I have a weird pretty eratic error:
Sometimes Videos get processed without any issue. The same Videos with the same settings produce a black screen when rendering preview while windows is still working (you can still hear youtube videos playing without any issues). Sometimes preview works fine, but export get´s me a blank screen.
According to the log file the video card wasn´t available anymore. I´ve set the max memory usage to 80% und acitvated the “low power” option. I use the latest NVIDIA Studio driver.
I have an RTX 3090 and VRAM usage never get´s over 4 GB. This error occurs since VEAI 3.1 and the introduction of Tensor Core usage.
I also train GANs on that card with nearly 100% VRAM usage (24 GB) and no such error ever occured. The GPU is not overclocked.
Anybody else here with such crashes?
edit: by blank or black screen i don´t mean VEAI going black/blank, the whole Win10 GUI goes black and the monitor goes into sleep mode. You even can´t reactivate the GUI by pressing shift/ctrl/win/B.
I was getting patches like those with Chronos fast. They stopped when I started using the slomo mode. So if I want to make a 30 fps movie 60, I set Apollo to slomo 2x and the output to images. (You can do it with Chronos too.) Once it’s done, I use another tool to convert the images back into a video and add the sound, chapters, and subtitles back in.
Doing it this way has fixed all the blur patches I saw before when using Chronos fast. (It’s been awhile since I’ve ran a frame interpolation video, but I see no reason why this should have changed.)
FEATURE REQUEST: A way to mark previews in the GUI.
I would love to be able to mark previews when I am going through a bunch of them to find the optimum settings for a project. Maybe a simple check mark or flag (keep and reject)? Or maybe a 5-star system? This feature would simplify my workflow.
Some previews and videos are still getting stuck at 99% or taking forever to complete right at the end.
Also, duplicate frames removal doesn’t always seem to work if there are multiple duplicate frames in a row. I’m still getting a pause in certain videos that have this issue, even after some duplicate frames were removed. It’s smoother than before, but not perfect.
Do I need to change the framerate at all when removing frames?
Yes, I tried it. It doesn’t work now. If a file has been modified and you start TVAI, it launches the installation prompt to download that file again. Kinda like there’s a watchdog at the start of the application now that detects modified files.
Testing on a short image sequence, I can see the speed improvements, so I tried to do a full sequence except am running into problems.
Now when importing the whole image sequence, the speed drops to 1/4 what is was doing with the original sequence, will run for about 20 seconds, does 31 frames then “completes” with a green checkmark.
I tried with a smaller sequence again and speed is normal and it doesn’t stop processing prematurely.
Is there a limit to the image sequence size that can be imported before it can’t process?
Although I appreciate the new features in TVAI v3, I would be willing to give them up in exchange for a) real time preview frame-by-frame like in v2, and b) ability to use any older AI models along with the newer ones.
The reason for me saying this is just to make a point about how much I miss some of the previous features, and I’m sure other longtime users do as well. I can always open the v2 application to do that, but that is not as practical.
