Original/Enhanced image blend ratio ? #231
-
|
Original/Enhanced image blend ratio what this slider do ? I have slide up and down for testing like im barely seem noitce any different ? |
Beta Was this translation helpful? Give feedback.
Replies: 4 comments 10 replies
-
|
+1 I would also like to know that. Or what is a good Setting to Combine Max Face Similarity Threshold with Original/Enhanced image blend ratio |
Beta Was this translation helpful? Give feedback.
-
|
Image blend ratio is only being used when you turned on one of the face enhancers (GFPGAN, Codeformer, etc.). Original image means the one straight after face swapping (which has a poor resolution of 128x128 max). The ratio defines how much of the original image is blended (mixed) into the enhanced one. So a setting of 0.5 would mean half of the original image and half of the enhanced. Reasoning behind this is, the face enhancement modifies a lot of the persons likeness and looks oftentimes very artificial. |
Beta Was this translation helpful? Give feedback.
-
|
Does decreasing the percentage of enhanced image (thus increasing the ration in favor of original image) improve the overall processing speed? |
Beta Was this translation helpful? Give feedback.
-
|
Not in my experience. But if so, that would be a very small, not really noticeable percentage. To my knowledge, if you choose an enhancer, the image gets enhanced fully first. Blending with the original image is a second stage. |
Beta Was this translation helpful? Give feedback.
-
|
Exactly like that. The enhancement process always runs and it will always be blended together even at a value of 0.0. There will be no speed difference. |
Beta Was this translation helpful? Give feedback.
-
|
I'll add the next Question here: What does Offset Face Top ?? |
Beta Was this translation helpful? Give feedback.
-
|
It offsets the beginning of the original face on the target face. Useful to skip low hairlines for example. The face images are upscaled to 512x512, meaning if you set the slider to 256, you will probably see the eyes and half of the nose of the target and the rest should be from the original face. You need to manually click refresh to see changes btw. |
Beta Was this translation helpful? Give feedback.
-
|
can we have offset face bottom, to use instead of masking (which slows things down) in cases like we have ehm "obstructions" around the mouth of the target face please? |
Beta Was this translation helpful? Give feedback.
-
|
Although it would be better to request someting in #87 this was really easy to add. |
Beta Was this translation helpful? Give feedback.
-
|
bless brother. and sorry, new to github. registered just to post about this request |
Beta Was this translation helpful? Give feedback.
-
|
do you have deleted the "mask" offset top face from tha newest version? just now i could really use it... -__- |
Beta Was this translation helpful? Give feedback.
-
|
No?
|
Beta Was this translation helpful? Give feedback.
Image blend ratio is only being used when you turned on one of the face enhancers (GFPGAN, Codeformer, etc.). Original image means the one straight after face swapping (which has a poor resolution of 128x128 max). The ratio defines how much of the original image is blended (mixed) into the enhanced one. So a setting of 0.5 would mean half of the original image and half of the enhanced. Reasoning behind this is, the face enhancement modifies a lot of the persons likeness and looks oftentimes very artificial.
Face similarity threshold on the other hand is only being used if you selected a target face and is something completely different:
#166