Use the prompt: extract the outfit onto a white background
Notes:
You may have to use a high lora strength, around 1.75 - 2.
I'm still perfecting this lora but it hasn't failed me yet. It just loses some of the finer details. This lora is not required but transferring person to person will sometimes result in more than just the clothes transferring. The hair may also transfer for example. I built the tryon lora around the extracted outfits layed out of a white background so better results are expected with when using this lora.
You can also use nanobanana for this step. A prompt like "remove the man/woman an extract place their outfit on a white background"
ComfyUI Workflow is the official Qwen Edit Workflow template tuned to fp8_e4m3fn + 4steps (but i usually do 6 to 8 steps) with an extra LoadLoraModelOnly node to load the lora. If you'd like a specific LORA trained, send me a message. Commissions are open.
Trained with AI Toolkit:
Description
Use the prompt: extract the outfit onto a white background
Notes:
You may have to use a high lora strength, around 1.75 - 2.
I'm still perfecting this lora but it hasn't failed me yet. It just loses some of the finer details. This lora is not required but transferring person to person will sometimes result in more than just the clothes transferring. The hair may also transfer for example. I built the tryon lora around the extracted outfits layed out of a white background so better results are expected with when using this lora.
You can also use nanobanana for this step. A prompt like "remove the man/woman an extract place their outfit on a white background"
ComfyUI Workflow is the official Qwen Edit Workflow template tuned to fp8_e4m3fn + 4steps (but i usually do 6 to 8 steps) with an extra LoadLoraModelOnly node to load the lora. If you'd like a specific LORA trained, send me a message. Commissions are open.
FAQ
Comments (21)
This is brilliant.
been waiting for this, was struggling too much to get consistent results but with this Lora so far all extracted outfits are perfect!
does it work on any qwen model?
i was trying with instapick and didn't worked well
Great lora! Little tip for usage, can say things like '2-piece' or '3-piece' to separate articles if it's having issues combining them, or you can name each article to clarify as well.
No available workflow to download?
Load my WF from my image in the gallery
@TecKnight I tried I think it failed.
@JustPara All you have to do is drag the .png file onto your comfyUI workspace and it will load the workflow. My workflow does require 1 custom node be installed called sage utils, but it's easy to install
@JustPara this image definitely has the workflow embedded. https://civitai.com/images/99483859
@TecKnight Ohh I have the censor to NOT show those images (I thought I would at least show me the them and let me uncensor them if you know what I mean? But no they did not even show, the only images I had where the oen that re not NSFW and these did not have the workflow I think) See what I mean? haha OK I will try that image thank you man
@TecKnight Hey tec, when i tried to run that image, it said i needed SAGE nodes? And manager did not find any corresponing missing nodes?
This Lora and it's companion Lora work awesomely. It works great on my system and I only have 12GB of VRAM
Could you maybe update to extract the shoes, hair if possible and other items as well like a bag or other objects?
When I try extracting bikinis or lingerie, it extracts it as a longer piece of clothing, like a crop top or a one piece bathing suit. What can I do in cases like this? I followed the guide with the strenght/cfg etc.
Wow super useful lora, thanks so much!! For the next version, could you create one that can show back and side of an outfit? That would make this perfect...
Does this lora work for extracting hats/headwear?
Could you please tell me the size of the dataset you used to train this lora?
Thanks for making this sweet lora! Tried it with nunchaku qwen without luck, but works perfectly with the normal version
Hi I noticed it does not work as well as before with Qwen-Image-Edit-2511. It often keeps the original person in the shot. Is there any way you could update it for Qwen-Image-Edit-2511?
really? I've been having great success with this lora and 2511. I'll look into it
why does output loose the clothes embroidery, prints, patterns quality? any help please
Details
Files
extract-outfit_v3.safetensors
Mirrors
extract-outfit_v3.safetensors
QWEN_OutfitExtractor.safetensors
extract-outfit_v3.safetensors
extract-outfit_v3.safetensors
extract-outfit_v3.safetensors
extract-outfit_v3.safetensors
extract-outfit_v3.safetensors
extract-outfit_v3-mid_1940557-vid_2196307.safetensors
extract-outfit_v3.safetensors
extract-outfit_v3.safetensors
eo3.safetensors