Video Enhance AI v1.2.2 - Stability improvement & old devices support

The Titan RTX has 1,200 more cuda cores than the 1080ti, higher boost clock, and 24GB of DDR6 vs the 1080ti 11GB of DDR5

me too


Ja, naturlich.

Have you tried using PNG instead?

Hi,

  1. if you used the original file from the DVD (Mpeg2) ? your video source is interlaced. I would suggest you to do a de-interlacing before.

  2. if it’s a DVD rip made by yourself, the original frame rate can’t be at 24fps , except if you converted it to 24 (it’s not a good idea, as the original frame rate (if your DVD is from Zone 1 / NTSC) is necessary at 29.97 (of 25 if it’s PAL).
    something is wrong with your source file, i suggest you to check that.
    so if you put the Mkv with inside the Mpeg2/AC3 files, the destination file must be 29.97.

  3. unfortunatly the out of sync issue with VEAI is reported here since the beginning. but the reason can have different explanation.
    a) bug from the software to be fixed. we noticed that some frames are removed during the enconding process at the beginning or the end of the file and it can maybe be the source of the out of sync.

  4. maybe an issue with your source file frame rate, i really suggest you to check this 24/29.97 “problem”, because the only version available of deep space nine can only 25 or 29.97.

there is no bluray version of Deep Space Nine, only The next generation is availabe in bluray and i don’t know if they converted the 29.97 material to 24fps as i don’t own it.

  1. doing a 3 days conversion in VEAI is not i think a good idea. i would suggest you to do it chapter by chapter. several people complained here about having their render crashing during a long time. it’s very easy to avoid.

  2. if the audio being out of sync is constant, this is as well easy to fix in a video editor. but it would depend only if the out of sync is constant. rendering small parts of the whole video can reduce a lot the issue if the out of synch is growing with time.

Thanks, I’ll give that a try. First
 I already did the conversion with v1.2.1 and this problem didn’t occur. I was just doing it with v1.2.2 because I wanted to see if the render quality improved. I’m reading the FPS in MediaInfo via the K-lite package. It reads the frame rate as 24.121, but maybe I’m misunderstanding what I’m looking at. Finally, upsampling Star Trek DS9 and Voyager from DVD Rips are done, precisely because Bluray releases will never occur. v1.2.2 hasn’t crashed for me at all, so doing it piece-wise (at this point) doesn’t provide me any leverage.

1 Like

Doesn’t matter, all outputs have this issue.

Yes, i’m aware of the “no bluray” version of DS9 and Voy, we’re waiting this since a long time.

so with 1.2.1 version no audio out of sync issue ? well then stay with it. if the 1.2.2 make a better picture, then trying to resync the audio (if constant) is easy and maybe worth it.

using K-lite package is a very bad idea, like any codec package they bring more issue than help. If you want to do this seriously i would suggest to do it together with a nice video editor.

anyway, if you started from an original DVD source that YOU’re owning, the fps can only be 29.97 (America or Japan DVD) or 25 (Europe, South America DVD), it’s just impossible to have a 24fps, DS9 was done for TV, and back then in the 90’s the 24fps was only for cinema.

so if you have a 24fps frame rate you maybe used an another source (internet) and maybe the one who did the conversion messed it up and did (badly certainly) a conversion in 24 (but this must be check) and hes maybe de-interlaced it. check in media info, if it’s tell you it’s progressive or de interlaced. don’t base yourself about what you see on VLC or other video reader (MPC-HC, etc
) because they usually apply a shader of deinterlacing on the picture when reading the video.

It has all those extra things over the 1080ti you list, but that doesn’t translate to alot more cuda performance considering the cost. The RTX Titan costs roughly 3 times what a 1080ti costs at RRP and you’ll be lucky to get 15% more in cuda performance unless the 24GB of vram is needed for the task (which doesn’t seem to be the case for VEAI).

Well I took a recording of the input image as a video then upscaled it with png and tif and I couldn’t reproduce the artifacts (though in the past I tried tiff and it had some issues and I thought was a transparency error so stopped using that format). Maybe if I have the same input video I could try again to reproduce the issue.

Hi all. The docs say that Video Enhance AI can run on MacOS 10.12, but it crashes consistently on launch for me on 10.12.6. Can I send logs or help in some other way?

