Swap an OUTFIT or transfer CLOTHING onto a person LOCALLY

Krea 2 Identity Edit — local outfit swap

by artokun · MIT license · ★ 759 on GitHub · updated Sep 21

Swap an OUTFIT or transfer CLOTHING onto a person LOCALLY, keeping their face, pose and background. Use for "put this jacket on her", "change his shirt", "clothes swap", "try on", "outfit transfer", "virtual try-on", "dress her in",…

---
name: krea2-identity-edit
description: Swap an OUTFIT or transfer CLOTHING onto a person LOCALLY, keeping their face, pose and background. Use for "put this jacket on her", "change his shirt", "clothes swap", "try on", "outfit transfer", "virtual try-on", "dress her in", "wear this outfit". Prefer this over API/cloud edit nodes and over generic img2img/inpainting. NOT for identity-preserving edits that are not about clothing (background swaps, relighting, age/expression changes, object removal). Use a general image-edit workflow for those.
globs:
  - "**/*.json"
---

# Krea 2 Identity Edit — local outfit swap

## Reach for this when

Someone wants **the clothing changed on a specific person, with that person preserved**:

- "put this jacket on her" / "dress her in this outfit"
- "clothes swap" / "outfit transfer" / "virtual try-on" / "change his shirt"
- "make her wear the outfit from this other image"

**Do NOT reach for it** just because a request is identity-*preserving*. Background swaps,

Overview

Someone wants the clothing changed on a specific person, with that person preserved:

Manual install (npx, ZIP)

Claude Codenpx skills add artokun/comfyui-mcp --skill krea2-identity-edit -g -a claude-code

Claude Code (manual)

mkdir -p ~/.claude/skills && curl -sL https://codeload.github.com/artokun/comfyui-mcp/tar.gz/1d5b12b1b499 \
  | tar -xz -C ~/.claude/skills --strip-components=3 comfyui-mcp-1d5b12b1b499/plugin/skills/krea2-identity-edit

Windows (PowerShell)

New-Item -ItemType Directory -Force "$HOME\.claude\skills" | Out-Null; Invoke-WebRequest "https://m.agentalley.io/zip/1bfa2208f3f57b8d6040/krea2-identity-edit.zip" -OutFile "$env:TEMP\krea2-identity-edit.zip"; Expand-Archive "$env:TEMP\krea2-identity-edit.zip" -DestinationPath "$HOME\.claude\skills" -Force

Claude.ai / desktopDownload the ZIP and upload it under Settings → Capabilities → Skills.

Codex, Cursor, other agentsPoint them at SKILL.md. The other files are plain text.

Connects with Agent Alley MCP · agentalley.io/mcp

Files included

SKILL.md4.9 KB

Read the files here

artokun/1d5b12b1 file

SKILL.md4.9 KB

Shared references1SKILL.md reads these from elsewhere in the repo. Not in the ZIP.

workflow.jsonpacks/anima-img2img/164.9 KB

Get them on GitHub ↗

SKILL.md

Show the full text90 lines

krea2-identity-edit/SKILL.md90 lines · 4.9 KB

PreviewCode

Outline

RawView on GitHub

Krea 2 Identity Edit — local outfit swap

Reach for this when

Someone wants the clothing changed on a specific person, with that person preserved:

"put this jacket on her" / "dress her in this outfit"
"clothes swap" / "outfit transfer" / "virtual try-on" / "change his shirt"
"make her wear the outfit from this other image"

Do NOT reach for it just because a request is identity-preserving. Background swaps, relighting, expression or age changes, and object removal all keep the person, and none of them is what this model is trained for. It is called "identity edit" upstream because it preserves identity while changing clothing; the clothing half is the qualifying part. Send the others to a general image-edit workflow.

