Hello everyone!
We have another beta release for y’all’s testing.
Downloads
Windows
Mac (DMG) | Mac (PKG)
Changelog:
Thanks for testing
Please upload problem videos and logs here: Submit files
Thank you
I was wondering if I could get some clarification on this.
In TVAI 3.1.7.1b, it said:
- Apollo is the default for slow-mo, and Apollo Soft is the default for frame-rate conversion
Is this still the case in this beta? Because from looking at the command line commands generated by TVAI, it seems Apollo (Normal) is used in all cases.
Or is Apollo used in all cases, and TVAI automatically switches between that and Apollo Soft depending on the slo-mo factor?
I made a report about this a few betas ago, but I’m still having issues with it so I’ll report it here again.
When trying to run TVAI ffmpeg from the command line on macOS, it doesn’t work. The reason: ffmpeg doesn’t have permission to do something (read?, write?) in the models folder.
Changing the permissions of the models folder by running chmod 775 "/Applications/Topaz Video AI BETA.app/Contents/Resources/models" fixes it.
System information:
macOS 13.2.1
Processor: M1 Pro
Installation method for TVAI: Download DMG, open the DMG, drag and drop TVAI into Applications folder.
Other potentially useful information: There is only one account on the Mac and it is the one I’m using.
Just throwing some benchmark numbers in the ring (and I found the separate benchmark thread after posting this, sorry)! Would it be feasible and make your life easier if there was a “send benchmark results to Topaz database” function? This would be great to have available online in the future as well to see how your current system ranks against others / how much of an impact a system upgrade might have on real world performance.
Topaz Video AI Beta v3.1.9.0.b
System Information
OS: Windows v10.2009
CPU: AMD Ryzen 9 5950X 16-Core Processor 63.909 GB
GPU: NVIDIA GeForce RTX 3090 23.77 GB
Settings: device: 0 vram: 1 instances: 1
Benchmark Results
Artemis 1X: 16.59 fps 2X: 7.37 fps 4X: 1.95 fps
Proteus 1X: 13.38 fps 2X: 6.4 fps 4X: 1.95 fps
Gaia 1X: 6.8 fps 2X: 4.43 fps 4X: 2.42 fps
4X Slowmo Apollo: 17.56 fps Chronos: 14.32 fps Chronos Fast: 16.18 fps
Good idea, would love to understand what criteria would influence fps, of course CPU, GPU, RAM, model+version, settings but also video resolution, color profiles…
Also thinking 2 videos of the same resolution could still be very different in complexity, so could we establish some video templates we would all use for bench-marking? This would make the comparison more meaningful. You’d have to find videos without copyright claims, I am thinking there would be plenty available online that would be usable for such comparison.
If Topaz would provide a link to some videos to test and settings to run I’d be happy to run them on my system and report results. Hopefully others would do the same so we better understand the product, what hardware choices to make and Topaz can improve VE AI models and engine.
Thanks
Edit: just found the performance bench-marking thread, I am now having a RTFM moment:-)
Happy to see option to set the default preview, thank you
@gregory.maddra
Topaz Video AI Beta v3.1.9.0.b
System Information
OS: Mac v13.0201
CPU: Intel(R) Core™ i5-7600K CPU @ 3.80GHz 32 GB
GPU: AMD Radeon Pro 580 8 GB
Settings: device: 0 vram: 1 instances: 0
Benchmark Results
Artemis 1X: 1.45 fps 2X: 0.99 fps 4X: 0.39 fps
Proteus 1X: 1.61 fps 2X: 1.12 fps 4X: 0.46 fps
Gaia 1X: 0.53 fps 2X: 0.37 fps 4X: 0.28 fps
4X Slowmo Apollo: 1.52 fps Chronos: 0.21 fps Chronos Fast: 0.46 fps
Topaz Video AI Beta v3.1.9.0.b
System Information
OS: Mac v13.0201
CPU: Intel(R) Core™ i5-7600K CPU @ 3.80GHz 32 GB
GPU: AMD Radeon RX 480 8 GB
GPU: AMD Radeon Pro 580 8 GB
Settings: device: 2 vram: 1 instances: 1
Benchmark Results
Artemis 1X: 2.89 fps 2X: 1.64 fps 4X: 0.7 fps
Proteus 1X: 2.7 fps 2X: 1.82 fps 4X: 0.65 fps
Gaia 1X: 1.06 fps 2X: 0.75 fps 4X: 0.57 fps
4X Slowmo Apollo: 2.31 fps Chronos: 0.4 fps Chronos Fast: 0.98 fps
@gregory.maddra Uploaded Test Files 1080i and files smaller than HD.
Blockquote
Topaz Video AI Beta v3.1.9.0.b
System Information
OS: Windows v11.2009
CPU: AMD Ryzen 9 5950X 16-Core Processor 63.928 GB
GPU: NVIDIA GeForce RTX 3080 9.8311 GB
Settings: device: 0 vram: 1 instances: 1
Benchmark Results
Artemis 1X: 15.27 fps 2X: 6.68 fps 4X: 1.9 fps
Proteus 1X: 14.39 fps 2X: 6.24 fps 4X: 1.95 fps
Gaia 1X: 6.9 fps 2X: 4.6 fps 4X: 2.55 fps
4X Slowmo Apollo: 16.56 fps Chronos: 14.96 fps Chronos Fast: 21.07 fps
Preview still broken.
Hi,
Is this happening with the release DMGs as well, or only the ones for the betas?
So, where did Apollo Soft go to?!
Yes, I am having this issue with the release versions as well.
Although I didn’t notice because version 3.0.X worked fine, then I switched to 3.1.0 beta (where I first had issues) and didn’t switch back to the release branch until now just to test it.
Edit:
I just installed TVAI 3.0.0 to retest the old version. And I’m getting the exact same issue there. And chmod 775 /path/to/models/folder fixes it there as well. So it seems the issue is most likely my system doing something weird, but I don’t know why.
I might try factory resetting my Mac soon and letting you know if that fixes the issue. Unless you don’t want me to factory reset so we can investigate this issue, see if there’s something TVAI could do to fix it automatically?
By default the models directory is writeable for the installing user, though something seems to have gone wrong with that for you.
We will make the default permissions for that folder more expansive in the next release.
Could you let us know when we can use a new old templates and manager as before? 3 years ago you told us that the manager will be back, we would like to know when you will be back? Thanks
In Preview when I press “Back to input” the input video starts playing automatically. I don’t think I have seen other versions do that. It is annoying when I am trying different models / settings on the same 5 or 10 seconds part of the video - the automatic playing means the next preview will not be starting from the same place as the previous preview unless I move the video back to that exact point, which I need to copy and paste from a completed preview.
Artemis model with Low Quality / Noisy setting is colourising face of person in the background. The same does not happen with either Proteus or Artemis when Medium quality is chosen. The input is an entirely black and white movie from 1956.
Severe Chronos Fast mishaps appeared during a 1080p movie I did, using 59.94 fps, using Proteus.
DxDiag.txt (98.0 KB)
These floor deformations are quite severe, and seem to come out of the blue, but don’t occur with Apollo.
Not to complain too much, but a process like this takes over 24 hours (on an i9 12900K and RTX 3080 Ti), after which you can toss the entire job as failed. 
