High to sharpen the faces of all persons

[inkdolls]

(11-12-2024, 06:46 PM)Krikor Wrote: Does it work on Windows? Does it require extra files or does it work as a plugin?

Is there a website that already provides this service?

Yes, you can run on any platform where Python runs, including WIndows. The extra files it needs are the trained models and you could say all the huge dependencies it requires. It's not a plugin, but some AI interfaces add it as an option to refine faces after image generation.

Yes, there are online demos, I guess you can find the same links that I can. The problem is sometimes there's a long queue and you get long delays, if anything at all. I used it extensively online when I first heard about it.

(11-12-2024, 06:46 PM)Krikor Wrote: Could you explain the steps on how to run/install this GFPGAN?

I can try to help if you really want to install it, but I'm sorry to say local installation is difficult if you haven't got some knowledge of Python and console commands. These are the instructions, but those will probably not make any sense to the casual user. It will take a lot of disk space and dealing with all the versions of all those Python extensions (and Python itself) can be a real chore.

There's also the option of running it on Google Colab, if you can't make a local install, but you need the knowledge anyway.

(11-12-2024, 07:36 PM)denzjos Wrote: @inkdolls, I only made the faces that Stephin Liu asked. Doing the workflow without the selection, then the whole photo match. You can then correct the colors using the (automatic) levels or using the white, gray and black points levels. It is always difficult to fix small faces (not the group ) because there is usually no graphical information about these areas in the photo and some filters just add noise to give the impression that the photo is sharper. That's why restoring faces using AI sometimes looks great, but the result is usually a look-alike person. Thanks for the GFPGAN link.

Yeah, too much information is lost, AI can't do any more than what the trained models provide, you'll hardly get any resemblance to the actual people from such a poor source.

You're welcome. And thanks to you, I didn't know about NAPS.