Topaz Video AI v3.0.4

I’m doing very much the same thing resurrecting ancient DVDs and tape. I have found that doing the upscaling in steps does give a better result. And, the first thing you will need to do is get your source truly deinterlaced. Next pass, you can work on taking out noise. Please be careful not to overdo that step because you can remove detail, too. BTW: Initially every old, low-res image is different, at least until the clean-up phase is completed.

After you have a clean, progressive (hopefully lossless) 100% image, you can start enlarging in small increments. Initially, don’t push the sharpen/detail settings too hard.

A few runs and some trial and error are necessary experience to get the hang of it.

1 Like

17 Seconds for 22 minutes is awfully slow, although you may have to check your job settings and look for possible bottlenecks (constrictions) in your system.

Are you using the Nvidia-specific codecs to render with? Depending on your enhancement and upscaling settings the frame rate will vary. Typically, you should be able to get a frame rate between 5-18 fps. - Some operations can happen much faster than real-time…

If you can send up a brief description of what you’re trying to do and some input and desired output information that might be helpful so we can help.

Let us begin with the foundation, the old V2.6 and even the V3.x are not even build with the M1 CPU FULLY addressed. I have this info from good authority that if using a Mac M1/PRO/Max and even Ultra the speed is almost the same…so on the foundation level if you’re using a MAC system, forget about speed enhancing.Even running on the M2 the speed improvement is almost zero.
The same goes for the PC market, as you mention, even a 4090 (price!) will not make big shifts on deliverables.
So many companies make ‘enhancements’ and adding ‘features’ without looking at the BASE in their code.
And as it goes …after adding more and more features…THEN (after client feedback) they look how to ‘improve’ their application.
These kind of heavy calculations applications (see Davinci Resolve who by the way, FULLY supports all M1/PRO/MAX/ULTRA & M2 cpu’s AND Nvidia high end cards) demands heavy duty from these components.
I fear from what I read in all the beta users that Topaz is busy ‘tuning up’ the too fast delivered production version on V3.x.that’s good but not on the foundation level.
So if are where you, wait on investing a lot of money on 4090 until they find a way to fully optimise their application code for the components that can be fully addressed to high demand tasks.
This is my view of course, I hope it finds a way to the dev teams to re-allign their focus on this product.
The potential is there, the goodwill also, so come on dev teams, surprise us with a good performant application :- :blush:

Thanks! One thing I have noticed is when exporting/previewing an avi file, even with a 60fps setting, the result doesn’t look as smooth as an mp4 of the same video. Yet the avi detail looks far better. Maybe this issue is individually variable?
I’ll try out your tips. Thanks again.

Do you use a utility called MediaInfo to view your files? The difference you’re seeing could have to do with the bitrate, whether it’s interlaced or progressive or some mix of the two. (And yes, there really is such a thing as progressive, interlaced video.)

If you have video in files copied from tape or DVD, what is the MediaInfo on them? Their original state will dictate what you must to first to prepare them for enhancement.

1 Like

Thanks again. It’s so addictive trialling with the settings. Every day’s a school day!

Do you have MediaInfo? It is available for both Apple and Windows machines. If you don’t have it, you need to get it.

Here is the link: MediaInfo Website Download Page

1 Like

Thank you for this info. I’ll nab it tonight. :+1:

Good. It’s like a flashlight in the darkness.

There is also good documentation on it, but there are a few leetle details that make it especially valuable

1 Like

You answered the wrong person. Stop sending me useless messages!

I know this is a tiny deal but when processing, the fps rate is stated as spf with prores selected.
Don’t want anyone to get a sunburn out there :smiley:


Hi Paul, in settings, you can choose either seconds per frame ‘spf’ (without sunburn😉) or frames per second ‘fps’.

My apologies. I will try not to do it again.

For all who thinks that this new models are slow: try to enable tensor rt - in my case it speed up model for 50 %.


How to do that?

I followed instructions from Installation Guide :: NVIDIA Deep Learning TensorRT Documentation.
But I am a programmer so I used this manual, and I dont know if there is some other way or source for this runtime.

1 Like

:scream: I’m afraid of pythons and other snakes. :snake: 33% off.

I might try this after doing a backup of my system.

I also found this page which might be useful.

Also, do you know if there’s a list of supported Nvidia Cards?

I would think that my RTX 3080ti would be a candidate.


I am running it on RTX 3060 so yours is fine (using windows)

So just having that installed, speeds things up? Seems odd that it would do anything… unless ffmpeg is just silently trying to use it and than doing work-arounds when it doesn’t find it.