I really have no idea how any of this works but I looked around on some of the forums I have access to and Topaz Video Enhance AI seems to be a very popular one. Also none of the following is my own but here is something that might help.
Theia-Detail algorithm in Video Enhance AI is giving the best results yet. With the proper settings, you can achieve a more convincing upscale with less artifacts than the Gaia algorithm. I'm upscaling DVD content after ripping and de-interlacing with QTGMC. Here are some tips:
- "Restore Detail" should do most of work, start at 80, if it still looks pretty soft go up to 85 or 90.
- "Sharpen" often does more harm than good and a little goes a long way. Set it at 5 or 10, never more than 20. Do a few previews starting at 0 and going up increments of 5. What works well depends on a lot of factors but especially the source's contrast levels. A video that looks a little dull and underexposed might do well with sharpen at 20; but brighter or high contrast videos you may not be able to get away with any sharpening. Sharpen is where the "demon eyes" and gross vagina artifacts come from. It's very preferable to have a slightly soft looking video as opposed to these artifacts.
- Much of the same can be said about the "Reduce Noise" feature, a little goes a long way. Even at just 10, you may remove some legitimate noise but you're also going to remove some natural features like body hair and freckles or skin folds/wrinkle patterns on the hands and face. It just makes people look weirdly plastic and airbrushed, almost like mannequins. 5 is usually plenty for a decent quality source.
- If using QTGMC, don't let it interpolate to 60fps. AI upscaling takes forever, but obviously takes even longer with a higher framerate. It's much quicker to upscale at the native framerate, then use SVP or one of the many AVIsynth-type solutions to interpolate the upscaled video. Better yet, use DAIN or Twixtor which are smart algorithms that do predictive motion blur and look more lifelike and natural.
