I’m getting better upscaling results with v6.3.3 than v7
Could you show me the results that you’re receiving with v6.3.3 and v7?
Sorry I meant to say that it doesn’t do well when upscaling old images. As you can see in this image, their faces does not look right.
Thanks
Original image
Try it with the High Fidelity v2 model and set face recovery to something high like 75 to 90%.
I can only really see two faces with obvious problems in the result then, one being the face of the woman that seems to have some strange double exposure issue or something like that in the original image. The other is an eye that seems smaller and a little out of place but it’s not so obvious.
Yeah, I see what you mean. This looks like a scanned image as well, which we don’t technically support since we train the models with digital images from digital cameras. Do all models give you this result or is it just one model?
what AI model you use in v7 and what AI model you use in 6.33 that you claim results are different? that way we can do an apple to apple compare
it says in the filename of the uploaded image, low res scaled up by a factor of 4.
You can save the original image and run it to compare.
I looked at it out of curiosity to see what models mess with the faces and he obviously did not use face recovery, or not very strong face recovery. That’s why i suggested him to try with the High Fidelity v2 model and set face recovery to something high like 75 to 90%. I didn’t process the image but I looked at it in the previews with those settings.
