ComfyUI workflow swaps 1-4 characters in one pass using MiniMax H3
GkoChan · reddit · 2026-11-09
Reddit user GkoChan shared a ComfyUI workflow built on MiniMax H3 that swaps 1-4 characters in a single image in one run; the JSON is available on Civitai and Dropbox.
How it works:
- Worker nodes handle logic, comparison and prompt assembly; the author recommends pairing the main model with qwen3vl8bint8convrot as an auxiliary CLIP
- Settings node: set character count (1-4), reference image size, optional fizgig toggle (off gives 20% speedup), and a "Reinforce relight" option that auto-adds relighting prompts so subjects blend naturally
- Additional overall instruction field supports extra edits like background changes, pose fixes
- Characters node takes reference images with optional notes ("she's a minigirl, downsize her", "age her up")
- Preview nodes let you verify subject identification and character ordering, and inspect the final assembled prompt
Requires custom nodes: ComfyUI-Fizgig-H3-Still and rgthree's pack.
More from Multimodal
- Local pipeline showcase: MiniMax H3 refMod generations from reference image to 4K upscale — edisson75 · 2026-11-13
- Gaussian splats may be the best way to capture how 3D displays like LKG Portrait look IRL — nptacek · 2026-11-13
- AI-generated 'not canon' clip of Schwarzenegger driving with Michael Jordan — alanskimp · 2026-11-13
- H3-Image API gotcha: server keeps billing on client timeout — set max_retries=0 — xiaohu · 2026-11-13
- MiniMax open-sources H3-Image: native 16MP output, up to 16 reference images — xiaohu · 2026-11-13
- Daily AI painting made with GPT-6 Images 2.6 shows painterly winter village — DeryaTR_ · 2026-11-13