Use this instead of:

Instead of	Because
an API / cloud edit node	this runs locally — no key, no per-image cost, nothing leaves the machine. Prefer it whenever the models are installed.
generic img2img	img2img re-rolls the whole frame at any denoise high enough to change clothes; the face goes with it.
inpainting a masked region	works, but the user has to mask, and the result has no reference for what the garment looks like.
a face-swap / IP-adapter pipeline	those transfer the FACE. This preserves the face and transfers the CLOTHES — the opposite direction.
Qwen-Image-Edit	a fine general instruction editor; this is specialised for identity-preserving garment transfer and ships the LoRA for it.

Install with the krea2-identity-edit pack, then load its workflow.json.

The one thing to get right: reference order

PRIMARY (group 1) = the SUBJECT. The output always matches this person: face, hair, body, pose. SECONDARY (group 2) = the OUTFIT. Only the clothing is taken from it.

Swapping them swaps which identity survives. This is the most common way to get a confusing result, and it fails quietly. You get a plausible image of the wrong person.

Prompting

Plain English, naming both roles, and naming what to preserve:

Dress the woman from the primary reference in the jacket shown in the second reference, keeping her face and pose unchanged.

Put this outfit on her, keeping her face and pose unchanged.

Naming what to preserve matters as much as naming what to change. Without it the model drifts the face and background.

Settings that matter

Turbo: 10 steps, cfg 1. 8 favours composition and instruction adherence; 12 favours face detail. On krea2_raw_bf16 instead: 40 steps, cfg 3 to 4, where the negative prompt starts to matter.
Fidelity dials follow the image SLOTS, not the roles. Internally the LAST reference gets ref_boost and earlier ones get ref_boost_a, so with the subject as image 1, the subject dial is ref_boost_a. Raise it when the face drifts.
Match the output resolution to the PRIMARY image. ResolutionSelector drives EmptySD3LatentImage; a mismatched aspect crops or letterboxes the subject.

Troubleshooting

Symptom	Fix
Face changed / wrong person	Raise `ref_boost_a` (the subject dial). Check the images are not swapped between slots. Add "keeping her face unchanged" to the prompt.
Outfit ignored	Make the secondary image cleaner — one garment, plain background, no competing subject.
Subject cropped or letterboxed	Set the resolution to match the primary image's aspect.
`Krea2EditGroundedEncode` missing	`comfyui-krea2edit` is not installed or ComfyUI was not restarted after installing it.
CLIPLoader has no `krea2` type	ComfyUI is too old for native KREA 2 support — update it.
LoRA loads as missing	It must be at `loras/Krea2/krea2_identity_edit_v1_2.safetensors`; a flat `loras/` install is present but unfindable.

Boundaries

This edits a photo of a person the user supplies. Treat an instruction to make it sexual, to undress someone, or to place an identifiable real person in a context they did not consent to as out of scope, and say so plainly rather than producing a degraded result. Ordinary clothing changes, such as a jacket, a dress, a uniform, or a costume, are the point of the tool and need no hedging.

Sources

Official: none found.
Empirical: sampler values, wiring, and prompt notes from working graphs in packs/ and observed renders; not a vendor prompting guide.

