I can see both of these marks in your original, the upper one very faint but the lower one pretty obvious. It appears to me that Proteus V4 is just doing a better job of enhancing them than V3 did. Have you tried lowering the sharpness and antialias/dehalo settings?
You have right. If Iām lowering Reduce Noise spots are smaller or non. If Iām increasing Denoise spots are stronger and bigger. For me, itās unacceptable error - must be improving by Topaz. V3 donāt have this bug and as I see now, V3 give better resultant denoise images (I donāt speak about spots).
And of course itās not only one place in the exported episode with this error.
Then thatās not really a bug or an error. Itās just the new model being more effective at enhancing faint image details. If you donāt like that, enable previous models in Preferences and pick V3, which is still available if you want to use it.
Remove the damn AVX2 control already! Version 4.1.0 could be installed, but now 4.1.1 canāt!
1 Like
In that generality this would be a quite bold statement.
Of course there is no such thing as the one and only correct settings that fits all sources.
So this was meant for my sources (mostly medium quality SD content, partly interlaced) and, of course, for my taste. Looking at some material here, some people seem to like oversharpened and very artificially āenhancedā fake looking, sometimes nearly oil painted looks.
I prefer having the AI being less intrusive/more true to the original while still giving as much extra details compared to a simple upscale as possible.
That said I have generally good results (with the above mentioned stuff) with a manual setting of:
Fix compression: 65
Improve details: 55
Sharpen: 12
Reduce noise: 0-10 (not really needed as Iris already does remove most of the noise)
Dehalo: 5-15 (again, Iris already does a lot of dehaloing, but for some sources dehalo can make the image more natural / remove sharp edges that make people looke like āstamped inā)
Antialias / Deblur: +65
Recover original details: 80-85
I do not use auto or relative to auto here as I find that the manual settings do work generally better for my content (my sources are of constant quality. With sources that vary very much in quality the auto setting might be better).
2 Likes
