I have skipped the 2.2.x versions as a result of the numerous complaints in this forum but, given the good reviews of 2.3 over here, I thought I’d give it a try. And I must say this version is really, really worth the download! The Proteus model in itself is a game changer for many of the videos I process. I mostly process stuff of already good quality (such as digital satellite TV recordings) and while Artemis HQ already gave me good results in most cases, Proteus not only keeps up with AHQ when using the automatic settings but fine tuning of the suggested settings very often results in a model that beats Artemis. However, the auto settings Proteus uses in many cases seem to be a bit “brutal” (too high)as to dehaloing and somewhat “weak” as to compression artefact removal but both can easily be tuned manually - especially when having both models on display in “comparison view” which is an enormous help as to choosing model and setting. You can even use two different parametrizations of the same model - for example, I use AHQ as the first preview screen, Proteus auto mode for the second and for the third screen Proteus with manually altered settings that I believe to perform even better. That is a so much more effective workflow than the older version provided.
A small problem has popped up with 2.3, anyway: Starting the preview, sometimes starting the processing as well, results in the program being closed every now and then. Not a big deal because usually it works fine after restart, even the settings for the job before the crash are saved and can be used again. However, this did not happen with 2.1, at least on my machine. The Proteus mode seems to be particularly susceptible to these crashes. Apart from that, 2.3 is a great update and a heck of a job by the VEAI dev team!
No friend Johnnystar, even because from what I understand Proteus is also for SD video both interlaced and progressive, because this example I presented above was using PROTEUS AUTOMATIC PATTERN, I did it quickly, without really understanding the terms used, which was the my proposal in a post above, and I found that with some changes the video has improved a lot, I presented this automatic pattern only as a didactic one to present some problems that arise in the original videos and that can be improved through this model using correctly the various parameters it presents here, so I emphasize once again: we have technicians that are understandable by those in the area, as if I were to use technical terms of IT systems analysis, many who are not in the area will not understand, so I asked the technical staff and a professional who would be kind enough to put an image of what each term represents, so that we can quickly understand what we can or not to do with these parameters of Proteus, I believe that now I have been clearer, I hope, so everyone has to gain here in this renowned community.
P.S. Friend you’re right, Proteus works with progressive video, I just did a deinterlacing with Hybrid QTGMC and now there will be no artifacts (comb effect, serrations) nor the lines around the spotlights .
Yes my friend Johnnystar I saw the video, but I confess that my English is basic for listening and writing too, so much so that I use google translator and many may have already noticed the bad English I write here :).
Thank you very much for answering my questions, it’s still not ideal, it would be ideal with images, but it should be noted that you are a professional or even a person who has didactics and understanding to put yourself in the position of the neophyte person aka noobie who is in doubt and so be able to better help you, my sincere thanks, few people have this patience and didactics to teach, maybe everyone here could have their knowledge of communication, and although perhaps no one will come here to point out this quality as well, I am sure that many here will be benefit from your answer addressed to my question.
I know that many here don’t have all the time in the world, to spend a few minutes of their time to collect images of the terms used here and in the Proteus parameters and what we usually discuss here, to improve everyone’s understanding, but still I continue to to encourage whoever can, to do it for the common good, in this way, even recurrent doubts that always appear here in the community are spared. For example, some terms can get confused when talking about artifacts, comb effect, serrated effect, because without an image of them, they can be totally different patterns or same patterns and we will never know for sure; another example is the gouache paint effect (which I use to make the faces look strange in the Dione DV and TV model) and/or another one used by the folks here, which is the plastic effect of the faces, which for me is the same thing, anyway I’m waiting for the images, and thank you in advance if some good connoisseur can give us this knowledge :).
But friend, you are answering my question exactly with the assertion that was contrary to my proposition, let me see if I can make myself understood, maybe you could put yourself in my position as an apprentice/neophyte:
You say at the end of your answer: “regardless of what the devs or other people write about it.”
Do you see where the solution I am pointing to is? “than the devs or other people WRITE ABOUT IT”.
You noticed the word WRITE, because this is my problem, a picture is worth a THOUSAND WORDS, no matter how hard someone tries to write it in words, an image will show what it can take a whole text to write, I think it should have stayed now clear to you.
My friend Johnnystar understood, but even giving the answer that might help me, he still didn’t do it in an ideal way, but his teaching was definitely better :).
I am currently on a trial for a A.I. upscaler, called Topaz Video Enhance AI (v. 2.3.0). Looks very promising, except… it can only do output lossless to either .tif of .png files. (sic!) What the heck am I suppposed to do with a trillion png files?! I had really hoped it could output either H264 or H265 losslessly (for finetuning, say, x265 parameters on it afterwards). Or any kind of contiguous lossless format, really. Could this be added in the future? Thanks.
ProRes is lossly or almost lossly. and if you put the H264 CRF at 1, you’ll have something very very close to lossless. kind of impossible to see the difference.
Edit : i suggest you to put the CRF to 17, good result with that already, if not just use ProRes if you want the best result possible (be careful, the files are very big)
I wasn’t trying to teach you, except to say you can solve this in about 15 minutes. Who cares what is written about it? The only thing important is to SEE firsthand what it does.
I suppose a low CRF will do (typically, QP=0 is the way to go lossless with x264). Thing is, project itself is already going to take 2 days; spending time on H264 compressing as well (whereas I really just want a lossless, processable output), would take like a week.
What is curious to me, is that y’all have such a brilliant upscaler, but then no way to just output stuff losslessly, to be parsed to something else (like x265). Ideally, output to STDOUT would be perfect, so ppl, like me, can use their own procesing tools (with all custom tweaks), instead of relying on what I assume is nVidia’s (somewhat interferior) NVENC internal encoder Video Enhance A.I uses.
Would love the ability for us to properly view (including in split view) portrait videos with a “rotate” metadata flag. Happens often with portait videos recorded on iphones.
like jphilip said , juste use ProRes then. you’re on a trial… just try the software and enjoy. you’ll not find anyway, anywhere else a better product for doing upscalling actually. this software is pure magic.
I was thinking - how would Nvidia’s DLSS 2.2 (eliminates a lot of ghosting compared to 2.1) fare in VEAI as a scaler or a helper/supporter in scaling? Or AMD’s FSR?
If you want lossless then PNG/TIF output is your choice. That’s what I do for my serious projects.
After that I save a video file in Premiere Pro with a lossless codec (MagicYUV) with RGB colorspace to avoid any colorspace conversion when re-importing the video.
FYI, eventually I went with H264 + CRF = 0 (only parameter you can set, it seems), as I don’t have a free 4T drive at the moment to store all intermediate tiffs. Also, seems the upscaler lets my 1080 Ti do most of the work, so eta is rather reasonable, considering.
As for HVEC license, that is not what I meant. I meant to plead for an option whereby the upscaler outputs its data to STDOUT as, say, y4m, to be caught by a pipe, so a third-party encoder (x265, in my case – a public, free encoder) can pick up the data and compress it (with all kinds of detailed, custom tweaks). Especially since this upscaler is so awesome: worthy as a ‘filter’ in a longer chain.
.
