There is a longstanding bug in which the facial recovery outputs the faces nudged one pixel down and one pixel to the right. This bug is now also happening on faces that are output via Very Compressed model as well. (In past versions, the Very Compressed output was offset in it’s entirety - this bug was fixed but now is occurring in facial areas only). For those of us who composite outputs from multiple models to achieve better results, this requires workarounds and is slowing down the workflow.
Well this sounds odd.
Would you mind sharing a screenshot or two of the conflict so I may better understand just exactly what is occurring?
Thanks for responding. Please see attached JPGs. If you layer them in photoshop, you can see what I mean.
Hi @steve6,
Thank you for the report. I see what you mean!
I don’t have an ETA as to when we can commit the time and resources to a proper fix for this, but I wanted to let you know that we are tracking it.
