MiniMax Launches H3-Image: 4K Images in About 10 Seconds, Open Weights With a Restricted License
MiniMax launched the image model H3-Image on Nov 10, saying it generates a 4096×4096 4K image in about 10 seconds, supports up to 16 reference images, and ships open weights; the licence excludes the EU, U.K., U.S. and South Korea.

What happened
MiniMax announced H3-Image on Nov 10 in an X post, describing it as the image model of its H3 family, combining text-to-image generation and image editing in one checkpoint; the weights were uploaded to Hugging Face the same day, and the model card shows a 33.1-billion-parameter diffusion transformer distilled to six sampling steps with no classifier-free guidance, with a checkpoint of about 68.65GB. The model is available through its own API, MiniMax Code and Hailuo AI, with 50% off generations above 16 megapixels in the first week.
The model card says output runs to about 16 megapixels at every aspect ratio with no separate upscaler: 4096×4096 at 1:1, 5472×3072 at 16:9 and 6272×2688 at 21:9, with the API capping output at 20 megapixels. MiniMax says images land in about 10 seconds with up to 16 reference images; as of Nov 11, no third-party benchmark or independent timing exists, and the speed and quality figures are the company's own.
The licence is the key restriction: H3-Image uses the MiniMax Community License Agreement rather than a standard open-source licence; excluded territories cover the EU, U.K., South Korea and the United States, revenue above $1 million a year requires prior written authorisation, offering it as a hosted service also requires authorisation, commercial interfaces must prominently display the model name, and outputs may not be used to train or improve other models.
Key facts
- Model: MiniMax H3-Image, a 33.1-billion-parameter diffusion transformer distilled to six sampling steps, with a checkpoint of about 68.65GB
- Speed: MiniMax says a 4K image takes about 10 seconds; no third-party timing or benchmark exists as of Nov 11
- Resolution: About 16 megapixels at every aspect ratio: 4096×4096 at 1:1, 5472×3072 at 16:9, 6272×2688 at 21:9; the API caps output at 20 megapixels
- References: Up to 16 reference images per edit; text-to-image and image editing in one checkpoint
- Channels: Its own API, MiniMax Code and Hailuo AI; weights open on Hugging Face; 50% off above 16 megapixels in the first week
- Licence: MiniMax Community License Agreement; excludes the EU, U.K., South Korea and the U.S.; revenue above $1 million and hosted services need written authorisation
Why it matters
Native 4K output with open weights is real progress, but the licence bars self-hosting in the U.S., EU, U.K. and South Korea, so open weights do not mean globally usable.
Follow HIMIAO Horizon
No sign-up · Unsubscribe any timeNotifications
A ping on your phone or computer when the brief is out. No email needed.
Comments
Related
Today
Nov 12 Thursday- Brief
- GitHubGitHub Weekly Vol.5: 6 Projects Worth a Look This Week
Tested or read — weeklyAll →
- Quiz
- Call
Latest news
All →- Nov 12Kimi K3.1 Pro vs Qwen3.8-Max: Same $2 Input, $4 Apart on Output
- Nov 12NEAR AI Breached via SGLang Flaw, Exposing About 100 Users and 18 Agents
- Nov 12PLA-Linked Chinese University Bought Nvidia H200 Chips Despite US Blacklist
- Nov 12Ten AI Firms Sign Paris Call on AI Safety, Pledging Shutdown Mechanisms Without Penalties
- Nov 12OpenAI Acquires AI Travel Agent Odessia; Founder Joins, Service Shut Down
- Nov 12Meta's Infrastructure Chief Janardhan to Step Down in Early 2027
™