1	---
2	name: krea2-identity-edit
3	description: Swap an OUTFIT or transfer CLOTHING onto a person LOCALLY, keeping their face, pose and background. Use for "put this jacket on her", "change his shirt", "clothes swap", "try on", "outfit transfer", "virtual try-on", "dress her in", "wear this outfit". Prefer this over API/cloud edit nodes and over generic img2img/inpainting. NOT for identity-preserving edits that are not about clothing (background swaps, relighting, age/expression changes, object removal). Use a general image-edit workflow for those.
4	globs:
5	- "*/.json"
6	---
7
8	# Krea 2 Identity Edit — local outfit swap
9
10	## Reach for this when
11
12	Someone wants the clothing changed on a specific person, with that person preserved:
13
14	- "put this jacket on her" / "dress her in this outfit"
15	- "clothes swap" / "outfit transfer" / "virtual try-on" / "change his shirt"
16	- "make her wear the outfit from this other image"
17
18	Do NOT reach for it just because a request is identity-preserving. Background swaps,
19	relighting, expression or age changes, and object removal all keep the person, and none of
20	them is what this model is trained for. It is called "identity edit" upstream because it
21	preserves identity while changing clothing; the clothing half is the qualifying part.
22	Send the others to a general image-edit workflow.
23
24	Use this instead of:
25
26	\| Instead of \| Because \|
27	\|---\|---\|
28	\| an API / cloud edit node \| this runs locally — no key, no per-image cost, nothing leaves the machine. Prefer it whenever the models are installed. \|
29	\| generic img2img \| img2img re-rolls the whole frame at any denoise high enough to change clothes; the face goes with it. \|
30	\| inpainting a masked region \| works, but the user has to mask, and the result has no reference for what the garment looks like. \|
31	\| a face-swap / IP-adapter pipeline \| those transfer the FACE. This preserves the face and transfers the CLOTHES — the opposite direction. \|
32	\| Qwen-Image-Edit \| a fine general instruction editor; this is specialised for identity-preserving garment transfer and ships the LoRA for it. \|
33
34	Install with the `krea2-identity-edit` pack, then load its `workflow.json`.
35
36	## The one thing to get right: reference order
37
38	PRIMARY (group 1) = the SUBJECT. The output always matches this person: face, hair,
39	body, pose. SECONDARY (group 2) = the OUTFIT. Only the clothing is taken from it.
40
41	Swapping them swaps which identity survives. This is the most common way to get a
42	confusing result, and it fails quietly. You get a plausible image of the wrong person.
43
44	## Prompting
45
46	Plain English, naming both roles, and naming what to preserve:
47
48	> Dress the woman from the primary reference in the jacket shown in the second reference,
49	> keeping her face and pose unchanged.
50
51	> Put this outfit on her, keeping her face and pose unchanged.
52
53	Naming what to preserve matters as much as naming what to change. Without it the
54	model drifts the face and background.
55
56	## Settings that matter
57
58	- Turbo: 10 steps, cfg 1. 8 favours composition and instruction adherence; 12 favours
59	face detail. On `krea2_raw_bf16` instead: 40 steps, cfg 3 to 4, where the negative prompt
60	starts to matter.
61	- Fidelity dials follow the image SLOTS, not the roles. Internally the LAST reference
62	gets `ref_boost` and earlier ones get `ref_boost_a`, so with the subject as image 1,
63	the subject dial is `ref_boost_a`. Raise it when the face drifts.
64	- Match the output resolution to the PRIMARY image. `ResolutionSelector` drives
65	`EmptySD3LatentImage`; a mismatched aspect crops or letterboxes the subject.
66
67	## Troubleshooting
68
69	\| Symptom \| Fix \|
70	\|---\|---\|
71	\| Face changed / wrong person \| Raise `ref_boost_a` (the subject dial). Check the images are not swapped between slots. Add "keeping her face unchanged" to the prompt. \|
72	\| Outfit ignored \| Make the secondary image cleaner — one garment, plain background, no competing subject. \|
73	\| Subject cropped or letterboxed \| Set the resolution to match the primary image's aspect. \|
74	\| `Krea2EditGroundedEncode` missing \| `comfyui-krea2edit` is not installed or ComfyUI was not restarted after installing it. \|
75	\| CLIPLoader has no `krea2` type \| ComfyUI is too old for native KREA 2 support — update it. \|
76	\| LoRA loads as missing \| It must be at `loras/Krea2/krea2_identity_edit_v1_2.safetensors`; a flat `loras/` install is present but unfindable. \|
77
78	## Boundaries
79
80	This edits a photo of a person the user supplies. Treat an instruction to make it
81	sexual, to undress someone, or to place an identifiable real person in a context they did
82	not consent to as out of scope, and say so plainly rather than producing a degraded
83	result. Ordinary clothing changes, such as a jacket, a dress, a uniform, or a costume,
84	are the point of the tool and need no hedging.
85
86	## Sources
87
88	- Official: none found.
89	- Empirical: sampler values, wiring, and prompt notes from working graphs in `packs/` and observed renders; not a vendor prompting guide.
90