Edit:

dyld: launch, loading dependent libraries

Dyld Error Message:
  Symbol not found: _OBJC_CLASS_$_MTLCaptureManager
  Referenced from: /Applications/Topaz Labs/*/Topaz Video Enhance AI.app/Contents/MacOS/../Frameworks/QtGui.framework/Versions/5/QtGui (which was built for Mac OS X 10.13)
  Expected in: /System/Library/Frameworks/Metal.framework/Versions/A/Metal
 in /Applications/Topaz Labs/*/Topaz Video Enhance AI.app/Contents/MacOS/../Frameworks/QtGui.framework/Versions/5/QtGui

No bro, it’s a 20% increase over the 2080ti and about a 45% increase over the 1080ti. In no way should the RTX Titan have the same rendering performance as a 1080ti. Is there a clear benchmark made by the devs of expected upscaling time?

1 Like

Who said anything about rendering performance? We were talking about cuda (compute) performance (rendering and compute are different things), and I never said they were the same. Yes the Titan RTX is a bit stronger than the 1080ti in cuda, but not 45%. In general rendering performance they’re is a bigger jump over the 1080ti, but the 2080ti and the Titan RTX perform very close to each other.

Edit: Perhaps I instigated a bit of confusion using the term “cuda” in reference to the “cuda api” that can be used for compute tasks on nvidia gpus, rather than the gpu calculation units “cuda cores”.

As for benchmarks, they’ve only given a couple of vague metrics (quoted from the faq) such as: “On a PC with Nvidia GTX 1080. For a video from HD to 8K, it typically takes 3-4 seconds for each frame. If you upscale a video from SD to HD, it will take about 0.4 seconds for each frame.”

Gaia-HQ
Standard upscaling for high-quality progressive footage.

Gaia-CG
Upscaling for progressive computer-generated content. Use this for digital animations, cartoons, etc.

Artemis-HQ
Enhancement and upscaling for high-quality progressive footage with better motion consistency. Use this to upscale higher-quality footage, like stock video clips, films, and drone footage.

Artemis-LQ
Enhancement and upscaling for low-quality progressive footage with better motion consistency. Use this for old family recordings, historical footage, etc.

Artemis and Gaia are nice names but these names do not describe what the best use of these modes are. Basically one has to memorize what they stand for. Names should be Functional Descriptions, the way LQ, HQ, and CG are meaningful.
Based on the explanantions it sounds like Gaia-HQ is for lower but still HQ footage whereas Artemis HQ is for even better, higher Quality input.

Aside from this VEAi is astonoshingly GREAT, and only needs a few tweaks to make it even better.
Congrats to all the people who worked on this.

Thanks

Bowen,

Some people have suggested that the 100% DeNoise DeBlock be used and then another pass with a different Mode for input footage with the poorest quality.

Does this make sense?

Thanks,

Dan

Bowen,

I often would like to get 4-5 still frames from one particular place and another 5-6 still frames from another section of the footage.

Right now I have to close down the input video and load it again to accomplish this. Could we get a way to do this without the extra steps ?

Thanks,

Dan

I’m guessing that it’s because VEAI puts a little checkmark on the video thumbnail to indicate the video has finished processing and then it doesn’t let you work with it further until you reopen it. Is that what you mean? If so, click on the little check mark and it will disappear. Now the video is ready to process again.

Bowen,

For a longer video, it is sometimes very cumbersome to find the exact position for still frames. The Left and Right buttons move only 1 frame at a time so moving around is very slow. Using the keyboard L and Right Arrows is a little better but it can still be v frustrating to to get to where you want to be.

Could we have Ctrl- L or R Arrow in order to move by 10 or 20 frames at a time ? Or something of the sort ? Maybe Ctrl+ Shift with L or R Arrow would double the number of frames with each click ?

Thanks,

Dan

I wasn’t aware of the checkmark. I will try out your suggestion and report back.

Thanks

I followed you suggestion and unfortunately it does not work. The Select Start Frame cannot be changed, even after the check mark is gone.

*Even with 2080ti still pretty slow
*When using MP4 the audio is getting changed to AAC-LC (original was ac3) Remuxing the original audio back makes video work in normal windows 10 player otherwise I had to use vlc
*Using tiff uses alot of space not much can be done about that I suppose
*Would be nice to set a default setting that would apply to all queued items changing setting for each video is a pain

1 Like