Topaz Video AI v3.1.9

Hello Everyone!

A new release of Topaz Video AI is now available.

Released March 14th, 2023

Downloads: Windows | Mac

Changes from v3.1.8

  • Adds tool to benchmark performance (Process > Benchmark or Ctrl/Cmd + B)
  • Updates Apollo model
  • Prevents running the app from DMG
  • Adds preference for disabling anonymous data collection
  • Links to the support page when an unknown error occurs
  • Adds feedback button and options to the Help menu
  • HEVC saves as HVC1 instead of HEV1. This allows videos to be played in more apps on Mac
  • Improves tooltips
  • Fixes Full-frame Stabilization with Reduce Jittery Motion 5 passes
  • Fixes model directory being read-only for some Mac users
  • Misc bug fixes and improvements

Known Issues

  • Windows 22H2 users may experience a delay when they run models for the first time

Submit files here: Dropbox

NOTE: Please share your benchmark results only in this benchmark-related post.

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!


Posted this idea in the benchmarks thread but also posting here for visibility - suggest that VAI version # be included in the benchmark results output to eliminate guesswork.

Edit: The text output does include the version # already. The benchmark GUI doesn’t however so it would be good to have it added there as people like posting screenshots of it.


What’s changed with the Apollo model compared to in v3.1.8? Is it just a bit more accurate?

With the new preference for “disabling anonymous data collection” - was it previously sending anonymous data without asking?

Also doesn’t the “reduce jittery motion” use the Chronos Fast model? If Apollo is the best model for accurate interpolation why does “reduce jittery motion” use use Chronos Fast and not Apollo?

Also why on the “roadmap” page does it say Chronos Fast is better for slowmo >8x? What’s that doing differently and why would that be better? So Apollo always creates 8x the frames and discards some - so if Chronos Fast takes less source frames into account shouldn’t that mean it’s less accurate at non-linear motion? Maybe if you need >8x the frames there could be a different Apollo model for that unless that would be much too slow.

Is there going to be download links for each set of model files for this version (to save having to download >30 files for each model if the app can’t download them itself)? And if so, when will those be made available?


I think, especially for AMD CPUs, RAM speed would be helpful to show in the benchmark results. I know for a fact it makes a difference.

1 Like

Getting live clock/frequency numbers for CPU, RAM, GPU etc. are surprisingly complicated. It is on our TODO list.


I’m noticing that too :slight_smile: Should be there in next weeks release. Text is easier to parse and compare, but people will post screenshots.


Yes, it has been trained further to be more accurate

Currently the only time our servers are contacted are at the time the app is opened, user login, model download and if the GUI crashes. This option will allow users to opt out of any detailed tracking we add in the future and GUI crashes tracking.

Apollo and chronos fast work differently. Apollo always generates 8 frames between any 2 frames, Chronos/Chronos fast have the ability to generate frames at random positions between frames. This feature is needed to reduce jittery motion. So Chronos Fast is used.

Since Apollo at present can only generate a maximum of 8 frames between 2 frames. Use other models for more than 8x slowmo.

Ideally, the app should be able to download the models. If it is not, contact support and we can give you a script to download the models.


I don’t want to say obviously…

1 Like

When will we have more control of audio settings? I can only get auto to work.
If its easy for a start I would like to see 320 instead of 192 be the auto AAC default please.


you can modify that and adding any setting of encoding you want, by modifying or adding a line you want in the file encoders.json it’s in ProgramData/Topaz…/Models/

just replace 192 by 320 in this line :

“transcode”: “aac -b:a 192k -ac 2”,

1 Like

If it does turn out to be too great of an undertaking, I’m sure we’ll survive without it. Worst case, if we care that much, we can paste in the number reported in Task Manager or the like.

Apollo current focus is to reduce the loss of image quality?


Thank you for the quick response.
How would I add aac 320k encoding to V9 best.
This is the line now
“transcode”: “vorbis -ac 2”,
Would this be correct to substitute this line?
“transcode”: “aac -b:a 320k -ac 2”,

Hi this message is for Suraj. Suraj I really enjoy your product but am wondering if there is a way to optimize rendering and frame rate through better coding of the software? Some folk here said he owned a nvidia gtx 4090 and the gpu was at its maxiumum and only getting 14 fps. What kind of machine would Video Enhance need to render 60 fps? Or does this machine not yet exist. Rather than spending a bunch more money on graphics cards can the software be fixed to perform better on gpus? Thanks for your time.

1 Like

What info does it currently send if it crashes? Surely it should be asking for our permission before sending any crash reports, especially detailed ones like the stuff shown in the log files. It’s good if it’s now asking but it sending info like that shouldn’t have been done previously surely without permission (due to data protection).

Okay but the Chronos ones don’t handle non-linear motion as well from what I’ve read. Maybe there could be a version of Apollo that can output more than 8 frames, eg. double that for more precision if needed (and take more frames into account).

So isn’t that another readon for a version of Apollo that can generate more than 8 in case that’s needed, since that’s the one that’s best for non-linear motion (and I assume it should work fine for linear motion as well since I think it takes more source frames into account than the Chronos ones).

Thanks. I’ve contacted support multiple times. I’ll ask for a script for that that I can run in windows then thanks. Hopefully the script will only download the ones that are needed. I’ll ask them if they could supply that thanks.

Hi - can you please send me the original video and let me know the exact setting used? Dropbox

Hi, is there a topic that talk and or ask about futur functionnality for Video AI?
I’m asking because I’d like to propose the addition of a PAUSE somewhere. Using command line never worked for me (pause function).

1 Like

I’m pleasantly surprised no other data collection was done while the option was gone.Excuse my cynical expectations :slight_smile: