I’m looking forward to more configurable models like Theia.
As long as your expanding the list of AIs to fix videos issues is it possible to make one to fix dirt and scratches from vintage film? I know filters and plugins exist that claim to solve this but in my experience they only work so-so at best for dirt and are pretty mediocre when it comes to scratches.
I’d also love to see a Dione model tailored to fix poorly deinterlaced clips with comb artifacts. The current models only help out maybe 30% of the time and mostly just make them look worse.
Many people think that ‘quality’ is an objective determination. On the contrary, video quality can’t be measured scientifically as it exists in the eyes of the beholder. In addition, in this case it’s a compromise between perceived quality and file size. In other words, you test it and choose the number that works for you.
Oh, and I do cook eggs until they look good to me, don’t you? In my case ‘over easy’.
Well, isn’t that how you cook a lot of things, though? Anyway I think you’re dwelling too much in it. I do think that the “compression factor” in VEAI relates to CRF, perhaps not directly, but in a similar manner. I would agree with johnnystar that 15 is usually a good compromise on file size/video quality, perhaps even quality overkill in most cases. Personally, I usually re-enconde it later in x265 using other applications for more reasonable file sizes while keeping quality, anyway.
Taylor this behavior seems to have changed. Admittedly, this morning I only tried it twice. On first opening the behavior was as I reported. Same after a quit and relaunch. Then I reinstalled 2.1.1 and ran some videos while at work. Tonight the UI in 2.2.0 is no longer stuck on that one model. I’ll do some more messing around but it seems OK!
Works great! However when I try to use “MOV - ProRes 422 HQ (Fast)” I’m getting a “Cannot Write Video” error. No such error when it comes to “MOV - ProRes 422 (Slow)”.
The new comparing view gives me four identical looking preview images in every part of my test movie. I can not see any difference in them no matter which modes I compare. Or is it a using mistake? Do I have to render a preview sequence before the 4 images show the differences?
Artemis Low-10 is better than Low12. It keeps the details on bad quality videos whereas v12 blurs everything a lot more especially Peoples faces and shades on them and blurs shading on cloths too. Medium-12 blurs stuff too the Image look to blur. Video Enhancer 1.9 has best Artemis Low quality AI. I see you going backwards with AI models on other apps too. Stop destroying your own Soft
The last three versions (since 1.9 when I started) Gaia HQ has been exactly the same for me. I’m using a 1050ti, so that’s very slow but it’s been extremely consistent - about 2.1 seconds per frame.
1 Like
viktorz3008
(Z...z...z 😴 I'm sleeping. And I love Sumire 😍.)
61
Honestly, this new version v2.2.0 should not be released. It’s better to stay at v2.1.0 because v.2.2.0 is such a downgrade to me.
The UI is noob friendly? Not quite. It took me longer to navigate things and double check to see if I have the correct options.
Speed is reduced 30%. Might not be a huge deal to others. But for people who are working on movie projects like me, 30% time is about more than 1 hour running for a 25m episode, you see the problem right? High power comsumption, longer time for each movie/episode, slower preview frames. I really want to go back to v.2.1.1. But I also want to improve VEAI, so I decide to stay with v.2.2.0 and give out the opinions to the devs and hope they would make some changes. 0.31s/f => 0.40s/f you must be kidding me.
Models are not getting better, but getting worse. I prefer having a beta testing channel for new models without installing a new VEAI beta everytime. Just the models right into VEAI Beta. So everyone can just download it and test it with the older models. Also let people keep the old models in the beta version.
A model like Gaia with customizations like Theia will be a thumb up for me. It’d be better to have that than updating the current good models (Artemis especially). Let users control the sharpness and colors would be a huge improvement than trying to improve Artemis models which have too many artifacts here and there. That way, everyone is satisfy with the result because AI can never be better than human eyes. At least for now.
Dione should have a model like Gaia CG which is stable, and without sharpness. I really like to deinterlace, but the artifacts are holding me back. I had to use Gaia CG and let Youtube deinterlace the movies for me. But still, I prefer doing it myself. Anyway, a 1080i deinterlace without doubling the framerate would be nice (if that’s possible).
Theia is great, but the customization range is short. I mean… I can’t find the sweet spot at all. It’s like choosing to smooth out the details or having artifacts. Kinda sad. I would like to have the balance between them.
I’m not sure if it’s possible to have this, but an auto detect feature from AI could help a lot of people asking for help because they don’t know what to choose. We’re here for AI, at least to have AI suggestions based on the quality, video type and bluriness some kinda stuff. You know what I mean.
2.1.1 (and all prior versions): Each frame (for a test video) takes ~0.7s per frame using AHQ.
2.2.0: This has increased to over 6 seconds per frame (for the same video).
2.2.0: Importing a test VOB (progressive, 25fps), this has gone from 0.8 seconds per frame to 12-13 seconds per frame. Why? The models appear to be exactly the same, with the exception of interlaced-focused models.
2.2.0: Why does JPG have a video quality settings box when it’s set to compression of 97 (as stated in the popup) and an image sequence? I would strongly welcome a compression slider for JPG so I can change it to compression of 99, for example.
I’ve re-downloaded 2.1.1 and the times have gone back to under a second per frame for the exact same models.
The difference is very clear. Unfortunately Artemis LQ10 has also destroyed surfaces, grass, small stones etc and also when used to clean 1080p has deformated noses on faces. Very strange behaviour. Personally I also wait for deinterlacing improvment of 1080i. QTGMC is still ahead of VEAI. It would be great if VEAI is at least as good.
In my case ‘over easy’.
