I actually like the Dione models as they seem to have a temporal element to reduce flickering on the exact fields, but currently it still exhibits a lot of combing (similar to the 4:2:0 chroma subsampling thing in Resolve someone mentioned before that was fixed), and the actual AI upscaling part usually ends up looking a bit subpar to Proteus since there’s no control over the parameters. For me it usually looks a bit oversharpened and a bit watercolour painting looking.
I imagine if Topaz fixes the combing and if it’s possible at all to do some kind of Proteus deinterlacing model then that would probably be the one I’d use.
