MiniMax released open weights for H3-Image, an image generation model. The system can produce 4,096-by-4,096-pixel images in under 10 seconds. It supports text-to-image generation, reference-based generation, and image editing.
Availability and access
The weights are available on Hugging Face. MiniMax also offers the model through its API Platform, MiniMax Code, and Hailuo AI. Developers can use up to 16 reference images in a single request.
Performance and quality
MiniMax says H3-Image delivers realistic details and strong prompt adherence. The model can reach up to 5 times the speed of mainstream alternatives for 4K output. Standard mode is 1.5 to 2 times faster, while Highspeed mode reaches 3 to 4 times.
Pricing and launch offer
MiniMax prices H3-Image at 50% to 70% of comparable mainstream offerings. During launch week, image generation requests at 4K resolution and above are 50% off the standard price. The company announced the launch on November 11.
MiniMax built H3-Image on the H3 pretraining foundation. The company improved data, reinforcement learning, sampling-step reduction, and deployment optimization. It says the combined changes deliver speed, image quality, and prompt adherence.