I would like to see Face Enhancement added to video enhance ai. This would really improve the quality of videos by having the ability to add detail to them, instead of only removing compression artifacts and smoothing out lines.
Super Quote… <3
I agree. In my videos, face detail is very important. If you could add something similar to the Remini app for face enhancement, that would be appreciated. But the GAN should be constrained to face details like skin, for best results, Remini, for example, is very good at enhancing skin detail, but it sometimes produces grotesque monster teeth when enhancing photos. I appreciate that this is a very big ask, but frame to frame temporal consistency matters in video. Photoshop has neural filters for photos, but Adobe has not yet figured out how to make them work in Premiere and After Effects.
Yes, we need this Face Recovery to enhance videos, too. Time and time again, when I upscale a video with a face close up, the outlines are improved, but the face itself, I mean the skin in particular, loses detail and almost looks fake. This is a big issue.
It is like I actually see less detail, not more.
You could always play with AVCLabs Video Enhancer AI. It has the face restoration function. Be warned, using that function is super slow and the technique is obviously in its infancy. Super dependent on the source material as well. Some results are mind blowing and others are horrifying. I think Topaz could improve on how it’s utilized and produce better results.
Really surprised by how good the new face recovery is on Gigapixel 6.1.0.
Obviously, something like this probably can’t be as good with Video Enhance AI due to frame consistency, but if something like it can be done to make it better than it is now, that would be great.
I think many users would not mind selectable options in the application that would increase processing time by two times if it would mean much better results.
A lot of users try to run their video through VEAI more than one time to try to improve the results, which IMHO is not the way the application is designed to be used, so they spend twice the time anyways for results that are actually worse than they could be.
But with the old footage I work with – 1960s for example, 2-pass is helpful, Once to denoise and deblock at 100%, then another model for detail.
You seem to be forgetting that Gigapixel takes about ten times longer with Face Recovery enabled and it increases with each additional face detected.