MiniMax H3 Image lands in ComfyUI day-one: 6-step sampling, native 4K output
Lexius2129 · reddit · 2026-11-11
ComfyUI announced native day-one support for MiniMax H3 Image, with a single checkpoint handling both text-to-image and reference-based editing.
Key points
- CFG-distilled: 6-step sampling at guidance 1, no negative prompt needed
- Native high-res output (not upscaled): up to 4096×4096 at 1:1 and 6272×2688 at 21:9
- Multi-reference editing via <Picture 1> / <Picture 2> prompt references in input order
- Up to 4096-token prompts; detailed descriptions work best with the Prompt Enhancer
- Shares the Qwen3-VL text encoder with the H3 video model, so existing H3 users need no extra download
Tips: leave width/height at 0 for auto-sizing (2048×2048 for t2i, image1's size when editing), and connect the VAE when editing so references are anchored as latents. Weights are on Hugging Face, templates ship in ComfyUI, and Comfy Cloud offers a no-download option.
Related event: MiniMax open-sources H3-Image: 16MP 4K images in about 10 seconds(20 posts)→
More from Multimodal
- Seedance 2.5 generates a luxury perfume commercial for a fictional fragrance — aftahi_ai · 2026-11-11
- BaoCut Explains Virtual Streamers Like Shadow Puppets — 3 Steps to Add One to Your Videos — dotey · 2026-11-11
- MiniMax H3-Image 2-step LoRA released on Hugging Face with sample gallery — Illustrious_Row_9971 · 2026-11-11
- Magnific Studios Launches Animated Series 'Before the Show', Shares Behind-the-Scenes — CurieuxExplorer · 2026-11-11
- 'Cara Love the Adventurer': AI-Generated Short Film Trailer Released — noLOOPai · 2026-11-11
- 'Digital Love': An AI-Generated Robot Dating Show Short — BotLordBotflix · 2026-11-11