I don’t really check the forum much so I don’t know what people tend to use for what but I have noticed that Dione models are far superior to any other models for upscaling older content. Artemis models just don’t look realistic and, Proteus models can’t get good sharp pixels out of low res stuff, and Theia models look like their shoehorned in new versions but they actually produce the otherwise most realistic results except they are too constrained and sometimes appear to dehalo the images. for say 720p to 1080p upscaling, Proteus works great as does Theia though Theia seems to take way longer.
I don’t use the Dione models that increase framerate, I don’t see the point. I use Dione models exclusively for older stuff, it’s hard to explain but it really brings the detail out and looks realistic. The issue with the Dione models is they seem to be heavily effected by the way the video was compressed. You can take an old 640x480 avi and upscale it 200% and sometimes it looks great but you can take that same video, compress it to mp4 and not see a difference in the source, but the video upscaled with Dione will have weird compression artifacts. Dione robust tends to look the best but sometimes has dirty stray pixels, Dione Robust Dehalo seems to handle deblocking a bit better but sometimes dehalo’s way too much.
What I wind up doing alot is running a Dione Robust @ 200% then running the output through Proteus or Theia @ 100% or 200% pretty much to just reduce compression/deblock and smooth it out. This obviously takes alot of time and if the source was really bad I can’t get rid of all of the artifacts that DTD left so I’m wondering if there’s any way to create a Dione type model like DTD that does more deblocking, maybe a settable level of deblocking/smoothing or whatever. Not sure how the models work, whether this is something that has to be done in a post step or what.
