MiniMax Open-Sources H3 Image: 4K in 10-20 Seconds, 16 Reference Images, Best-in-Class Instruction Following
新智元 · wechat · 2026-11-10
MiniMax released and immediately open-sourced H3 Image, its new image generation model, with API access live across MiniMax Code, the open platform and Hailuo AI. It generates 4096×4096 (16MP) images in 10-20 seconds — 5x faster than mainstream tools — supports up to 16 reference images and 7500-character prompts, and claims SOTA among open models on instruction following: CAD-style blueprints, parts explosion diagrams with correctly numbered callouts, 48.1% accuracy on American Sign Language (vs Moss's 20%), and consistent character/outfit/scene matching across 16 refs. Built on H3's unified-context pretraining plus RL-based post-training, step compression, prompt enhancement and deployment optimization. Licensing follows H3's model: free under $20M annual revenue, separate licensing above or for US/EU/UK/Korea deployment.
More from Multimodal
- Microsoft ships MAI-Image-2.6 Flash: 2.8x faster than GPT-Image-2-Medium, now Bing's default — JordiRib1 · 2026-11-11
- Animate real places in 3D with Claude and free map data from AWS and Overture — drbronk · 2026-11-11
- Fable 5.2 builds its own engine: 6cm FLIP wave sim and rigged galloping horses — AymericRoucher · 2026-11-11
- Redditors reproduce the disputed Fable 5.2 demo video, cost ~$20 in API tokens — timeport-0 · 2026-11-11
- Dev builds an AI paintbrush that repaints her first drawing in 55,185 strokes — repligate · 2026-11-11
- Minimax H3 Image hands-on: superb edits, but small repeated edits ruin images — k_from_HyperDraw · 2026-11-11