FLUX Kontext — Character-Consistent Image Editing (ComfyUI Workflow)
A ComfyUI node graph (not model weights) for editing clothing, background, or style of an image while preserving character identity, using FLUX.1 Kontext Dev as full visual context (not mask-based inpainting).
When to use it
Best for broad semantic changes (full outfit, background, style). For precise, localized touch-ups, combine with traditional inpainting rather than replacing it.
Tested parameters
Denoise between 0.55 and 0.70 gives controlled changes without breaking character identity — outside that range the character degrades or the edit barely applies.
Requirements
- 12GB VRAM recommended (16GB+ unrestricted).
- Viable on 8GB with a quantized GGUF Q4/Q5 build, offload enabled, resolutions ≤768px.
- Time: 40-120 seconds per image.
- Nodes:
UNETLoader, DualCLIPLoader, VAELoader, Load Image, Kontext Image Encode, Flux Guidance, KSampler, VAEDecode, Save Image.
Full write-up
License
CC BY 4.0 — please attribute by linking back to the original article. This repo contains only the ComfyUI workflow JSON (node graph); it does not bundle or redistribute any model weights.
Files
flux-kontext-character-edit-comfylab.json — the workflow, load directly in ComfyUI.
