I took at look at the Sayaka video, and I noticed it is less sharp than your Yuri Takase one. But I guess that's due to the peculiarities of the source? I think it might be interesting to do a side-by-side video comparing new and old TGMC.
What you saw was exactly what the source was like. That's the point I've been making about accuracy vs quality. The Yuri vid is a clean, sharp fairly recent vid. The Sayaka vid is from 2004 and looks almost analog, much smoother and with more noise / less detail.
In Yuri's vid, the new mode helped pull out exact detail, e.g. skin tone, eyelashes, individual hairs etc. In Sayaka's vid, the new mode helped to remove the plastic-like oversharpening that TGMC always puts on foliage. However, it also honored the noise, so I did have to run an additional denoise pass on that one. So post-processing becomes relevant. You may want to do additional deblocking, debanding, denoising and sharpening. Although if you need too much of those, then you might as well stick with the original modes since there won't be additional detail to gain. I've considered adding such post work in the script... maybe a later version.
Actually, you can already control denoising with the existing settings (noise bypass, tr2), but sharpness is trickier. Of the two new modes, the first removes halos and adds a little more detail, in this mode you can control sharpness too (at the risk of returning the halos). The second mode is even more precise and adds more detail, but with that precision you lose most of the internal sharpness control. There's also the full lossless mode, where clearly you lose most image control (wouldn't be lossless otherwise).
Regarding post-work, it's also worth noting that edge cleaning is rarely required on the new modes, since much of the halos were being introduced by TGMC. Now you just get the edge effects in the source. However, if you post-sharpen you run the risk of haloing again. I considered post-sharpening Sayaka's vid, but didn't because of this. Many people will not like the new modes because they tend not to be oversharp - some people
like oversharpening as a psycho-visual improvement.
As you can see from watching the vids this is no panacea that will make all vids perfect and comparable. There will never be jaw-dropping improvements in quality because existing algorithms are already good. However, it's what you don't see that's important, the rips should have just looked
right. I hate the slightly "plastic" look that standard TGMC gives, and that can be in part due to the source, but it's also due to the oversharpening/denoising of TGMC.
I haven't posted comparison shots yet because I'm still developing. I've already improved the algorithm since the Yuri vid.
