ComfyUI: Outfit Changer Workflow | Using A Reference Image

hello there in this video I'll be showing you an exciting workflow that allows you to change a person's outfit based on the reference image so let's start but first you'll want to make sure you got a few things downloaded click on the manager install custom nodes search for IP adapter you want to download this IP adapter plus node also search for clip seg node and download it all right let's see the results from this workflow first and then we'll start to build it from scratch let's load the default workflow from the manager and then we'll start to build from there we will need an inting model for this workflow and I have one already downloaded in my system this model is available on cvti for download this is a good uh inpainting model to be used I will put a link in the description delete this empty latent image node and double click to search for this inpainting model conditioning node connect the positive and negative pipelines through this inating model conditioning node connect the latent to latent image and also the vae now the starting image that we want to work with uh need to be loaded here along with the mask of the clothing area so double click to search for a load image node here and upload the image here and connect the image pipeline to pixels so we need to get the mask of the clothing area from this image for this right click on the image and open in mask editor now let's brush out the area to be masked click on save to note connect the mask and that's it okay let's uh write the text prompt for the clothing that we want to change uh I'm trying black sweater this time okay let's try this okay I think the black sweater worked out quite well so let's uh suppose that uh we need to do this automatically because this may be running at the back end of some web app or something like that so in that case uh we will need another uh node that's called clip seg node uh that can do this automatically for us so double click to search for clip seg node connect the image to this and let's uh clear the mask in The Mask editor because this is going to happen automatically now this clip node autonomously handles the masking process we just need to input the area to be masked in the text field let's add a preview of The Mask here the clip seg node has uh successfully masked the area you can see this is working out in the for the for changing the dress of the image I think the black sweater worked out quite well so now let's load the uh image that we want to use as a reference for the clothing so double click anywhere in this blank space and search for IP adapter so here it is let's try this uh ADV Advanced one because uh I like to have more options to play [Music] with connect the model pipeline through this node so now we have uh connected the model pipeline through this node so now we need to load the image here so double click anywhere on the blank space and search for load image [Music] load the weight and we'll also need to connect the IP adapter and the clip Vision model there's a new node in IP adapter V2 for this uh double click to search for IP adapter unified loader we need to pass the model pipeline through this connect the IP adapter and it will take care of the clip Vision model let's use IP adapter medium strength [Music] here okay let's uh Cube prompt let's see the results the colors are matched quite well but the design doesn't seem to be the same but uh nevertheless this can be useful in exploring new designs let's try some more reference images you can see it's it's not uh exactly the same but uh it's taking some features and styles from the reference image [Music] I think this one worked out quite [Music] well yeah if you're wondering uh why I get these images from uh I don't use copyright images uh go to pixels.com and you can find many free to use images there uh some of the IM that I used in this video are from pixels and some are just uh created by me using AI uh those faces don't exist in the real world but are uh created by just using AI so that's it for this video uh I couldn't upload more frequently because of my freelancing workload but but I'm trying to upload more consistently now to share my knowledge so consider subscribing and stay tuned for the next video

Channel: Zuhaib R

Views: 1,000

Keywords:

Id: 3eSQpAR3DIY

Length: 7min 41sec (461 seconds)

Published: Mon May 06 2024