I think that adjusting with ANTI-ALIAS / DEBLUR (cmd: preblur) will produce better results than adjusting the sharpening with SHARPEN (cmd: blur).
preblur
Adjusts both the antialiasing and deblurring strength relative to the amount of aliasing and blurring in the input video.
blur
Additional sharpening of the video. Use this if the input video looks too soft.
The value set should be relative to the amount of softness in the input video - if the input video is already sharp, higher values will introduce more artifacts.
For the ANTI-ALIAS / DEBLUR, a PLUS will set the value to DEBLUR and make the image sharper. Conversely, a MINUS will result in ANTI-ALIAS, making the image blurred.
The old perpetual licence model worked on major version releases. So you could buy V1 and 3 months later V2 is released and you donât get any updates. Topaz is much better as you get 1 year of updates, which is a much fairer way of doing things.
huge upgrade to performance so thank you. just got one question. any word on av1 support for amd 7900 xtx or xt cards? also smart access video or ai accelerator support for amd?
Although I am eternally grateful for your tweaks, and the woman with the red lips looks a whole lot better (when does a woman with red lips ever not look good? ), I ran some extra experiments. A nose closeup with your current settings:
Jup. Same issue here. System is Mac OS Ventura on a 16GB Mac Mini M1. That preview thingy is slowly getting nerving. Sometimes preview shows in split screen left and right completely different videos, parallel upscaling of >1 Videos is in progress.
Sometimes I really miss the quad preview from V2.6, and also itâs flawless preview during encoding. Please further optimize that in 3.x somehowâŠ
Otherwise the the last few releases was worlds away from the initial - super buggy - 3.0x releases- The 3.x branch is heading towards the right direction ;o)
What your seeing from scene to scene is why just using the estimate in proteus never seems to work over an entire video. The real trick and this takes some trial and patience is using âRelative to Autoâ. Hopefully the video is fairly consistent with the type of problems from scene to scene. What wonât be consistent is the values to repair the problems. Relative to Auto goes a long why to help with the problem.
I also try to limit my upscale to 2x or a max of 225 percent for a single pass with TVAI. I also if Iâm working with a very low res input file I will use conventional methods to do a slight upscale for generally a better result. Say 512 width to 640 or 720 to 854. Itâs not much but I find that it helps. I also do these at a lossless or very close to lossless conversion.
I posted this in another topic but it didnât get any traction. Could someone test downscaling a video? Version 3.1.0 is taking magnitudes longer than version 3.0.12 to the point that itâs unusable. Upscaling is also much slower. Thanks.
Yeah, I always tweak ârelative to autoâ. I just had no idea sharpening had such detrimental effect on the overall picture quality (for some reason, I got it in my head it would just work on edges).
Yeah, same here. In fact, with video that is too low-res, I scale up a bit (Lanczos), then let TVAI take over. The trick, I found, is to allow for enough scale âroomâ, as it were, for TVAI to perform a usable, visibly repaired upscale, but not let it come from such low res that it starts mangling faces.
I see no incentive for anyone to pay to receive incremental updates to 3.x with the program in itâs current state since the release of 3.
Currently, the incentive is to not upgrade, or participate in any kind of testing of the updates and maybe consider upgrading once the bugs are ironed out and performance has been improved.
I would have agreed with you, until I realized that this is âbleeding edgeâ software, or software based on largely experimental new technologies. As such, Topaz is doing a better job keeping it working, and stable, and speedy, and effective, than any comparable commercial competitors out there. Are there bumps and setbacks along the way? Yes. However, with the price of admission, we get to ride this A.I. ship, and boldly go where no video program has gone before.