I tried the Cross-Image Try-On LoRA that runs on FLUX.1 Kontext. You keep one image with the outfit on the left and the target person on the right. The LoRA reads both sides and tries to move the clothing style from left to right. It is clearly marked as experimental, so expect misses along with the hits.
What changed in v0.2
- New v0.2 safetensors is live at about 344 MB
- Fresh workflow JSON for v0.2 is included
- Dataset expanded to 53 paired images
- Two trigger prompts now, one for full-body and one for upper-body
- Training done with AI-ToolKit at 5e-5, around 65k steps
- These updates landed within the last day on the model repo, and the README lists the exact changes.
How to Run It
The easiest way is with ComfyUI, and the creator already shared a workflow file here: workflow JSON.
Quick start in ComfyUI
- Load FLUX.1-Kontext-dev as the base model.
- Load the Cross-Image Try-On LoRA v0.2.
- Open the provided workflow Flux Kontext LoRA.json and drop in your inputs.
- Prepare a single side-by-side image. Outfit reference on the left. Target person on the right.
- Use one of the trigger prompts below and add a few clear clothing words.
- ComfyUI workflow and usage notes are in the repo.
Trigger prompts
- Full-body
Change all clothes on the right to match the left - Upper-body only
Change the upper body clothing on the right to match the left
The author calls these out as the reliable starters. Add explicit details like right woman wearing a red dress to steady the result.
Prompts that worked for me
Use a simple pattern. Keep the subject specific, and say the color or garment clearly.
- Change all clothes on the right to match the left, right woman wearing a white shirt and blue jeans
- Change the upper body clothing on the right to match the left, right man wearing a black bomber jacket
- Change all clothes on the right to match the left, right woman wearing a red kurta with gold trim
These follow the repo guidance about explicit clothing phrases.
Download and files
Grab the safetensors from the Files tab. The current file on the page is:
Cross-Image Try-On Flux Kontext_v0.2.safetensors at 344 MB.
FAQ style
What kind of images work best
Regular photos where the outfit is easy to read. Keep the left side as the reference outfit and the right side as the person.
How specific should the prompt be
Name the garment and the color. Short and clear works better than long and vague.
What usually fails
Illustration or anime, heavy occlusion, and shots with glare or blur.
