For me, it was an amazing discovery that video enhancement programs like these exist.
And of all of them I tested, Topaz left the rest far behind - well done.
I like v3 UI and parallel work.
Now I’m using the Proteus model without zoom.
Processing two videos at once is about 30% faster than v2.6.4 when processing one after the other. I like it
But…
There is something wrong with showing the image in the input, preview and output windows. This is best seen in the lines - they have flaws, they are jagged, although the input material does not have it.
Fortunately, this flaw is visible only on the screen - the AI engine gets a well-processed image and the saved file is correct (as in the picture).
This makes it difficult to judge whether I set the parameters correctly, because I do not know whether the image has been corrected well or not.
Is anyone having issues with Main10? Anything I export with Main10 has artifacts/noise like little dots everywhere. I have tried several different drivers for my RTX 3070 and no change. It’s only x265 having the issue if I use like Jpeg2000 10bit it’s fine. x265 is also fine Main only Main10 is having issues.
Hi. Here’s what I’d like to note. I’m currently converting interlaced PAL TV material to 4K 50pfs from the late 60s upwards and no model works as amazingly well as Dione Interlaced TV V1 from v2.6.4. Although there is some kind of interlace in the result, it is not comparable to any other model in naturalness and resolution. If I could choose this model in v3… then I could give up on v2.6.4. Thank you.
Is that preview image at 100% view? I noticed the preview can get pixelated or jagged when it’s not at 100%. For example, when you set it to ‘Fit’ window.
I’m also confused. Ok, if you manually select -50 in ‘relative to auto’, then it means 50% of the 50 auto value = 25. But, what if you set the slider +50? What is the adjusted percentage and the resulting value? I’m guessing it’s 150% of 50 = 75?
It might be simpler to add and subtract instead of using percentages, which would be the same logic used in Proteus ‘Manual’ parameters.
Remember the fixed parameter value changes with each frame so the idea is a percentage can be more flexible I guess. If you had a frame value for a parameter that was 5 in one frame but later in the video was 40 a you could have a problem if you were adding fixed values to the parameter. What we don’t know is are the parameter value linear. Is a change from 1 to 10 the as 31 to 40? We don’t know if the scale has a curve. This is why IMO you still need to look at different parts of a video and use the ‘estimate’ as a guide to set the ‘relative to auto’ percentage. It is also why I preprocess videos to try to get a poor quality video better before feeding it to VAI. The more stable the parameters are across an entire video the easier to get a the percentage close or in some cases just use Auto alone.
If you look at the VEAI command line for Proteus Relative to Auto, you see something like this preblur=0.03:noise=-0.05:details=0:halo=0.1:blur=0.04:compression=0.
This makes me think it’s a simple addition.
Also, based on my experience of selecting parameters for Proteus in the previous version, for any parameter changes of ±5 lead to a minor change in the picture. It would therefore make little sense to apply a percentage correction to the calculated parameters, especially when their values are no more than 20. An authoritative statement on this issue for the current final version of veai from the developers is sorely needed.
Vote for this post to get the developers’ attention.
Again I can only go by what the developers have said in the past. Until they update us on exactly how the sliders work or how to select values for optimum use we only have those statements that were made. The last one that I could find was October 12th.
I was about to report a preview mismatch, but it’s already there, I wouldn’t have even written that it’s a VOB file.
I find the UI great, within a few seconds I found the right model.
Performance:
I get 4.6 FPS in preview 352x576 → 1080p, Interlace Progressive, Dion Robust dehalo, 25 fps.
TR 3960X - Radeon Pro W6800, 5GB GPU memory load.
I run multible previews at the same time (its nice that there is no slowdown), and throw out the bad ones.
Its also nice that when a preview fits my needs that there are still the settings and i don’t have to recreate them.
+1 for color coding previews. For example, green for good, red for not good and yellow for maybe. Or, at least just highlighting the best preview somehow.
My comment had to do with what would be more intuitive for the user. For example, if the auto setting for the ‘dehalo’ parameter is actually 5 (unbeknown to me because it’s invisible), how would I know to crank the ‘relative to auto’ slider to 100 to change the number to 10? Seems extreme for a relatively small adjustment. If it was additive, I already know from experience that moving the slider by 5 points has a small but noticeable effect.
EDIT: Although I know this is mostly a user to user forum, whenever I post, I am also hoping the devs read it and potentially take it into account for future development.
