[FACE NO GLASSES] - Add a No Glasses Option

[andymagee-52287]

Just as we now have a Preserve Text option, we need a No Glasses option.

FYI, he was not wearing half a pair of modern glasses.

[steveallen243 (steveallen243@gmail.com)]

Lol, good, it’s not just me I see the same thing with an old photo. It looks like they eye that gets the glass(es) is a little darker due to shadows and this gets turned into half an eye glass…

I can’t quite sort out how to send just this picture to the Topaz data set (I don’t want to export/upload all my photos - maybe a “upload this” button would be a nice addition (with a comment about it also maybe?)

[lhkjacky (Jacky)]

I don’t think TPAI is designed to upscale such low resolution image.

At such low resolution, it require using stable diffusion to regenerate the detail back.

[GinTonic (Gino)]

I fully agree: developments should be concentrated on these things and not on things already present, and well, in other image editing software

[tim.he]

Correct, the AI models rely on existing image data so if the original lacks detail or is too low resolution, the program will struggle as well. That seems to be what’s happening here.

[GinTonic (Gino)]

I agree but I think that artificial intelligence should also be, and can be, helped: this is exactly what this post asks for, that is, telling AI whether you want glasses or not. Among other things, apart from the ghost glasses, the face recovery is remarkable even if the image is low resolution

[GinTonic (Gino)]

The image on the right is very fascinating: could you tell us what tools and how you obtained it? Thanks

[lhkjacky (Jacky)]

@GinTonic

The image on the right is very fascinating: could you tell us what tools and how you obtained it?

You may refer to this post

V2.0.1 faces look like wax museum

The “Others” images are upscaled using the latest AI upscaling algorithm called DiffBIR, developed by the Chinese Academy of Sciences and Shanghai AI Laboratory. They recently released their research paper in arXiv. DiffBIR utilizes stable diffusion and controlnet to upscale images and regenerate realistic details, including hair, skin texture, clothing texture, text, trees leaves, grasses and more. Based on my testing, their face recovery surpasses that of GPEN, GFPGAN, CodeFormer, and VQFR,…

After process the image with above software, I futher improve the eyes with Stable Diffusion.

[andymagee-52287]

Some models are more prone than others to create glasses out of nothing. It would be nice if we could just let the model know in the beginning not to try adding any. Perhaps add a “No Glasses” toggle switch? Alternatively there could be a specific model to choose from that is trained without glasses.

I have various old family portraits from the 1890s where I know that no one was wearing glasses. The current set of models (in version 1.2) all want to add on a pair of glasses here and there (or sometimes even just one lens, but not a true monocle) based on something they think might hint at glasses in the original.

I can hand edit the phantom glasses back out in post, but that is very time consuming.

The next complication to really get this right is that images with multiple faces would need to have a “No Glasses” toggle per face.