You can’t use another program that also needs GPU memory or CUDA computing when VEAI is running even with high end hardware. I haven’t try the power usage set up, try set it to the minimum value but even if it works then VEAI will be extremely slow in that case.
And if you use CPU only it’s the same, it’s heavy computing and memory use that it left few resources left.
Your game is simply cutting out VEAI resources and then it stops running.
In one case watching a video on YouTube was enough to make Video Enhance AI stop working!
1 Like
Yep. So far I manage to play games while using VEAi but only ones that have really poor graphics. Also any app that uses CUDA needs to be turned off or you’ll get a warning.
VEAi doesnt warn you, it just crashes.
I bought Video Enhance AI last night and must say that this application does a really great job. Well done. 
You know I am still quite unhappy with the GigaPixel AI which still feels too faulty in terms of usage and results.
The Video Enhance AI works almost flawlessy but it sometimes stops working or drops a warning when too busy.
I also tested jpeg to raw and Sharpen AI but both felt not yet ready for use to me because of the uneven results I saw.
VEAI Works pretty well so far, but there’s a few things that would be nice to have for future versions.
-
Better face/hair handling for progressive non-CG video. Right now eyes, mouths, hairlines and eyebrows seem to give VEAI a headache.
-
Some kind of high bitrate x264/x265 or even lossless output, or at the very least more control over output options. You’ve got ffmpeg bundled with the install, let us pass configuration options to it manually if you don’t want to add additional output profiles (variable framerate, audio codec selection, etc).
-
Utilizing multiple GPUs for single jobs if feasible. If not, then the ability to execute multiple jobs simultaneously and utilize more than one GPU that way without having to run multiple instances of the software.
-
A zoom option to scale the input file the same amount as the output using the usual upscaling that a media player would use. This way we can see in real-time how the chosen algorithm compares to a regular playback upscale.
1 Like
To shorten the processing time it would be useful to enable the application to use CPU and GPU together.
2 Likes
Hi,
Here are some observations with the latest version of VAIE.
- The “LQ 100% deblock denoise” mode does not give the same rendering as the previous version (see the three screenshots).
- in LQ 100% Deblock Denoise mode, there are texture analysis problems which often give a flowing paint effect as we can see on the example of water on the ground (two original screenshots and treated).
On that of the tiles, we can see on the right that part of these, however correct in the original, become blurred (screenshot of VAIE during calculation). - in HQ Gaia mode uspcale 1080p at 4K, on the example, there is noise which is added (screenshot of VAIE during calculation)
Does speed (500% instead of 100% of resources) affect the quality of the rendering? All calculations with 1.2.1 are 500%. I do not know if this parameter was on 1.1.1, but if it was the case it had to 100% not having adjusted it.
I just did a test with version 1.1.1 on the water in LQ mode 100% Deblock Denoise and the result is much better. The paint effect is much less pronounced than with 1.2.1 which makes the plan usable.
For tiles the problem was already there with 1.1.1. On the other hand, the vehicles on the left are better rendered in 1.1.1 than in 1.1.2.
In addition, with 1.1.1, I tried to reprocess 4K shots from a Phantom 4 drone that had huge artifacts due to compression. To reprocess them, I had to reduce the resolution to 1080p, switch them to LQ Denoise Deblock mode then switch them back to HQ 4K mode. The result is spectacular. The ideal would be to have a Deblock Denoise mode directly in 4K so as not to lose quality (reduction in resolution) and computing time.
OpenVINO speeds up CPU processing by about 3 times for Artemis, and by about 2 times for Gaia. There does not appear a way to turn off OpenVINO as there is in other Topaz products.
1.2.1 Artemis LQ produces noticeably more artifacts than both 1.2.0 Artemis MQ and LQ. It also has less of a denoising and deblocking effect. There are noticeable ringing and halos around objects that are not present in 1.2.0 Artemis LQ and MQ. Depending on source playback, these ringing and halo effects can be unstable. The 1.2.0 Artemis-MQ mode would stabilize the source to minimize this effect. That’s gone now. Overall Artemis is definitely a downgrade in my humble opinion.
The Gaia models are also different. Gaia-HQ is mostly the same. Gaia-CG is somewhat improved in that it will not produce dither type patterns as often. It also produces less artifacts and an overall softer image. I would say Gaia-CG is improved.
All filters seem to brighten the video, noticeably lowering contrast. Gaia-CG does this the least. Not good!
Here are updated comparisons. It occurred to me that 28 Days Later is a good “stress test” since it was shot using late 90’s “pro-sumer” digital video cameras in PAL (720x576). They over sharpened their footage and the quality is horrible, so AI improvements are very limited due to lack of detail in the source. As before, the source is the PAL DVD from 720x576 and scaled 2x to 1440x1152, then resized to the correct aspect ratio of 1920x1080 using spline144 resize in AviSynth, recoded to high quality h264 via crf18/veryslow preset.
As before, I recommend you download the images (Zippyshare link posted above for convenience) and use IrfanView to do “in place” comparisons. In addition to using the space bar to go forward one image and the backspace key to go backward one image, you can also open multiple images in multiple instances of IrfanView, then switch the open windows by using your start bar. This is the best way to clearly see differences.
Source sample clip, 28 Days Later:
Processed clips, 28 Days Later:
Source sample clip, Martin:
Processed clips, Martin:
Gaia CG often handles the video footage best for me. Gaia HQ and Artemies often make it too crisp, add strange noise and do not remove compression artifacts that well.
1 Like
I agree CG gives the cleanest results, if you are willing to sacrifice true surface texture from the source. For that reason, none of the modes are “there” yet for me. I await future quality improvements before doing any serious upscaling.
Actually CG is the only setting that can be used for the moment, or you have to go back to 1.1.1.
1 Like
That may be true, but I’d rather go forward than back 
. Hopefully, in a few months it will be altogether better, and faster.
Are there some benefits to output in 16 bits tiff videos that are 8 bits ? Does it improve the dynamic range, the colors ? for a better color correction postwork.
I tried once a 10 minutes SD 480p clip at 200%, it has bring a folder about 180GB of size 
and to just learn after that that you can’t import 16 bits TIF image on Premiere Pro 
Fortunately VirtualDub2 can do it and then you can export a 10/12/14 bits output video. But does it really worth it knowing you will have a gigantic master file in size ?
I read a post in the FB VEAI beta testing page where someone did a comparison between 8 bit and 16 bit output. If memory serves, his conclusion was that there was a small difference where 16 bit gradations looked a bit smoother, but for him it wasn’t worth the much larger file sizes.
What annoys me is that Gaia HQ and CG often adds unwanted noise to skin textures, especially non-white skin. This makes young skin apper like old skin, and the unwanted dotted texture. The older CG mode has less of this issue as it had more smoothening.
1 Like
VEAI is fine for beta testing purposes and it looks promising, but it’s not ready for serious upscaling yet. I would give it 8 to 10 months of further development; if it’s still not improved much by then, I’m sure another competitor will come along with a better software offering.
Hello,
I had my hopes up for trying this product but it doesn’t seem to work with my computer with the following specs:
GPU: RTX 2080 Super
CPU: AMD 3700x
RAM: 16GB DDR4 3600mhz
OS: Windows 10
I just get the “AI Engine Failed” message and changing the resource percentage makes no difference. Any ideas?
2 Likes
