An all-in-one ComfyUI workflow for making still images with MiniMax H3 (a video model): text-to-image and image editing in one graph, 4 to 16 megapixels, 6 steps, with automatic prompt expansion.
What's inside
Prompt box, color adjustment, function switches (SeedVR2 upscale, detail sharpening) and resolution presets (4–16 MP).
- Editing LoRA: the biggest quality jump. It also works in non-editing mode, with a slightly Eastern look.
- Model area: keep the 6-step sigmas set in the workflow. The official numbers are not accurate.
- Prompt expansion: don't change the two system templates. If your prompt is already finished, switch expansion off; it's much faster.
- Reserved box for trigger words and special terms; the expander won't rewrite them.
- Up to 6 input images (images or video, same as MiniMax H3). Turn all inputs off for pure text-to-image.
- SeedVR2 upscale: keep it on with the multiplier at 1. It adds detail, no enlargement.
- Four example prompts included: colored-pencil character, character sheet, xianxia costume, poster (text test).
Tip: plain-language prompts are fine; the built-in expander rewrites them for H3.
Video walkthrough: YouTube channel YJ LAB, EP06.

