Qwen Image 2.1 Outfit Swap

Dress the person in your photo in the outfit from a second picture. Nothing to mask or paint. The face, the hair, the pose and the background stay where they are.

Why use this workflow

You can do an outfit swap with a plain two-picture Qwen edit and a typed request. This workflow adds four things:

• It describes the outfit for you. A vision model looks at the outfit picture and writes the request. You type nothing.

• It fits the outfit to your picture. No legs in your picture: only the top half goes on. No feet: no shoes. Without this check, Qwen tries to fit trousers and shoes into a close-up and zooms out. Before I added it, about half of my close-up swaps came out wrong.

• It keeps your picture in place. It runs with my Outfit Swap Consistency LoRA. On 16 test swaps the result sat 0.1 px from the original picture. Without the LoRA it sat 3.6 px off, with a softened face and a repainted background.

• It takes the old clothes off. "Replace everything they wear." is added to the request, so old jeans or tights do not stay on under the new outfit.

How to use it

1. Load your image.

2. Load a picture of the outfit. Optional, but it helps: drag the blue handles to crop around the clothes you want.

3. Run. The request sent to Qwen shows in Show Text, above the result.

For a group photo, name the person in Who to dress, like "the woman on the left". To dress two people, run it twice and load the first result as your picture for the second run.

For a close-up photo, crop the outfit picture to its top half. A full-length outfit on a close-up can zoom the picture out or leave the head too big.

The outfit gets described for you, and your picture decides how much of it goes on. No legs in your picture: only the top half. No feet: no shoes. To type the outfit yourself, switch Describe the outfit for me off and use What to take from it.

What happens to the old clothes

Add to the request starts with "Replace everything they wear.", so the old outfit does not stay on under the new one. To keep their own shoes or tights, change it to "Keep their own shoes and legwear."

Shoes are the stubborn part. Sometimes a pair stays on either way. Run it again.

Faster, optional

The Viggle Turbo row is switched off. Download the file below, switch the row on and set Steps to 8. A run takes about half the time. The picture still stays in place. Plain fabric can come out a little grainy.

What you need

• AusBoss nodes 2.5.1 or newer. In ComfyUI-Manager, search "AusBoss". Already have them? Update first.

• ComfyUI 0.38 or newer.

• The Qwen Image 2.1 models:

  • qwen_image_2.1_int8_convrot.safetensors → models/diffusion_models · 7.26 GB

  • qwen3vl_8b_int8_convrot.safetensors → models/text_encoders · 9.35 GB

  • qwen_image_2.1_vae_bf16.safetensors → models/vae · 676 MB

• The Outfit Swap Consistency LoRA: qwen-image-2.1-outfit-swap.safetensors → models/loras · 159 MB

Optional, for the speed row: Qwen-Image-2.1-viggle-turbo-v0.3-6step-lora-r128.safetensors → models/loras · 680 MB. The workflow runs without it as long as its row stays off.

Tips

• Any outfit picture works: a person wearing it, a flat lay or a product shot. Crop it to the part you want.

• Name people by where they stand or by their hair, not by their clothes. The clothes are what changes.

• It keeps the pose. An outfit that needs a different pose will not get one.

• Something lying over the clothes can go with them, like an earphone cable over a jacket.

• About 23 seconds a run on an RTX 5090, 13 with the speed row.

Credits

Qwen Image 2.1 is by Alibaba Qwen, under the Qwen Research License: non-commercial use only. The optional speed LoRA is Viggle Turbo v0.3 by Viggle, under the Qwen Research License: research use only.

v1.1 (2026-10-04): the outfit picture loads through a crop tool, so you can crop it to the clothes you want.

v1.0 (2026-10-04): first release.

Questions or bugs: GitHub issues or the comments here. More from me on X @Zanzibased and GitHub.