What problem does it solve?
Changing the outfit on a photo of a specific person usually breaks their identity: img2img re-rolls the face, inpainting needs manual masks, and cloud edit APIs cost money and send images off the machine. This Skill runs a local Krea 2 identity-edit workflow in ComfyUI that swaps clothing while keeping the subject's face, pose, and background intact.
Core Features & Use Cases
- Local outfit transfer: Takes a primary subject image and a secondary outfit reference, and renders the subject wearing the referenced garment with no API key or per-image cost.
- Identity preservation guidance: Documents the critical reference-order rule (subject in slot 1, outfit in slot 2), prompt phrasing that names what to preserve, and fidelity dials (
ref_boost_a) to fix face drift.
- Troubleshooting and tuning: Covers turbo settings (10 steps, cfg 1), raw model settings (40 steps, cfg 3-4), resolution matching to the primary image, and fixes for missing nodes, outdated ComfyUI, and misplaced LoRA files.
- Use Case: A user says "put this jacket on her" with two images; the Skill wires the Krea 2 edit graph, sets the subject as primary and the jacket as secondary, and renders the result locally.
Quick Start
Install the krea2-identity-edit pack, load its workflow.json in ComfyUI, then ask the agent to dress the person in the first image in the outfit from the second image while keeping her face and pose unchanged.
