MiniMax H3-Image
Intro
MiniMax H3-Image is MiniMax's open-weight image generation model built on H3, supporting text-to-image, reference-based generation, and image editing. The company posted the official announcement as @MiniMax_AI on 10 November 2026, and published a research blog, “MiniMax H3-Image: Bringing 4K Image Generation into the 10-Second Era,” dated 11 November 2026. The previous radar slot deferred this name because an official product page was still missing; with the announcement and blog now public, it is recorded here.
MiniMax says H3-Image can generate a native 4K (4096×4096, 16-megapixel) image in under 10 seconds, with richer, less over-smoothed detail, strong prompt adherence among open-weight models, and support for up to 16 reference images. The blog claims up to 5× the speed of mainstream alternatives at full 4K. On the API, Standard mode is described as 1.5–2× as fast as mainstream alternatives and Highspeed mode as 3–4×, with pricing at 50–70% of comparable offerings. For launch week, 4K-and-above generation requests are offered at 50% off the standard price.
The model is available through the MiniMax API, MiniMax Code, and Hailuo AI, with weights on Hugging Face at MiniMaxAI/MiniMax-H3-Image. API docs are at https://platform.minimax.io/docs/guides/image-generation . License scope, regional restrictions, and local-deployment requirements should be read from the official repository and license text, not from third-party summaries.
Highlights:
- Official blog: https://www.minimax.io/blog/minimax-h3-image ;
- Official announcement: https://x.com/MiniMax_AI/status/2120179737695826420 ;
- Hugging Face: https://huggingface.co/MiniMaxAI/MiniMax-H3-Image ;
- API guide: https://platform.minimax.io/docs/guides/image-generation ;
- Distinct from already-listed names such as P-H3-Image and Pika H3 Flex. This term is MiniMax's official H3-Image.
Explain
Why it matters
Open weights, API and product-surface availability, and a claimed sub-10-second native 4K path are a concrete set of product claims in image generation. MiniMax is also documenting multi-reference input (up to 16 images), editing, and a launch-week discount, which makes this a formal public release rather than community early-access chatter alone.
Speed-up figures, quality, and prompt-adherence claims come from MiniMax. We have not independently benchmarked them. Weights were already visible on Hugging Face and discussed before this slot; the official post and blog are what turn the name into a public product launch for this radar.
Limits
“Under 10 seconds at 4K” and “up to 5× faster” are company figures; wall-clock time depends on hardware, resolution, and Standard versus Highspeed mode. The open-weight license may include regional or other restrictions — read it before commercial or local use. Do not treat third-party fine-tunes or quantizations such as P-H3-Image as the same name.
Usage
- Read the official blog https://www.minimax.io/blog/minimax-h3-image and the announcement https://x.com/MiniMax_AI/status/2120179737695826420 ;
- Get weights from https://huggingface.co/MiniMaxAI/MiniMax-H3-Image, or call the model via https://platform.minimax.io/docs/guides/image-generation, MiniMax Code, or Hailuo AI ;
- Keep this distinct from P-H3-Image, Pika H3 Flex, and other H3-related names ;
- Read the official license and any regional terms before using the open weights; launch-week 4K discounts follow the official pricing page.
Advertisement