Do you think Placebo would generate a better quality picture and maybe this issue wouldn’t occur?
1 Like
QTGMC with Placebo preset is the BEST Deinterlacing method for the best final video quality.
I hope that in the next version of VEAI 3, with FFMPEG, it will be possible to use it, or at least that the developers will improve the program’s deinterlacing method (such as the “Dione” method)
1 Like
No, it’s not related to this. The issue is related to lack of details / poor source material (some of them are even DVD, but not great ones). some are converted laserdisc digitally recorded, or VHS. i noticed the issue only on concerts with black or uniform color background (like blue or dark blue , or even other colors but all uniform). one ofthe solution was to inject in the source more noise to the picture, so when upscaled, VEAI see them as details to enhance and not noise to erase (kind of). the issue happens mainly with Proteus, and less with other artemis models.
solutions for now is a mix of script / deblock / noise generator done with the help of QTGMC with added script feature under Avysynth/Vapoursynth + Neat Video plugin. some guys in the VEAI community are really pure genious, two of them helped me but it’s a lot of work as i want to upscal not good original material at 704x480 / 720x576 to 1920x1440 and then to crop it to 1920x1080.
can’t try on a very long video because after 20mn, if we don’t click on the window, the computer shut down (it’s a shadowpc , a cloud computing offer,).
i tried on a video, here are the results. i don’t know where i’m doing it wrong :
test
A/ 2 x VEAI at the same time (Card with 16Gb of VRam )
first VEAI 1 - 5mn02
2nd VEAI - 4mn41
= 9mn43s
B/ Same video with one VEAI running twice one after the other = 7mn23
What are the computer specs. I’m sure you’ve said but I don’t remember 
I’ve seen that when I’ve run out of resources Windows tries to compensate which is rarely good. If I oversaturate the cores usage say I was running 2 instances at 78 percent cpu and I try a 3rd or other things running in the background. The worst time I had was when I run out of system ram on 16GB machine because of automated background tasks and VEAI. That just slowed VEAI incredibly. Windows instead of limiting certain low priority processes just start using the page file. Even with a fast NVME I went from 0.35 sec/frame to 1.65 sec/frame. That’s when I decided to not have less than 32GB of ram in any computer that might do AI or virtual machines. I wish VEAI was available for Linux but Linux doesn’t get along well some graphics cards.
if i could use the computer i tried to build the last two months, it would be great, i7-12700 + z690 motherboard + 64gb ram + rtx 3060 OC + Nmve samsung 980 as system + 970 nmve 2Tb for stockage + 6 x 2.5" SSD.
but life is always a b*** lol, so
on this actual computer (online one in a datacenter, that i take control from here), there is :
- Unfortunatly 12Gb of system Ram, and i can’t change it. it will be updated in october up to 16 i think.
- Graphic Card is a Quadro P5000 with 16Gb of Vram
- Cpu is Xeon E5-2678 V3 @ 2.5ghz (4 cores only).
- 256Gb of SSD for Stockage. (i can have more but i don’t need it).
so by reading what you tell , the issue is on the cpu and much more important, the very low system Ram. unfortunatly can’t do anything except waiting october and jump on their new offer, which will have a RTX 3070 + 16 Gb Ram and a CPU AMD EPYC™ 7543P (4 Cores and 8 MT)
video card can be a AMD Radeon™ PRO V620 too, not sure we can choose.
Price will grow up, but high chance that before october i’ll be able to have my own new machine running, with 32Gb Ram + Z690 + I7-12700 + RTx 3060. We’ll see…
I think your just completely hardware limited right now. Upgrade when you can but for now one instance at a time.
1 Like
I dont rememeber graphic driver that I used, and I don’t kwow how install a old driver nvidia.
hi, ok, it’s easy to do
1/ if you need VEAI 2.6.3 application and you don’t have it anymore, you can download it here :
at the top you’ll see the 2.6.3 thread/download.
2/ you can try to uninstall your nvidia graphic driver, and install a previous version. but you told that the speed issue you had was related to VEAI version only, so uninstalling the 2.6.4 and installing the 2.6.3 would make it working again maybe ?
