MiniMax open-sources H3-Image, 4K images in under 10 seconds
The open-weight model combines text-to-image, reference-based generation and editing in one pass, taking up to 16 reference images per call.
MiniMax released open weights for H3-Image on 11 November, a model that generates 4K images, 4096 by 4096 pixels, in under 10 seconds. Text-to-image, reference-based generation and editing all run through the same model, which can take up to 16 reference images in a single call.
MiniMax says H3-Image runs 1.5 to 2 times faster than mainstream alternatives in its Standard mode, rising to 3 to 4 times in Highspeed mode, at 50 to 70% of the price. Weights are up on Hugging Face now, and the model is also live through MiniMax’s own API, MiniMax Code and the Hailuo AI app.
huggingface.co/MiniMaxAI/MiniMax-H3-Image
Why it matters: a fast, open 4K image model changes the economics of anything that generates images in bulk, product mockups, catalogue shots, batch edits. Self-hosting means no per-call API pricing and no vendor queue, and the weights are there to fine-tune or drop into your own pipeline.
The caveat: the speed and price comparisons are MiniMax’s own claims against unnamed “mainstream alternatives”, not an independent benchmark.