This workflow is for creating consistent realistic characters using SDXL and face swapping.
You can see a tutorial on how to use it here:
This workflow is also great for creating a dataset for LoRA training.
Description
FAQ
Comments (10)
INFO: the IPAdapter reference image is not a square, CLIPImageProcessor will resize and crop it at the center. If the main focus of the picture is not in the middle the result might not be what you are expecting. The heads of people in the pictures will be cropped. Especially above the forehead, the area above the forehead is completely cut off How can i do?
If you have access to an Outpaint workflow, I have had some success by growing the vertical or horizontal dimensions to create a square input image. Flux Outpaint is well suited to realistic/photo images, but may need a few generations depending on the original input's complexity/how close the subject is to edge of frame can negatively effect results. GL
dude my dumb ass thought it's a consistent character with a reference image and downloaded whole stuff just to realize it was for generated images!!
sorry for asking this but is there anyway to make consistent characters with a reference image ? I'm pretty noob. tell me my time and downloading all of this models wasn't in vain :(
no full consistency from reference image , still trying to figure that myself as well from any workflow i can havent found one yet. (one on my radar but the owner is charging 60 bucks for it lmfaooo yeah right)
I am getting pretty good results, but only as far as the face detailer, the workflow errors, "zero division error 212 face detailer" cannot workout why. Any guesses?
Im having this issue:
EmptyLatentImage
Storage size calculation overflowed with sizes=[268540049343247, 4, 128, 128]
yeah it took noob 3 week comfyui guy about 6 hours to make this workflow runing :D am , all those anywhere nodes where not connected for some reason i had to figure it out. allso i had memory issues on my 5070ti face detailer hit cura limits, but offloading one model to cpu made it work. Am also it cant make same face from loaded image... and this huge complex workflow with so mutch loras and detailers can simply be replaced by some simple basic qwen image edit workflow. and use reactor to swap faces to keep consistancy.
I have the exact same problem. It doesn't work out of the box.
pls do new tutorial
the tutorial video is private