Note, the model tends to lose details on high-quality videos ![]() This model is trained to work on all three categories - progressive, interlaced, interlaced progressive.I tried it and it works surprisingly well on this type of content. ![]() With this AI Model they finally got rid of those creepy faces and that typical "AI enhanced" look.Īlso works great on interlaced and heavily compressed or noisy content. In fact it works great on all lower quality content that needs enhancement. Screenshot is not enf/cmnf related but just to give you an idea. Thanks for bringing this to my attention - the more recent iterations of Topaz Video AI (the 3.0 series) was quite cutting edge and not something I was planning to use until it became a little more stable and it looks like now is a good time to jump in! It's still probably worth keeping 2.6.4 installed as well for backup situations where the current version doesn't work as well. The face situation with Iris is quite interesting: faces have been a weak point for AI upscaling for a while now and I wasn't entirely convinced that we were going to see much progress on that front in the short term.įor my test I used JYH-0001 (starring Yuma Asami). ![]() This is a 720x480 video from a DVD and it is not the highest quality original clip, so it will serve as a good test to see how well it fares with a relatively low quality source (that currently doesn't seem to have a better quality version out there). Because I wanted to really put the algorithm through vigorous testing (and also because I got a bit carried away), I took a lot of comparison shots at a 2x upscale multiplier and the results are.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |