Topaz Video AI v3.1 – faster exports, faster playback, and new Motion Deblur model

Just whait, i’ve found something that does slow down AMD cards and maybe GTX cards.

I hope that the developement will respond today.

Maybe you will se a 4x increase in speed when its fixed with AMD.


Feature? It was always there before and now it’s gone.

Then stick with v2 until it comes back.

I was correcting your post.

Well, if you want to be pedantic about it, it’s still a feature, just one that’s missing (along with the model manager) from v2.


that is great hahaah i was start to think that i was the only with this problem

MacBook Pro with M1 Max and 64 GB ram is getting memory issues that weren’t happening before this version. I sent details via Dropbox.

1 Like

Any chance you could point me to one of these with libx265 encoders? and/or a guide on how to do it/run it/use it/anything lol

2020 iMac,128 GB Ram, 10th gen 10 Core i9, 5700XT 16 vram, source dv to progressive, enhance and upscale from 640 to 1920 with both Artemis low quality and and Proteus manual. Source on very fast SSD and Exports on Very fast SSD. I’m am seeing amazing performance and quality improvements. I’ve seen 8 frames per second exports when previously I might get .something other output. That’s incredible. Also, I’ve been working on a creative directors spots and the problems with faces for example has dramatically improved. He may be able to show at 1920 progressive what was once 640 interlaced to clients now given the Improvement you folks have made. Well Done! And godspeed to further quality improvements. Beta v3.1.1.1.B. Regarding the Motion Deblur Model, I’m having no luck with it. It’s still killing faces-lots of ghosting artifacts. Thanks!

I compiled the ffmpeg with libx264 myself using the instructions in the file which can be found here: Release Topaz Video AI Redistributables · TopazLabs/FFmpeg · GitHub, specifically inside the file.

Warning: it’s a lot of effort and it helps if you’ve got scripting/coding experience. Not for the faint hearted. If I have time this week I could have a go compiling for libx265 if you don’t fancy it.


If you compiled a version of FFMpeg with the Topaz libraries + x264, x265 and AVS script support you would be a hero to many in these forums. To be on the safe side a link to a page with a link to the binaries would be best to avoid any problems for Topaz rather than linking anything directly here.

Thank you so much for the starting points! If you do have time and are willing, I would be immensely grateful if you could compile one with libx265 (and especially if you could document the process of doing so so that I could keep it up to date as topaz updates come out over time). Additionally, I’d be more than happy to provide some type of compensation for the time/effort

I am transferring VOB 480 interlaced (29.97 framerate) files of 80’s and 90’s videos which are recorded from analog sources. I am trying to upscale using 2x deinterlace to FHD 59.98 frame rate. I have tried all the presets and seems they all look the same. Anyone else doing these types of upscales?
I also tried keeping the framerate at original 29.97 the using chronos and the other ones.
Still unsure of the results.

1 Like

QTGMC! :slight_smile: dione is bad…

1 Like

I watched and hour video on how to install this. I don’t have the time for that.
I tried hybid and it wouldn’t work. How did you install this?

Is it normal to get a smaller video window after upscaling?

Yes, I would love that for exactly the same reason.

As of 3.1.1, this feature has never worked.
While tasks are running, running “powercfg /requests” returns None.

I have a AMD Radeon 5700 XT and 10 Core, 19 (10 gen) iMac 2020 OS Monterey and I have found the last Beta Version to be up to 3 times faster (although slow with Deblur selected) using the 5700 XT. Could you tell me what circumstance that you found it slower with an AMD? Thanks!

Sorry, Eric… Although I like the direction you are generally taking with TVAI 3.1.x, the current implementation of Themis motion deblur is nowhere near the almost magic sharpening of your photo app Sharpen AI. The Themis model needs either more training, retraining, or perhaps an adjustment slider to temper it down. It is much too aggressive for most videos.

1 Like