Workflows / MiniMax H3
Premium workflow · new
H3 Image Edit — up to 8 references
Edit a picture with other pictures, in 8 steps.
P-H3-Image turns MiniMax H3 into an image editor: Picture 1 is what you edit, Pictures 2–8 are references — an outfit, a product, a face, a style. Say what to take from which picture. Runs as a 2 GB LoRA on the H3 checkpoint you already have, no CFG, 8 steps.
One-time purchase. Lifetime updates. Included in Pro.
You get
- H3 Image Edit (P-H3) workflow built like Pruna's own edit graph: Bypass LoRA, euler and the prompt upsampler that turns a plain request into the edit brief
- One reference wired; add more as image_3.. with a <Picture N> block each
- Tested on outfit, background and style edits
You need
- The Gmanski RunPod template with ComfyUI 0.44+
- 48 GB VRAM for 4 MP
Model packs · Edit an image · Characters · Text to image
minimax-h3-flowminimax-h3-pimageminimax-h3-pimage-upsamplerOn the template: gmanski-packs install minimax-h3-flow minimax-h3-pimage minimax-h3-pimage-upsampler
Pairs well with.
MiniMax H3newCharacter Video Swap — one photo, any performance
A guide video gives the moves. Your photo gives the person.
Krea 2newBulk Character Swap — Dynamic LoRA
A folder of photos in. Your character in every one — and the right LoRAs picked per picture.
Krea 2newBulk Reference Swap — no training
One reference photo. Every picture in a folder, re-cast — no LoRA needed.