If you want to see further development please consider supporting me.
🧥 Outfit Transfer Helper LoRA for Qwen Image Edit
The Workflow and the openpose image is uploaded as Training Data.
💡 What It Does
This LoRA is designed to help Qwen Image Edit perform clean, consistent outfit transfers between images.
It works perfectly with the Outfit Extraction Lora, which helps for clothing extraction and transfer.
Pipeline Overview:
🕺 Provide a reference clothing image.
🧍♂️ Use Outfit Extractor to extract the clothing onto a white background (front and back views with the help of OpenPose).
👕 Feed this extracted outfit and your target person image into Qwen Image Edit using this LoRA.
⚠️ Known Limitations / Problems
Footwear rarely transfers correctly — It was difficult to remove footwear when making the dataset.
🧠 Training Info
Trained on curated fashion datasets, human pose references and synthetic images
Focused on complex poses, angles and outfits
🙏 Credits & Thanks
Workflow uses Outfit Extractor Qwen Edit LoRA by KingRoka — many thanks for sharing!
Thanks to KoHya-SS for the amazing training tools that made this LoRA possible.
📦 Included Files
✅ ComfyUI Workflow (.json)
✅ Openpose image (To help extract the outfit with front and back views)
✅ LoRA (.safetensors)
☕ Support
If you enjoy this LoRA or want to support further development, feel free to buy me a coffee!
Description
FAQ
Comments (19)
Works very well, good job with the lora and the workflow
Which workflow?
@dominic1336756 After you click download icon. it will show you "Lora" and "Training Data". It says training data because it is not possible to upload workflow ".json" seperately on civitai. so i upload the workflow in a zip and it shows as a training data. After you download training data. it will include pose_ref openpose image and the workflow to use it with.
Works well most of the times but with lingerie almost all examples I tested the workflow adds a white sweater/cardigan to the extracted clothes images which will then also show up in the final image.
I have not found a way to avoid this
Sorry, Outfit extractor is not my Lora. If I can find free time I will train my own. But you can try to find the same outfit with a white background and bypass the cloth extraction part.
There is something I don't quite understand. There are three images input in the workflow, and it seems that the input "OpenPose" is not working?
Openpose image is for outfit extractor. It has two views of a person: Front and Back. Outfit extractor is not my lora. It is trained to output a single front view of the cloth. So I used openpose input to make it front and back. Then I made dataset using this outfit extractor and the openpose image. I included in the workflow because that's how lora is trained but it might rarely work on a single view of the cloth. You can bypass the openpose image. It will still work but sometimes it includes head and it transfers the cloth as well as with the head.
This works perfectly!
The costume change effect is quite good, but this Lora often introduces strange pixels at the bottom of the result image. Have you encountered this situation?
me too, have you deal with the problem?
@13190695790224 This problem is a defect in lora itself, and it may not be solvable.
所有生成的效果图中边缘都有添加奇怪的像素，有时上下和右边都有，有时右边和下边
我也是，你解决了吗
Hi,bro.Can update the workflow without outfit extract lora?And also can support various high heels
The best workflow and the best lore, thank you so much! It's lacking in detail, of course, because it cuts the quality, but adding SeedVR 2 upscaler to the workflow really helps. There's still a problem with slight inpainting and extra pixels on the sides, but it's not critical. I also added a few XXX lore for myself, and now it's a tool for dressing and undressing everyone in anything. It even handles movie stills and even brings anime clothes into real life.
Another cool thing is that you don't need any promts, just upload 2 photos and sometimes adjust the weight and Lora, and it works. If you're not satisfied with the texture of the new clothing, you can disable Lora's lighting and set 20 steps and 1.5 -2 cfg, and the texture may improve, but according to my experience, artifacts often appear.
Thanks, great work on this!
Note to others: remember to use Qwen Image Edit 2509 and not the original version. I was struggling to understand why in my generations the OpenPose image was getting superimposed over my Reference Clothing image... that was the reason. :)
This Lora is fantastic but it's not working as good with 2511. I use it to go direct from one reference image of a person in the clothes I want to my target person (no clothe extraction required) and it was working great. With 2511 it's also transferring the skin color of the source person in the outfit and not just the clothing. With 2509 it was PERFECT! Any chance you could train a 2511 version?
same issue here