MiniMax H3 Image lands in ComfyUI day-one: 6-step sampling, native 4096×4096 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 (connect images for an edit workflow, leave them off for t2i).
Highlights:
- 6-step sampling: CFG-distilled, guidance 1, no negative prompt, essentially nothing to tune.
- Native high-res output (generated, not upscaled): 4096×4096 (1:1), 4736×3552 (4:3), 5024×3360 (3:2), 5472×3072 (16:9), 6272×2688 (21:9).
- Multi-reference editing: attach several images and reference them as <Picture 1>, <Picture 2> in prompt order.
- Long prompts: up to 4096 tokens; detailed descriptions work best, with a Prompt Enhancer available.
- Shared text encoder: uses the same Qwen3-VL encoder as the H3 video model, so existing H3 users skip the download.
Practical node tips: leave width/height at 0 for auto sizing (2048×2048 for t2i, image1's size when editing); image1 is the image being edited, other slots are references; connect the VAE when editing so references are anchored as latents rather than only read by the text encoder. Update ComfyUI, grab weights from Hugging Face (Comfy-Org/MiniMax-H3-Image), or use templates / Comfy Cloud without downloading.
More from Multimodal
- Opus 5.5 vs Fable 5.2 side-by-side: Fable impresses on diversity and scene understanding — FinanceYF5 · 2026-11-11
- All-in-one H3 Image workflow with Pose Studio for easy character posing — foxdit · 2026-11-11
- Opus 5.5 generates a restrained, watchable line-art animation of streets and rivers — FinanceYF5 · 2026-11-11
- AI-generated holographic cyberpunk shop scene impresses with detail — Zard-Oz · 2026-11-11
- Seedream 5.5 edits hold character identity across 30 frames in user test — aziz4ai · 2026-11-11
- Tsinghua and Shengshu's APEX replaces fake-score nets with condition shifting for faster streaming video generation — Tsinghua · 2026-11-11