Topaz Video AI v3.0.7


This week’s release for Topaz Video AI includes some quality of life improvements. We’ve also compiled the major changes in the last few releases here:

Released December 13th, 2022

Download: Windows | Mac (.dmg) | Mac (.pkg)

  • Switch to more reliable web-based activation method
  • On Windows, app will no longer sleep when processing
  • App now correctly detects start frame for image sequences
  • Fix error when processing without any models selected

Submit files here: Dropbox

Please take a look at the Video Roadmap Update for more context on what we’re focusing on right now. Thanks in advance for your feedback!


You’re the best guys :heart_eyes:


Completely crazy! A crazy quality, I would not have thought that it could be improved so much. and it’s faster! :slight_smile:


Amazing, thanks! Guys, can you share your work with movie companies like Netflix etc
Just wondering what the results you have been done (if this is not confidential)


File drop in still don’t work on some pcs and any chance of adding sleep or shut down have video is done?

1 Like

Do I have to delete the old models and re-download them for the improved models to take effect?

That should be automatic.

1 Like

Ok, thanks. No new models were downloaded during preview so I just wondered.

The random numbers in file names are back?..

Initial impressions are that it’s worse, simply because I didn’t get far past login in my testing. Win 10, Nvidia RTX 3060Ti.


  1. New browser licence system forces you to be online each time you open the program.
  2. The username and password often has to be re-entered manually when re-starting the program even during the same computer session. Not always, but too often.
  3. Logging continues even when “detailed logging” is turned off.


  1. Watermark is now removed for users relying on the license extension for v 3.0.x versions, when starting to process a new video (but that’s more of a bug fix).
  2. Didn’t get any further because of the Minuses.

In other words, this software version cannot usefully be used offline either for those who choose to work like this or whose internet connection is down when starting the program. Or who don’t want log files cluttering up their system (we aren’t supposed to be Beta testers!). These things are not standard industry practice for consumer video software and should not be done.

I edited a few times to reflect my latest findings. I will be reverting to an earlier version.

*Late EDIT to add:

My licence expired after release 3.0.3 or so, I am relying on the posted Topaz undertaking to extend licensing for releases 3.0.x, for anyone who was correctly licensed for 3.0.0. This might explain why some are seeing the logon issues with 3.0.7 and others not.


This is completely unacceptable, IMHO.


Hi - the random numbers in the filename should only show up while the file is being processed (this file is actually a temp file and not the final one). Once the processing is done, the temp file (with the numbers) will be replaced by the final output.


It does not for me. Once it’s initially licensed I can completely disable my internet connection and it launches just fine.

I’m also not experiencing this. Are you using a trial version? Because it certainly seems you aren’t licensed correctly.

1 Like

I was upset when I saw what DaveL wrote. Can’t replicate it either, though. Works just fine here.


v3.0.7 works ok.
I just tested Proteus and Gaia with the same video clip which I use always for testing (1280x720 50fps to 3840x2160 50fps.
0.55spf with Gaia is 20% faster than 0.69 spf with Proteus on my PC (AMD 5800X with RTX 3060).
Gaia use up to 100% GPU while Proteus load max 12% GPU (AI processor set to auto)


I’m fairly new to Topaz Video AI and have been using Proteus exclusively. For some reason after reading & I think being directed to Proteus per it’s description in the software I’ve really only used it with very good results.

But reading your post makes me want to test Gaia, especially since I’m using an RTX 3080ti.

Of course, I’ve been doing 2 steps - 1. Enhancement then 2. Upscaling

Is Gaia as good doing enhancement as Proteus on most files?

Should I do 1. Enhance with Proteus, then 2. Upscale with Gaia since that would seem to make more sense if Proteus is better for enhancement and to save time with upscaling…

Also should I go with the latest Nvidia Studio Drivers or Game Drivers when using Topaz AI - any difference?

Thanks for any suggestions!

Loving this software that brings new life to some of my old favorite stuff. Sometimes like seeing it for the first time.



Hi Mike,
I can only explain what my observations are and how I get it:
Target size is always 4k (and I keep fps!)
Input video are either taken from TV Mediathek, MFT Systemcamera, or iPhone and range from 720x480 to 4k with all kind of compression and format
Therefore in TVAI first I trim a short but representative clip of the video I want to process, and look which presets work best in Artemis, Proteus or Gaia.

If satisfied by one of these 3 I’ll apply my choice to process the whole video.
If I’m not satisfied (face refinement, crop, stabilization, etc…) I preprocess that video tradionally in MAGIX and/or Mercalli befor I go back to TVAI.

1 Like

I don’t work for me, all models failed :frowning:

1 Like

Beta testers shouldn’t have to pay for the privilege of doing such work. I’m really disappointed that Topaz put this out and asked us to pay for it. It reminds me of a story I read once about a company that hired some so-called “whizz-kid marketing genius” straight out of college who decided to “fix” everything that wasn’t broken to justify their existence, and drove the company into the ground.
I’ve paid twice now for a program that remains unusable, and spent many hours testing and reporting results, which should be done by the company before a product is sold to the public.
But I’m just an old man who yearns for the higher standards of the good old days. This place was all trees and fields in my day…


Please tell me.
Does the improved quality of exported video by matching color space introduced in “Improved model quality” refer to “Quality now matches 2.6.4” which was supported in v3.0.5 or to something else?
In the FFMPEG warnings, before 3.0.4 it was BGR and after 3.0.5 it was RGB.

1 Like