Topaz Video AI v3.0.6

Yeah, NVENC (for TVAI) requires the newer implementations, that support B-frames. Your card will not support that (nor my RTX 3080 Ti, btw). The rest should simply work (not sure about VP9, though).

Also, they removed libx265 from their own version of ffmpeg, for GPL reasons. Valid,an und für sich, but then why go the ffmpeg route to begin with? I’d rather seen they had made their own exe take its own parameters, with the ability to output to stdout, (like VSPipe, from VapourSynth, for example), so you can pipe the output to whatever program you want (like an x265 command line, for instance). But hey, that’s just me. :slight_smile:

It should work, but it doesn’t. :rage:

If they removed it, why is it still in the software as an option for exporting then? That seems a bit…off. Granted, if they were in a real pinch timewise I could see it as a quick fix to be adressed later, but still a very strange decision.

Absolutely fine if they don’t want to have 265 as an option. But presenting the option to me, sets an expectation. Granted, it would feel a bit off to see premium prices and not have the most common file formats in there, but that would at least be more palpatable.

That is really interesting though, and more than a bit strange. I mean, sure there will be growing pains with the kind of success that they are having with it, I get that. But you’d think that licensing would be one of the first things to look at when you get to a stage where you have some financial wiggle room.

But yeah, I’m with you on 100% on that. It should be a given that this sort of product should have it’s own dedicated code base for encoding. I mean, it’s not like it’s hard to do either, hell even I’ve done that. The only time consuming part in terms of coding it is to look up what part of the stream you need to go to to read which data for the various file formats. The rest is easy.

I’d be lying if I said that this level of coding doesn’t scare me. Quality of the release aside, video formats are a genuine vulnerability, and with these sort of mistakes I don’t feel confident enough to say that they havn’t got their own security issues or haven\t gotten a library from an old blog with unmaintaned code and a soon to lapse domain name, or something similar.

It very much feels like the core idea is to take a free AI & FFmpeg and package it as a new product. I’m sure they haven’t just done that, but that is what it feels like.

Why not a raw format? Or webm if they want a free one?

In this case I don’t get why they don’t go with older more stable releases with wider support. Or at least have the good old option of an experimental, new fantastic features version and a stable release one.

That would make so much more sense.

At the end of the day, I could not care less about using the latest graphics card fancy shamncery, if it means I can’t run it on any fairly new machine. When given the option it’s not like people will see broken software, and go “oh - I can fix that, all I need is to go buy brand new $2000 hardware, and that will fix this buggy program!” over simply downloading something that works.

1 Like

Laziness? :grin:

Both x265 and libx265 require the GPL license; ffmpeg itself uses the L(esser)GPL license. So, for including xlib265, if they incorporate it, their derivative tool using it needs to be in the public domain too, and have its source code published. As a result, they now only support NVENC.

I’m pretty sure TVAI’s own A.i engine is very much not in the public domain (but no doubt based on common A.i knowledge).

Daniel you want a result test player and and standard vlc is not the best. I run MPC HC with external LAV video decoder it eats just about anything.

little off topic. Does anyone remember that you tube guy that did all those crazy deep space 9 and other videos with topaz? I think him name started with a K. It seems he ghosted youtube or went private and i wanted to him to see if anyone has any newer guides on how to get the best quality from videos.

With that being the case, you might as well output png or tif, then encode with Selur Hybrid (or whatever your encoder of choice is).

1 Like

Can someone kindly assist me with the question I asked? Please? I really would appreciate it!

I would suggest looking into Neat Video as it is the best video noise remover with extensive capabilities in all manner of noise removal. It comes as a plug in to most NLEs. I now use it for most all of my noise removal before going into TVAI.

I disagree with a lot of what you said. Digital video formats and their containers are the most convoluted and complex files in existence. The idea seems easy, show pictures in sequence fast enough to fool the brain, but the history of television and the means of recording it, complicates it even more. Their choice to use ffmpeg, gained them the decades of dealing with digital videos that cannot be gained any other way. It would take them decades to get to the same level. And even with that, there is no end to the requests on this forum about supporting more formats and features. (I only use Bink format because it’s free, why doesn’t TVAI support that? Then someone else pipes in about it not supporting some format that only gets produced by some ultra expensive video software.)

You probably got it backwards. It’s usually management that forces per-mature releases. I think they know by now what the consequences of that are for them.

AI is one of the most intense processing things one can do on a computer. Videos are some of the largest files. Combine AI with videos and you get hefty hardware requirements. Until the M1 chip, laptops have not been made with AI processing in mind. Maybe if you bought a top of the line gaming laptop, it would swallow it. The mobile GTX 1650 was the bottom of the stack when it came out. The ti variant, just one step above that. It should not be surprising that a light hardware equipped laptop couldn’t handle the stress. Even if it could, it would have to be left for weeks to process one movie.

I think the one thing Topaz could have done better with TVAI is stressed and made very plane that you also get VEAI. Several people on these forums have stated that TVAI was not worth the money they spent on it. I have yet to see someone say the same thing about VEAI. (Minus before the release of TVAI. There were regular posts about how it didn’t work with a particular video or it didn’t support this or that. [It was probably all those posts that lead them to create TVAI in the first place.])

This results are really strange… :unamused:

The real question for me is what was LTT running as a benchmark? His Topaz slide does not show enough information IMO and it should show lower tier cards which it doesn’t. The lowest is a 4080. I would love to have a copy of the video and the preset used for the benchmark. What version of AI was he running as so many have come out in the last month if you include betas? What hardware was the test setup? I must have missed it.

2 Likes


why can not login? Apple M1 Max mbp

Question is, can you install VEAI 2.6.4 next to TVAI 3.0.6?

1 Like

yes you can

2 Likes

See. They should have that explained all over the place—So that after installing TVAI 3, the user knows that they have the option to also install VEAI 2 and they know where to get it and why they might want it.

1 Like

Not Really.

7900 XTX has 122.8 TFLOPS FP16,
4090 has 83 Tflop fp16.

No wonders here.

So even if the TI comes out, its still slower than 7900 XTX by 22% in rough performance.

3 Likes

I am using a VapourSynth setup (Windoze), with a pipe to x265. I had found a ffmpeg way of processing png files, like -i %06d.png to the parameters, but I can’t have my cookie and eat it, as I also need the -i for something like -f vapoursynth -i “f:\jobs\test.vpy”. Will take a bit of sorting out. :slight_smile:

See, what I don’t get, is that my CPU (i9 12900k) is already at near full load, keeping my graphics card (RTX 3080 Ti) busy for like only 20%. My point being, what would a faster card do, when the CPU is already at capacity? (Aka, cannot process the GPU data faster)

What software do you use to monitor the load of the gpu.


update:

Here is the comparison between MSI Afterburner and Taskmanager for the workload detection.

The task manager does not even show half of the half although the GPU is fully loaded.

i did post this whyle the 3.0.5 beta.

The taskmanager did readout every time the gpu was at idle load between the frames.

But you need to set the afterburner to 100ms readout.

4 Likes

I normally use the Task Manager; GPU-Z gives yields slightly higher values, with peaks to 86%, but not sustained.