MiniMax H3-Image: Free 4K Image Model, With a Big License Catch
MiniMax released H3-Image on November 11, 2026: open weights, native 4K and 16 references. Its license excludes the US, EU, UK and Korea. Here is what to know.
Illustration made on Lumeta with GPT Image 2.5Make one like it
Chinese AI company MiniMax released H3-Image on November 11, 2026. It is an image model that makes and edits pictures at about 16 megapixels (4096 by 4096 for a square) without a separate upscaler, takes up to 16 reference images, and comes with its weights free to download on Hugging Face.
That sounds like a gift for anyone who wants to run a 4K image model on their own hardware. There is a catch, though: the license that comes with the weights does not cover the United States, the European Union, the United Kingdom or South Korea. If you live there, the paid API and MiniMax's own apps are the way to use it.
What H3-Image can do
- Native 4K. It works best around 16 megapixels, in shapes from square to 21:9 widescreen.
- One model for making and editing. The same checkpoint handles text-to-image and edits with references. You call your uploads Picture 1, Picture 2 and so on in the prompt.
- Up to 16 references. Useful for furnishing a room from product shots, or turning a 3D render into a photo.
- Fast for its size. MiniMax says a 4K image takes under 10 seconds in its faster Highspeed mode. Measured median times in one independent write-up were about 16 seconds in Standard and 10 seconds in Highspeed.
- Built on MiniMax H3, the company's video model, with the sound part removed.

Illustration made on Lumeta with GPT Image 2.5Make one like it
How much it costs
On MiniMax's API pricing page, checked on November 16, 2026, a Standard text-to-image picture costs about $0.022 at 1K and $0.023 at 2K. At 4K it lists about $0.069, cut in half to about $0.034 by a launch discount that ends November 17, 2026 at 16:00 UTC. The Highspeed version costs about $0.035 at 1K and $0.130 at 4K (about $0.065 during the discount). Edits with reference images cost a little more.
The license catch
H3-Image is open weights, not open source in the usual sense. Its community license lists the United States, the European Union, the United Kingdom and South Korea as excluded territories, and says use there is not authorized; MiniMax invites people there to ask for a separate license. Elsewhere, a company with more than $1 million in yearly revenue, or anyone offering it as a hosted service, needs written permission first. Commercial products must show the H3-Image name, and you may not use it or its pictures to train other AI models.
Can you run it at home?
Only with serious hardware. The image model alone is about 66 GB, and the text encoder it needs is another 67 GB or so. Community tests show a 4K image in about 56 seconds on an RTX 3090 with compressed weights, and about 9 minutes on a 64 GB Mac mini. MiniMax has not published quality scores, and early hands-on comparisons are small: it looked natural in portraits but struggled with small text.
What you can do with it now
H3-Image is in MiniMax's Hailuo AI app and on its API platform. It is not on Lumeta yet. For high-resolution images and edits with many references on Lumeta today, Seedream 5.5 takes up to 10 reference photos, and the video model from the same MiniMax family is here as MiniMax H3 Video.
Seedream 5.5Make or edit high-resolution images with up to 10 referencesWhen did MiniMax release H3-Image?
MiniMax announced it on November 11, 2026, with weights on Hugging Face and access through its API and the Hailuo AI app.
Can I use H3-Image in the United States?
Not under the free community license, which excludes the US, the EU, the UK and South Korea. MiniMax's own app and paid API are the options there, or a separate license from MiniMax.
How much does an H3-Image picture cost?
On November 16, 2026 the API listed about $0.022 for a 1K picture and $0.069 for 4K in Standard mode, with 4K at half price until November 17, 2026.
Is H3-Image on Lumeta?
Not yet. Lumeta has MiniMax H3 Video and other image models such as Seedream 5.5.


