6 sampling steps to produce an image, and it does so without classifier-free guidance.
The model also accepts up to 16 reference images at once. That gives users a way to show the model what they want, such as a character, a product, or a style, rather than describing it entirely in words.
Pricing on the API runs approximately $0.022 for a 1K image and approximately $0.069 for a 4K image. MiniMax is also offering promotional discounts until November 17, 2026.
Running the model locally comes with an additional requirement. Full functionality depends on a separate text encoder, Qwen3-VL-32B, which translates prompts into a form the image model can work with. So the download is really two large models, not one.
Integration arrived quickly. ComfyUI, the node-based interface widely used for building image generation workflows, supported H3-Image immediately upon release.
The licensing question #
The community license bars use in the US, EU, UK, and South Korea. On top of that, larger companies need written authorization from MiniMax to use the model.
A derivative already in the mix #
The ecosystem around H3-Image is already moving. A derivative model called P-H3-Image was announced on November 3, 2026, with claims of faster inference on consumer hardware.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.