Choose Face Model - type

[kalle07]

question

which kind of model?
self trained (face)loras ? iv trained them with kohya. which tool i can convert to fp16, float ?!?

all models i tryed

error1
...
*** Error running postprocess_image: D:\stable-diffusion\webui\extensions\sd-webui-reactor\scripts\reactor_faceswap.py
Traceback (most recent call last):
File "D:\stable-diffusion\webui\modules\scripts.py", line 856, in postprocess_image
script.postprocess_image(p, pp, *script_args)
File "D:\stable-diffusion\webui\extensions\sd-webui-reactor\scripts\reactor_faceswap.py", line 450, in postprocess_image
result, output, swapped = swap_face(
File "D:\stable-diffusion\webui\extensions\sd-webui-reactor\scripts\reactor_swapper.py", line 555, in swap_face
source_face_model = [load_face_model(face_model)]
File "D:\stable-diffusion\webui\extensions\sd-webui-reactor\scripts\reactor_helpers.py", line 195, in load_face_model
face[k] = f.get_tensor(k).numpy()
TypeError: Got unsupported ScalarType BFloat16

or error2
...
*** Error running postprocess_image: D:\stable-diffusion\webui\extensions\sd-webui-reactor\scripts\reactor_faceswap.py
Traceback (most recent call last):
File "D:\stable-diffusion\webui\modules\scripts.py", line 856, in postprocess_image
script.postprocess_image(p, pp, *script_args)
File "D:\stable-diffusion\webui\extensions\sd-webui-reactor\scripts\reactor_faceswap.py", line 450, in postprocess_image
result, output, swapped = swap_face(
File "D:\stable-diffusion\webui\extensions\sd-webui-reactor\scripts\reactor_swapper.py", line 600, in swap_face
source_face = sorted(source_faces, key=lambda x: x.bbox[0])[source_faces_index[0]]
File "D:\stable-diffusion\webui\extensions\sd-webui-reactor\scripts\reactor_swapper.py", line 600, in
source_face = sorted(source_faces, key=lambda x: x.bbox[0])[source_faces_index[0]]
TypeError: 'NoneType' object is not subscriptable

[kalle07]

i see selfcreat in third tab ...

btw why its not compatible to faceswaplab ? ;)

[Gourieff]

Faceswaplab saves only the "embedding" part of the whole face model (that contains much more tensors)
ReActor saves the whole model with all tensors it has

[kalle07]

i see ...

for blend, what is the best option for images with the same face, Mean Median or Mode ?
the weight for face restauration dont work for gfpgan ?
because codeformer often makes blue eyes...

you know Topaz Photo AI, i know its a big program, but if i process there the image without restauration its a better ;)

[kalle07]

another question...
can you please explane for many faces

"Source Image (above):
Comma separated face number(s); Example: 0,2,1"
(its which face would be replaced but how?)

is it maybe more easy (left middle right) or (detection Threshold 1,2,3)

[kalle07]

if i raise up Threshold only very very very small faces will be restored not the bigest and i am shure that is 0.8 and i set Threshold to 0.8