3/if it’s graphic card driver related, you can uninstall your driver and install an old one.
you can find plenty of tutorial or text about this on nvidia forum or on google search.
there is as well this tool :
4/ you can find older nvidia drivers here :
1 Like
Would you be kind enough to share your secret sauce for ridding the exports of the dreaded scary faces syndrome… Thanks!
2 Likes
Hi, I was using the trial for the topaz video enhance and it was great, the speed was around 0,20 sec per frame and I updated my macos monterey to the latest version and now topaz is very slow, rendering at 2,30 sec per frame. Do you guys know how to solve this? I have a macbook pro m1 2021
1 Like
I would like to know if there is support for dual CPU systems (like my HP Z840 dual Xeon) in the pipeline? I have noticed that only one of my cpu gets used (cpu 0) 12 physical 24 logical, I could expect a large boost if I could used both cpus, note that running another instance does not make VEAI use the other cpu 1.
1 Like
If I could make a suggestion from personal experience. The scary face (distortion) is from the ai not having enough frames to work from. I usually use Flowframes to up the framerate to 60fps before treating video files in VEAI. Also turning the slider down for high speed shots helps, with slower static shots you can get away with moving the sliders up.
2 Likes
It seems quite logical to add several things to new releases:
- The restoration of the old film footage in one model or building a model’s pipeline option:
-removal of dirts, scratches and other artefacts.
-stabilization of the footage.
-removing brightness and color flickering.
-color correction, and may be colorization of b/w footage. - Add convertation of the FPS option inside each preset but not only separate render or add a model’s pipeline option.
- The ability to configure more output parameters for each codec. Now, for example, there is no ProRes and ProRes LT(only HQ), no MXF(film and video industry standard), and you cannot specify the more settings for x.264 except CRF, it is not possible to use the codecs installed in the system and coding through FFMPEG.
Thanks
2 Likes
I feel like it’s taking my soul but i can’t look away.
1 Like
Hey there, trying to understand something since the beginning, but didn’t succed yet (lol).
it seems that depending how much a resolution is, when not “usual”, VEAI can slow down or go faster, depending of some factor.
for example, on my machine converting a 720p (1280x720) to 1080p (1920x1080) with a scale factor of 150% takes 1,21 s/frame
but converting 1184x826 to 1920x1080 at 131% scale take more than double the time.
I assume that to fix this by converting in 1280x720 the 118x826 would fix the speed issue, but just wanted to know why it doing this, and when it’s doing it (so i know which video i must re-convert / crop before sending it in VEAI and which not. Thanks !
If I have a 1280x720 video that is not very clean I will sometimes scale down to 854x480 to get a speed advantage. I will convert to mp4 using a quality of 0 in Staxrip. Then in VEAI I upscale to 1080p(225%) at a CRF of 5. Finally I take that file back in to Staxrip and process to the final size and quality I desire.
The vertical resolution is the driving factor in this case. Also remember that any scale factor over like 125 or 130 is automatically scaled to 200 percent by the AI and then downscaled to your chosen resolution, at least that was the way it used to be. If your working with very high quality videos you might be better off doing the final scaling from 200 or 400 percent in an external program where you have much more control.
Hi mike, yes i know this in big parts, i noticed that the switching from 2x factor to 4x model is happening exactly at 239%. to speed up the process, someone suggested me to stay on the 2x model and that do a conversion of upscaling when it’s close to the result i want, but i noticed that a lot of time, the x4 model even when it was downscalling after to go back to 1080p (like a 279% scale or even 300 one), was making a much better result. but i’m playing with “hybrid” since few days (the software), and there is a big work that can be done in such software already in term of denoise etc, and which can make the upscaling better, as the quality source matter so much in VEAI.
so if how you tell, it’s the vertical resolution the driving factor, i’ll try some test. i sometimes as well downscale too to 854x480 etc… 960x540 too. so i’m curious to know at which point in term of vertical resolution, the software slow down or go faster ! i’ll have to do some test so can harmonise better the workflow ! (thanks !).
here is an example with a very Bad vhs , recording in ntsc, transfered in Pal (lost the color of course), but it was just to see if in term of denoise, i could make something better to feed VEAI. so all this worth it to be investigated !
One thing is for sure I don’t think there will ever be a time in the near future when the process will be click a button and the AI figures everything out on a single consumer grade computer.
Have you heard of videoINR model? Do people use it to learn? How about the effect?
