MiniMax releases H3-Image open weights with 4K generation and up to 16 reference images

Announced November 11, H3-Image supports text-to-image, reference-based generation and editing. MiniMax reports 4096×4096 output in under 10 seconds. Deployment and commercial use remain subject to its community license.

Model releasesTboxIOPublished
MiniMaxH3-ImageAI ImageOpen Weights

MiniMax announced H3-Image and released its model weights on November 11, 2026. Built on H3, it combines text-to-image generation, reference-based generation and editing in one checkpoint.

Key capabilities

  • Up to 16 reference images for composing people, objects and scenes.
  • Native output of approximately 16 megapixels, including 4096×4096, without an upscaler.
  • MiniMax reports 4K generation in under 10 seconds. TboxIO has not independently verified this timing.

Access and deployment

Weights are available through MiniMaxAI/MiniMax-H3-Image on Hugging Face. The official repository points to SGLang, vLLM, diffusers and ComfyUI deployment guides. Hosted access is also offered through the MiniMax API, MiniMax Code and Hailuo AI.

License boundaries

The MiniMax-H3-Image Community License excludes the EU, UK, Republic of Korea and United States from its applicable territory. Annualized revenue above USD 1 million, or providing hosted services to third parties, requires separate prior written authorization. Check the complete agreement before use.

TboxIO perspective

High-resolution output and multi-reference control are worth evaluating against consistency, editing success, latency and deployment cost. Review both hardware requirements and license terms before downloading.

This report summarizes official materials. Vendor performance claims are not TboxIO test results.

Sources

Source labels distinguish official claims, third-party tests, and in-house tests.

← Back to AI News