This. I was just going to create a new thread about this, but no matter which model I use (Gaia, Artemis etc) there is randomly like a judder effect that happens. It’s like the program decided to skip some frames all at once making for jarring movement. This still happens even when not saving as MP4. Saving as PNG and combining the frames with an external program causes the same issue. The colour in all instances are also severely muted for some reason.
Comparing the source video vs the result, VEAI seems to be skipping 4-6 frames randomly. Audio sync is also a bit off throughout the end product ):
Playing through my 4k equipment it is much better than on the PC but still not as smooth as the original. I did notice that when I installed the update it was automatically set to Artemis HQ. Maybe that’s the problem? I’m now trying Gaia HQ & Theia Detail on an excerpt to se how it is.
I know this is more of a feature request, but a “Test” function is BADLY needed.
Something like: user selects a frame, then selects the number of frames to test (1 or more from the current point), then it goes through and automatically saves the frame/frames using every model, using the output method specified, and using any settings the user has set for Theia.
Otherwise, the way it is now, the workflow to do this is manual, tedious, and involves so many steps. Some automation is needed. This will greatly improve users’ ability to gauge which model works best for various different sources.
Have you figured out if there are actually frames missing? Total count of pngs should match the number that VEAI shows on input. I suppose you could check the input video frame count in another program to be sure that VEAI is interpreting correctly. Finally, I think it might be helpful if you list your system specs and program settings.
Interesting. It might be your source. I am getting an error in Virtualdub2 that it can’t read the source frame number 2. Edit it’s detecting a lot of bad or missing frames for the source.
The source vid above was reencoded into HEVC as it was cut from a longer clip. Here’s an untouched clip from the source. Only remuxed, no reencoding with avidemux.
Thanks. I’ll just reencode the source with max quality settings and that should resolve itself then. On another topic, what would get better results? Denosie/Deblock before upscale, or the other way round?
I myself Denoise and Deblock as much as possible before upscaling. VEAI does a decent job cleaning up noisy sources on upscale but it works better if the source is clean first. At least in my testing it does.
The reason is that H264/x264 codec wasn’t really meant for 4K and only powerful PCs can playback 4K smoothly when encoded in H264. That’s why we need H265/x265. Youtube for example not only encodes 4K in VP9 for bandwidth reasons but also for the known playback issues with AVC on 4K videos.
Depends on the source video (how it was filmed, compressed, etc), but Gaia-HQ or Artemis HQ should work well. In some cases, Artemis-LQ does a better job. It’s best to use the preview function to do some tests.
I am currently able to use older models, for example Gaia-HQ v2, v3, v4 (from the beta versions). A contact at Topaz told me he should be able to make Gaia-HQ v1.0.1 work as well… he will try.
I have noticed one improvement possible for Nvidia cards’ owners: use Studio drivers, not gaming ones.
I see a higher average load of GPU with Studio drivers.
