MiniMax H3-Image has arrived as an open-weight model that generates and edits native images of about 16 megapixels in approximately 10 seconds. MiniMax announced the release of MiniMax-H3-Image, based on the H3 family, without needing an upscaler for resolutions like 4096x4096 or 6272x2688.
What happened?
Released on November 10, 2026, MiniMax H3-Image unifies text-to-image generation and editing with up to 16 reference images in a single 33.1-billion-parameter checkpoint. It is distilled to just 6 sampling steps with no classifier-free guidance and uses the H3 DiT architecture (without audio layers) with the Qwen3-VL text encoder.
Weights are available on Hugging Face under the MiniMax-H3-Image Community License Agreement. The main model is about 68.65 GB in BF16, with native ComfyUI support, plus APIs, MiniMax Code, and Hailuo AI. During launch week there is a 50% discount on 16MP+ generations.
Why this matters
High-resolution open-weight models like MiniMax H3-Image accelerate access to professional-quality image creation tools for developers, artists, and indie studios. Multi-reference editing and practical 4K/16MP speed lower barriers in game art, concept work, and content production.
As a leading Chinese model, MiniMax H3-Image strengthens global competition in generative image AI, with a focus on realism and precise instruction following.
In practice
Users can download the MiniMax H3-Image weights and run them locally (with an additional text encoder) or use MiniMax APIs. Official ComfyUI templates are available for T2I and editing. The license has regional restrictions (excluding the US, EU, UK, and South Korea in some cases) and revenue limits for commercial use.
Sources: Hugging Face, Official MiniMax blog.
Image credit: MiniMax | Source: Official MiniMax blog
By Grok
