Updated presets and new 480p → 4K upscaling method with zero artifact + super resolution.
DVD/VCD to 4K:
Please watch some Youtube videos about Adobe Premiere Pro before using this method. It will help you achieve amazing results.
I use DaVinci Resolve and it seems upscaling is better in Topaz to be honest. Any suggestions?
I have started to use the following path as someone had suggested:
480p x4 AAA-9 and then downsampling with Artemis HQ to 1080p. The footaga is not very sharp but clean. Below I give you the original, AVI do ProRes 422, could you prepare 1080p footage according to your technique? I wonder if I could use DaVinci to check the path:
https://we.tl/t-mcl1p4eDiH
Later today I shall send you upscaled file. I also used GAIA CG but there are left some artefacts on surfaces.
Artemis AAA9 x4 and down to 1076p with AHQ 12
https://we.tl/t-GUCOSVa95Y
and GAIA CG but directly to 1080p and as someone once wrote, Topaz scales x2 or 4x.
https://we.tl/t-p2jJI3DctD
I just want to update this old post about my PSU. I just did an RMA with Corsair on my HX1200 and the new PSU arrived today. I tested the new one with 100% power at stock, no adjustments. Fortunately, it worked perfectly with 25% less power draw from the power source. I have a Kill A Watt meter and it really showed 900W with my old HX1200 (2019 model). Then the new HX1200 (2021 model) reduced my PC total power draw from 900W to 650-700W. Amazing. If you have any kind of problem like I had, please consider RMA your PSU and get a new one. You will save a lot of money and your PC components life span. I believe PSU brands did some upgrades for the current products to handle Nvidia 3000 Geforce series.
Just wanna let you guys know that VEAI v.2.2.0 still runs faster on my machine with “Reduce machine load” enabled. Low power consumption + PC runs smoothier + 30% performance boost.
I think that too.
For fast power switching GPUs like Ampere or Navi 2 you need bigger capacitors.
Is Gaia CQ a typo? Do you mean Gaia CG or Gaia HQ for the 480p → 4k upscaling method?
Yea, sorry, Gaia CG.
Thank you for this detailed guide! I was just going to ask if something like this existed.
Hope you find it useful. Thanks.
Is it planned to update your performance guide for Protus v1 & v2?
Because each video is unique. It would be hard for me to tell other users how to use it correctly because not everyone’s taste is the same. The model is just for people who want to customize the sliders. But for me, I still use Artemis and Gaia for now because those models satisfy me a lot already. Because Proteus is about customization, it depends on users themselves.
No, NVENC is no good in any form, used for anything, at any time. It makes unnecessarily huge files in all cases. It’s simple pointless to use.
I’m new to VEAI, bought 2.4 about a month ago in order to upscale downloaded old Mediathek films and some private videos taken with my Olympus MFT cameras in order to fit my monitor (50 FPS, 2560x1440).
The Material downloaded is not interlaced. Therefore I tried AI models: Artemis LQ,MQ, Gaia, Proteus, and Theia.
My PC configuration is: AMD Ryzen 5PRO4650G (6 Cores, 12 Threads), 32GB RAM, SSD, NVIDIA GTX 1650, which works fine consuming only 100 to 130 Watt/hour.
I experience around 1 sec/frame processing time with Artemis and Proteus. That is ok, charging the CPU at ~8%, and GPU at ~95% (at 2,2GB GPU memory,
<72°Celsius).
45 minutes film lenght takes about 40 hours processing (awfull, but tolerable for me because running in background while doing other Photoshop etc. work).
With Gaia an Theia I found no difference to original upscaling without VEAI models.
With Artemis I got awfull flat faces without structure.
For the moment I use Proteus with my own settings. That seems to deliver the best compromise with only marginal artifacts in faces.
What should I do for getting better results?
I bet your PC specs is just slow, so it’s kinda normal actually. You might want to upgrade if you really want to speed up VEAI.
Theia is not good imo. And Gaia is only for noisy footages, and it works… okay. I mainly use Artemis all the time because it’s better. Depends on your video, but if artifacts are presented, then you can try Artemis Aliased and Moire.
Thank you for your feedback.
Speed is not the primary issue but quality.
I tried Artemis Aliased and Moire but was not satisfied.
I found out that Artemis Dehalo (with Grain at 2 and 2) seems to be the solution for this material (faces not completly flat, not as sharp as Proteus customized but without artefacts).
Now I experiment to process 2 videos in parallel (CPU load 3%, GPU load 95% GPU memory 2,7 of 4GB). Processing time alternates 1 to 2.4 sec/frame, what I consider tolerable for background processing, enough capacity for regular foreground applications.
Not sure if this is the right place, but I did a lot of testing over the last couple of weeks. Since I still can’t post any pictures, you’ll have to take my word for it 
Artemis HQ v12
Artemis MQ v13
Artemis LQ v13
One caveat, though: The low-end Artemis models seem to have a tendency to sample-and-hold movement patterns for a few too many subsequent frames. This makes people ‘grimacing’ while talking and moving their heads at the same time. It’s particularly bad if the video source is of low quality and very soft (unsharp) or blocky.
Artemis LQ v13 is still afflicted with this ‘holding pattern’ and weird grimacing.
Gaia CG v6
Gaia HQ v6
I’m pretty sure Topaz abandoned Gaia models already because the v5 hit their limit of perfection already. Any changes will just make it worse. All models have artifacts, but Artemis HQ v11 and MQ v12 are what I’m using the most. Gaia HQ v5 is pretty good for DVD sources, but just don’t zoom in for the artifacts because it’s not worth it or visible at any viewing distance. Artemis + noise is the best bet for almost no artifacts. Gaia + little noise will also do the same, but not as good as with Artemis. Gaia CG should stay at v5 because it is for what it is, but still need more sharpness in overall.
What about Proteus v1 or v2? Any comparisons?
I think this model is not for comparison because the result is vary. But I can say Proteus v1 is better. Proteus v2 has more artifacts.
