Natural Female Faces (CLIP-L)
This CLIP-L model was trained to address an issue with captioning of RESNET-50, COCO and other datasets. Those datasets identified any average female face as masculine or male.
Example Images use this PONY model
Description
FAQ
Comments (4)
"This CLIP-L model was trained to address an issue with captioning of RESNET-50, COCO and other datasets. Those datasets identified any average female face as masculine or male." - Assuming this is correct, this CLIP model could be a big deal. I'll be sure and review it after testing. Good work.
The clip works with SD3.5L, thanks
What are the requirements for using it with SDXL? I get this error message with 832x1216 and other resolutions:
"mat1 and mat2 shapes cannot be multiplied (1x2304 and 2816x1280)"
I am on ComfyUI, using 12GB RAM.
The faces with this clip-L are exactly the same as with default one. (SDXL/3/3.5), other details are changing a little but faces are same.
So far it does nothing.
Why not do at least basic tests before posting such content to prove that it actually works as you claim?
https://prnt.sc/wIXjXqOraYwk