MiniMax released H3-Image: 4K images in under 10 seconds, with open weights
Direct answer
MiniMax launched H3-Image on November 11, 2026: an image generation and editing model with open weights that, according to the company, generates a 4096x4096 image in under 10 seconds and accepts up to 16 reference images. The weights are on Hugging Face and the code is on GitHub. But MiniMax's community license excludes the United States, the European Union, the United Kingdom and South Korea, and requires separate written authorization if your annual revenue exceeds $1 million or if you offer a hosted service.
What H3-Image is and what it promises
- One model, two jobs: the same checkpoint generates images from text and edits existing images with instructions, combining up to 16 visual references. Native output is around 16 megapixels (4096x4096 square), with no separate upscaler.
- 33.1 billion parameters, six steps: H3-Image is a 33.1-billion-parameter dense diffusion transformer, distilled to six sampling steps. The weights download is about 68.65 GB and deploys with Diffusers, SGLang, vLLM-Omni or ComfyUI.
- Speed is a vendor claim: MiniMax promises 4K in under 10 seconds. Its own API latency chart shows 16.0 seconds in Standard mode and 10.3 seconds in Highspeed mode. There is no published independent benchmark and no quality metrics for the model.
- The API costs $0.069 per 4K image: MiniMax charges $0.069 per 4K image on its API, with 50% off until November 17, 2026. The API is the usable route for a business in the United States; the weights, under the license, are not.
The license is the news: read before you download
- Excluded territory: the MiniMax-H3-Image community license grants rights only within the applicable territory, which excludes the European Union, the United Kingdom, the Republic of Korea and the United States. A business in Miami cannot use the weights under that license.
- Revenue cap and hosted services: if your annualized revenue, with affiliates, exceeds $1 million, or if you offer hosted services to third parties, you need separate written authorization from MiniMax. The license also prohibits using the model or its outputs to improve other AI models.
- No technical report and no metrics: MiniMax published no technical report, quality metrics or training datasets, and as of November 11 the model was not in the Artificial Analysis rankings. The claim of best prompt adherence has no published independent support.
- The API has its own safety check: the API guide documents a content safety check that can withhold an output (error 1027, and 422 if every image in the request is withheld). Open weights is not the same as an unfiltered mode.
What to do in your business this week
- If you are in the United States, use the API, not the weights: the license excludes your territory. Downloading the weights for production leaves you outside the license; the API at $0.069 per 4K image is the route with clear terms.
- Ask for the license in writing before integrating any open model: territory, revenue cap, hosted services and use for training other models. Four questions answered in ten minutes that prevent a legal problem.
- Measure speed with your prompts, not the vendor's chart: the under-10-seconds figure is MiniMax's, and its own chart shows 10.3 to 16 seconds depending on mode. Run your real prompts and measure time per finished image.
- Compare against what you already use: 16 references and native 4K are useful for product visuals and consistent editing. If your flow is simple text-to-image, the open model gives you nothing you do not already have in a closed API.
Note: this article summarizes MiniMax's announcement of November 11, 2026, the model card on Hugging Face and coverage by Times of AI, Neurohive and reAPI. The speed and quality figures are MiniMax's claims; the license cited is the text of the MiniMax-H3-Image Community License Agreement read on Hugging Face and GitHub.
Want AI that works for your business?
Message us on WhatsApp and let's talk about automation and AI agents for your business.
WhatsApp inicIARelated articles
- The open model you can run on your own server: Aleph Alpha Kolibri (Apache 2.0)
- Inkling 2 Small is downloadable now: the open weights arrived a week after launch
- How to use AI in your content without losing your voice
Frequently asked questions
What is MiniMax H3-Image?
It is a MiniMax image generation and editing model, launched November 11, 2026, with open weights: a 33.1-billion-parameter diffusion transformer that generates and edits images in 4K and accepts up to 16 reference images.
Does it really generate 4K in under 10 seconds?
That is MiniMax's promise. Its own API latency chart shows 16.0 seconds in Standard mode and 10.3 seconds in Highspeed mode for a 4K image from text. There is no published independent measurement.
Can I use the H3-Image weights in my business in the United States?
The community license excludes the United States, the European Union, the United Kingdom and South Korea, and requires separate written authorization if your annualized revenue exceeds $1 million or if you offer hosted services. The usable route in the United States is MiniMax's API.
Is H3-Image free?
The weights are published on Hugging Face under the community license, with its limits. The API charges $0.069 per 4K image, with 50% off until November 17, 2026.
