Topaz Video AI v3.0.7

Looks better on 2.6.4
There are a bit more small details.


Absolutely agree with everything that’s been said. Topaz should not pull a Skylum on a software that used to be exciting to use and was kind of a revolutionary tech when I first experienced it.

Now it just comes across as a gimmicky facade; being mediocre in its output coupled with that awful preview and unsatisfactory speed for rendering videos.

Whatever changes have been made under the hood need more than a tweak to attain the level of performance which was present in the past version of VEAI. You can do it Topaz!



Out of curiosity, what does the “add noise” sliders for? How is that different from the Grain setting. And finally, why would anyone want to add noise to video that the AI in this application is intended to remove?

Another Question that I find bothersome. Does the 2X deinterlace do any denoising as part of its process, or does it just bake it into the image more? I have discovered that it does give better results to deinterlace to 2x, Take the output and clean up the noise and enhance the double speed video while interpolating it down to it’s original framerate.

The real question is: Why isn’t there a reduce noise setting available for 2x Deinterlacing?

If anyone knows the answer to these questions, I’d appreciate hearing them.

1 Like


I think the evolution to a fully functional TVAI 3 is still in progress. I’m hoping it will continue to get better and faster. It has the potential; I believe they are gradually making it real.


As far as adding grain I only do that on old pre-color films that I know had a little grain to begin with. Usually in restoring some of these very poor quality copy of a copy of a copy you end up with a overly soft or flat look. I mean I remember going to the local theater in the early 60’s and watching many old films and ‘stories’ as some where called. The grain was there in many but not in your face grain so restoring some of these videos to get that original feel I add a very small amount. It is very trial and error and it doesn’t just involve TVAI but the pre and mostly post processing for each trial to get the right amount. Most color films from better studios rarely had visible grain when watched at the theater. It kills me when I see large grain all the way through many restored color films. I think most of the time the restorer is just covering artifacts that were not delt with properly.

1 Like

I’v found a solution to this problem for DDV3 model:

It does not happen if “Square pixel” is selected in “Video out” (instead of “Original pixel type”).

1 Like

It is related to the file format and the compression of the pixels. My experience to this day is to convert programs of 30 minutes with v.3.0.7 to a solid format.

I have been rendering Betacam Digital material every day without any problem…the notes sent by Gregorio.Maddra are executed correctly.

I review well and pass the rebel frames or sequences to V.2.6 4 in Tiff format and this program has enough tools to solve most of the problems.

To measure and quantify background noise I use Neat Image Pro, it has the necessary tools, There is a free trial version.

I remember your phrase that in computing garbage that comes in, garbage that comes out and therefore the old intertwined images and many other problems that accumulate, must be diagnosed in order to apply the correct medicine.

If you have a rebel sequence, render it to me in tiff of the original

with v3.0.7 and send it to me. I will return it to you with my executed diagnosis,

I have been in post VFX for more than 40 years and I am indebted to you for the interesting contributions made in this forum,

Regards Disculpa mi spanglish .

Absolutely brilliant App and love what you guys are doing. I have a few requests. I’m using the software to enhance some old Cine 8mm films that have been scanned and some of the frames are damaged beyond repair. The film also has a lot of dust, scratches and hairs etc…

Part of the enhancing process I want to increase the frames per second and when the software comes across dust and damaged frames, it tries to incorporate that damaged information in the extra frames generated.

It would be good if there was a frame repair option, that allows us to manually select and generate a new frame to replace the damaged or missing frame, based on frames pror to and after the damaged one.

Something that will identify hairs, scratches, dust specs, based on whether they are visible in a couple of frames either side.

And a feature that would allow a dynamic adjustment of exposure of a scene, as old cine films can go from under exposed to overexposed at times, something that would smooth that out would be great.

All of this would need to happen before the software carried out any work enhansing detail and increasing fps

Perhaps it needs to be a completely new App for cine films with features specific to those :slight_smile:

1 Like

And you aren’t even listed as a Beta Tester. Is this bug listed?

Or maybe to prevent color banding on compression when transferring to BluRay? In my own processes, I use a closing VapourSynth line to the script, like

vid = core.neo_f3kdb.Deband (vid, preset="veryhigh", dither_algo=2)

This will add grain (on gradients only), so color banding won’t occur, as a result of x265 (or any other compressor) thinking the colors on a gradient are too close to each other, and starts ‘eating’ them: aka, cause the infamous color banding.

Still don’t really get it, though. Studios have the best of the best equipment: no digitial noise should suddenly appear, accidentally, anywhere along their processing line.

Dev has said in here that it will be fixed in 3.0.8, and mentioned the bug thing when I sent in my logs to help resolve it.

I do this as well but only due to Audio being degraded as I do multiple renders in TVAI.
But I have never had delayed or audio trouble since using Transcode for Audio settings.

1 Like

Do a search for film restoration software. You’ll find a ton of programs that do what you need. Some are even free. This software is not designed for film restoration. No offense.

Film Restoration and Upscaling Video Footage are totally different applications.

You are completely right. It is vital so why is it still broken? I can’t use broken software no one can.



I have been tinkering with 3.0.7 on a a few Mac M1s and I am impressed with the progress.

It was odd that when I started the software, it briefly opened a web browser to and then closed it. Kind of caught me off guard. Actually – I am very uncomfortable with this software pulling up a web browser and doing something. Why is this happening?

I am definitely seeing a slight speed increase now on running parallel processes (at least with the same or similar options I was using on 2.6.4). I think I am – it is late here.

The interface looks good, feels more natural and I am excited again. Great job!

Eeek – oh - as I write this – perhaps a bug?

I am running two instances and it seems the split screen has a mind of its own with the first instance.

1 Like

Cant run proteus for any video. It shows model failed and AI error only.

Curious you captured via SDI right?

I’m starting to realise it’s not quite the software I’m after for cine restoration. It’s got a lot of features that I need, the AI resolution improvement is amazing, as is the stabilisation and ability to add extra fps, so it pretty much does most of what I need. These extra tools would make the software brilliant, but you’re right about
video up scaling and film restoration being done differently.

It’s perhaps an idea to extend topaz’s range of software as the core technologies seem to be there. In the meantime, if you know of any cine restoration software for a Mac, that stabilises, dynamically adjusts exposure, removes most dust and scratches. Allows creation of damaged frames based on frames either side, upscales and then increased the fps (using an optical flow technology and not just frame blending). I’d be very interested, because I’ve been looking for ages. :slightly_smiling_face:

My understanding is that “Add Noise” injects pre-process noise which thereby reduces the amount or severity of the chosen Enhancement Method by reducing unwanted artifacts or over-sharpening of certain features present in the video, leading to a more acceptable and pleasant result. This “Add Noise” is not the same as “Grain,” which adds film grain to the output, but rather it utilizes noise prior to processing, which helps prevent the AI from over-processing and causing extreme artifacts or excessive sharpening of some video elements that should be either left as-is or more gently processed. However, if you increase this “Add Noise” radically, you may end up seeing degradation of the output. I use it every time I think the AI process is going overboard by creating artifacts like striations that were not in the original video. Typically, I use between 2 and 5.

1 Like

Anyone know when 3.0.8 will be released?

And will it fix.

  1. unable to activate license after uninstall/reinstall - it asks for browser but never evidently contacts/activates via license server.

  2. RTX 3080ti Proteus model fails on 3.0.7.

Reduced to using 2.6.4 but audio fails to come across on it a lot.


Below does nothing and will not activate license after uninstall/reinstall in order to try to fix “model failed” in 3.0.7 using RTX 3080ti - card.