To be clear, I’m not suggesting to replace or add anything to the IRIS models to get perfect recovered faces.
But if it’s possible, I would like to see a special mode we could activate, and it would specifically use the recovery face models of Gigapixel/Photo AI…
I very often use this workflow for interviews, where the camera is static, and the face is relatively well defined already. But blurry, noisy, or with compression artifacts.
But it’s a bit of a pain to do because I have to isolate the face I want to really enhance, crop all the frames, then feed everything to Gigapixel. It works very, very well.
If this was included in Video AI, it would be perfect. And easier to manage.
recover the face of a static image is very different from improving a moving face in a video scene. I don’t think it will work that way. But you can give it a try: save all frames of the interview to images and process them via Gigapixel or Photo AI.
That is why I would like to see a special mode that can be activated for these particular situations.
The recovered faces are almost perfect with this method. Of course, it doesn’t work for everything; when a face already has small details defined, the frames are extremely consistent.
I often add a deflicker effect on top of it or pass the whole video with the new faces in Video AI, with less aggressive settings.
Of course, it can’t work with every single video, but then it just like any other video source. Some models work better than others, and some times the sources is just bad, and Video AI can just make them just a bit less bad.
But you can give it a try: save all frames of the interview to images and process them via Gigapixel or Photo AI.