use “Edit”
1 Like
Thank you! I really appreciate the insight.
Now I am encountering the unexpected result that Output Settings are not saved in user defined presets. Every attempt to save a preset reverts the Encoder and Container settings to the default item from the top of the list, which is not a useful result for me.
I will have to remember to manually change the output each run, or risk losing the rendering time to an incorrect file format.
1 Like
After you’ve modified your preset you need to tell the programme under “Preferences” >“Interface” again which of the presets should be your default preset.
1 Like
Thank you again.
I have since changed the default Output Preference which solves the trouble for me, but I would still expect the use of a user defined preset to override a default setting rather than see certain parameters not being saved to the preset (Output Settings), which is the current result.
I’m all fine with it’s current implementation, I only mention it as I think there will be others with the same expectations.
Absolutely. A preset should save EVERYTHING…no exceptions. Every single setting should be included.
2 Likes
When you investigate into the json files in preset folder it looks like all settings get saved
Thank you for the info MikeF.
Try to upscale the output to 1080p, I figured out that is the slow down.
If I output original scale, I get about the same as version 3.6.4
My test file: 640 x 480, Artemis, Low Quality Noisy, output to 1080P
Over 18 minutes. in v3
Thanks for the reply TPX,
I figured out it’s the Upscale output that is the problem.
Original is: 640 x 480 one minutes
Artemis, Low Quality, Noisy - output to original about 4 minutes in 3.6.4
Output to 1080p over 18 minutes in V3
About 4 minutes in v 3.6.4
My system is a 2950X Threadripper,
Mother board: ASRock X399 Taichi
32gb ram, geforce gtx 1080 ti
I hope the upscaling issue has been reported.
Thank you
I’ve found the same issues as you, and still using version 3.6.4
Definitely keep that until v3 gets fixed, My upscaling test show that v3 is 5 or 6 times Slower than v2.6.4
Steve,
Are you upscaling to 1920x1080? Try upscaling with a scale of 225 percent and a width of 1440 and height of 1080. Your original video is a 4:3 or 1.333 ratio video not 16x9 or 1.777. When I upscaled in V3.0.0 using 1920x1080, artemis lq noisy my time was 7.5 minutes. When I switched to the proper scaling and width that time dropped to 2.4 minutes.
1 Like
I have now had the opportunity to compare the processing of VEAI 2.6.4 and VAI 3.0 with several input video clips encoded in ProRes 422, h. 264, and DV on my fully maxed out 16 in. M1 Max MacBook Pro. In each case the exact same clip was processed to upscale to 1920X1080 with no other other type of processing. In 2.6.4 the Dione Robust model was employed whereas in 3.0 the Dione 2X was used. With 2.6.4 the processing time ranged from 0.07 - 0.09 sec per frame. In 3.0 the processing time ranged from 0.23 - 0.40 sec per frame. On this high end MacBook Pro Video AI was 3-7X slower than Video Enhance AI 2.6.4. In its current version Video AI is so slow as to be essentially unusable. Furthermore, I see no difference in the video output quality of 2.6.4 (which is very good) and 3.0. In summary, until and unless the processing speed of Video AI 3.0 is SIGNIFICANTLY improved I will have no choice but to continue to use 2.6.4.
1 Like
Could someone verify if we are or are not using RT cores on Nvidia 30 series cards currently. I ran 3 test with the same settings on 2.6.4, 3.0.0.11.b and on 3.0.0. The results from Task Manager is inconclusive but you can see there are differences in what part of the card might be used. GPU-Z does not for my card show what cores are being used. I’m at a loss as to how to determine for sure about RT cores. The crazy thing is I am getting the approximately the same speed with 2.6.4 and 3.0.0 but 3.0.0.11.b is a little bit slower. The high system memory use is because of a VM running in the background.
Thank you for the reply MikeF
Here is an update:
Original 640 x 480 one minutes video with 1937 frames
Artemis, Low quality, Noisy
V2.6.4
1440 x 1080 is 5 minutes 0.19 sec/frame
1920 x 1080 is 5 minutes 0.18 sec/frame
V3.0
1280 x 960 is 6 minutes
1440 x 1080 is 6.5 minutes
1920 x 1080 is 18 minutes
That did help, I just wonder why is it still slower than v2.6.4 can output to 1080p so much faster?
I hope it can be added to the Fix list.
Let me know if I can run an other test ideas you have 
Sad as this software is going downhill. These models are becoming worse and worse, except convertion to 60 fps, which worked very nice in 2.6.4.
It is nice to see parallel processing - very good move.
The UI is confusing now.
Artemis LQ was the best overall for most videos. I was expecting a lot - amazing upscale and super sharpness. Unfortunately got blur at slow speed.
we know that it is slower than 2.6.4… that’s how it is, you have to deal with it. I realized this evening that each model of veai can work miracles, provided you modify its front sequence with hybrid, and play with noise, halo, grain, and sharpness, (with hybrid or other software). I deinterlace with hybrid, and apply three “dehalo” filters, a “line” filter, two “sharpen” filters and “rec 709” then I open my video with veai > proteus in SD and finally with vegas pro > project in HD to refine the quality, and change the colors.
Here is my result with these two comparisons:
4 Likes
I verified I get the same slowdown when scaling to 1920x1080 vs 1440x1080. I can only guess it but this is FFmpeg based and there might be something there. Maybe they are not using the GPU to generate the bars on the sides. I don’t know.
I did an experiment with Staxrip just know where I upscaled using the CPU only through FFmpeg to 1440x1080. That upscaled at 225Fps. When I added borders to the video to make it 1920x1080 the Fps dropped to 51Fps. That may say a lot. I think it is a CPU generated border in 3.0.0 but not in 2.6.4 maybe.
2 Likes
in 2.6.4 all 4/3 videos i was upscalling in 1920x1080 without doing a manual crop were slower.
so i assume that VEAI doesn’ recognise if there are black bars or not.
tests on a P5000 quadro (close to a 1080 Gtx) :
Video used : 640x480 / CGi video of 219 frames. 25 fps
Gear : P5000 Quadro Nvidia + 16Gb Ram + Intel Xeon 2.5ghz
2.6.4 Artemis Medium
640x480 219frames 25fps → 1440x1080 0.20s/f Total: 49s
640x480 219frames 25fps → 1920x1080 0.21s/f Total: 51s
3.0 Artemis Medium
640x480 → 1440x1080 0.21 s/f Total : 54s
640x480 → 1920x1080 0.60 s/f Total : 2mn22s
possibles explanations :
- VEAI 3.0 render very slowly when 4/3 video is upscaled with no crop to a 16/9 output . (letterbox mode). maybe the black bars are the cause.
- Known Issues : Speed can be slower than 2.6.4 with some models on some machines. “We’ll be working on speed optimization in the next weeks” (Devs)
- MikeF explanation
3 Likes
Cool, they must have addressed this in Beta then as this is definitely not functioning as such in v3.0.0.
Not sure if I have a setting wrong or if this is an issue. Anytime I render out anything it makes two copies even if only doing one video. I have tried to do set of videos and it works but still make two copies of the first one before doing two separate videos. Looks like two different models even though I opened TVAI and went to my preset.
American Dad! - S01E06 - Homeland Insecurity [360p-SD] [x265-10bit] [MP3 2.0]_317742471_prob3
American Dad! - S01E06 - Homeland Insecurity [360p-SD] [x265-10bit] [MP3 2.0]_557192069_tvai
1 Like
Huh? The release is the latest version. There has been no new beta since the release.
