Custom workflow for the Clothes Try On LoRA by kingroka.
You have the following input options to choose from:
Enable Outfit Extraction : extracts the outfit from your input image (slot 1) and puts it on a white background;
Enable Outfit Transfer : transfers the outfit already extracted you provide (slot 2) to the character (slot 3);
Both : extracts the outfit from your input image (slot 1) and transfers it to the character (slot 3).
With version 1.2 you now also have the following upscaling options to choose from:
Enable Upscale Extraction : upscales the outfit extracted
Enable Upscale Transfer : upscales the final result
Qwen-Edit being limited to a size of 1 megapixels means you lose a lot of the fine details during the outfit transfer. That is a built-in limitation of the model and the LoRA that no workflow can ever fix, unfortunately. However, upscaling can hopefully restore some of the details. I'm using Real ESRGAN 2xPlus but feel free to experiment with other models (kingroka recommends SeedVR2, but this one requires a LOT of memory so be warned).
With the upscaler, you may want to play with the Denoise Ratio. For realistic images, something between 0.2 and 0.3 should be the sweet spot. For anything digital, you should probably put it back down to 0.
It may seem a bit conterintuitive but I recommend using the regular Qwen-Image Lightning LoRA that gives significantly better results than the Qwen-Image-Edit Lightning (and is the one used by kingroka in his own workflow). Once again, feel free to experiment to find what works best for you and the style you want to achieve.
I'm still trying to figure out how to automatically crop the outfit from the final result. On top of making it instantly usable, it should make the upscaler more efficient as it wouldn't have to deal with the outfit on the left of the image.
This workflow uses quite a few custom nodes so I recommend using ComfyUI Manager to get them all easily.
Description
You now have the following input options :
Enable Outfit Extraction : extracts the outfit from your input image (slot 1) and puts it on a white background;
Enable Outfit Transfer : transfers the outfit already extracted you provide (slot 2) to the character (slot 3);
Both : extracts the outfit from your input image (slot 1) and transfers it to the character (slot 3).
FAQ
Comments (17)
It's interesting, but the overall loss in quality is too much. I haven't yet managed to get an image with Qwen Edit that doesn't distort the entire image.
Unfortunately that's unavoidable with Qwen Edit's 1 megapixel size limit... I've added two upscale nodes here (one for the outfit, one for the final result) to mitigate some of that and the result is fairly decent, I think. If I manage to crop the result in the next version that will also make the final upscaling better.
Automatic cropping can be done by simply cropping the right side according to the size of the target image, and then enlarging only it.
Great job with this workflow !! It creates quality output. I had to change the loader you used, as I am using Qwen_Image_Edit-Q4_K_S.gguf (less than 12GB). I only have 12GB of VRAM and a GTX3060, but I am able to load up everything including 2 loras, my checkpoint, the VAE, and the Clip without getting an OOM. BTW, my regular Quen Edit Workflow has complete metadata support, so I can upload my Quen images to Civitai and they are immediately validated. https://civitai.com/images/99506554
Notice my outfits save to civitai actually show up under "Outfit Extracter" Gallery
I'm glad you like it! All the credit goes to kingroka for the amazing LoRAs, I'm just trying to make them easy to use for people and get experience in crafting workflows in the meantime.
This is very good, just has a few problems, it doesn't recognize shoes or boots, it puts the photo of the clothes next to the character, the output image quality is not good
I'm glad you like it! Shoes and hats are indeed hard to make work, that is a limitation of the Clothes Try On LoRA. The outfit is stitched to the character because that is how Qwen Edit handles multiple images but I'm working on a version that would automatically crop the result. Regarding the image quality, Qwen Edit can only handle 1 megapixels which is pretty limiting but with a decent upscaler (I personally use RaelESRGAN x2) you should still get good results I hope.
I am unable to make it work :/
It's working on my end with a similar photo pulled off Google, do you mind screenshotting the entire workflow? My guess is you haven't downloaded every required model or not plugged them in properly. https://i.imgur.com/Wb5C9lH.png
@JnTenkei Strange but I cant even extract the clothes (I cant get the red shirt from cristiano), look https://mega.nz/folder/fU1jVBTD#eqqx7lQayZUbX2fiOu8viQ
I dont need to add the image in the node above right? I just need to insert critstainao on the left and I can aalready extract his clothes?
@JustPara Version 1.1 fixes a few quirks of the previous workflow, I think it might fix your issue.
i had a similar issue, turns out i was using Qwen Image instead of Qwen Image Edit
Hi again, why the result was not that good (for the T shirt from cristiano?) I put all assets here: https://mega.nz/folder/2NUkFLBC#Es6jTYBOhHGkw6e-1UsW7Q
See if you can make it better?
Qwen Edit has a size limit of 1 megapixels, making you lose some extra fine details unfortunately. Plus, the Clothes Try On LoRA is trained to try and match the art style of the target (in your case, a video game character). Not much you can do, those are just some limitations of the model itself.
I'm still a beginner, I only started to get interested in all this a month ago, but I can see that your scheme is very difficult to understand and the result is worse than in the scheme of the author of the model. I mean, transferring clothes from one person to another. Try to transfer complicated clothes, for example, an outfit from the 17th to the 18th century, a fur coat or something like that.
This workflow is indeed a bit more complicated but it also gives you more control over the output result. If you don't know how to tweak the settings, they're already set to the values recommended by kingroka anyway.
What you're talking about in terms of result has nothing to do with the workflow though, but might just be limitations of the LoRA itself. The workflow doesn't do anything "special" that could mess with the result, it's just a different way of organizing the nodes.
@JnTenkei this may be true, but I can't get an acceptable result in your scheme, which is obtained in a simpler scheme by the author of the model.