Video Enhance AI v1.9.0

I wasn’t saying anyone was incorrect. That’s just what I read in VEAI.

The clarification is appreciated. Thank you for elaborating.

That’s odd. I get a faster speed with a ‘lesser’ card, the RTX 2070 Super. I get 0.39 sec/frame upscaling 720p to 1080p with Gaia-HQ. It does 1080p to 4k at 0.85 sec/frame. That’s with VEAI 1.9.0. My system is Intel i7-7700K, 4.2Ghz, 48MB RAM, SSD 256.

I have heard that VEAI isn’t fully utilizing the RTX 30 architecture yet, perhaps that is the reason?

I get faster processing speeds with downloaded models, but I was under the impression there was no difference in quality, and was actually told that by more than one Topaz developer. It would be interesting to demonstrate any difference by comparing output frames from only the base models vs. the downloaded models, if such a difference indeed exists.

I believe tensor support was enabled at one time, many versions ago, and then removed.

4 things that to improve Video Enhance AI.
First: The artemis, dione and gaya filters could be improved a little more.
For example, Dione TD does sometimes weird things on faces or people in general. Sharpness and strange. Artemis lq is a little too sharp on some details. Artemis HQ v10 lets a little bit of grain through that v9 doesn’t let through.
Second: A very important point. Add the intermediate frames. For example for movies or series with a frame rate of 23,976 per second. Because Topaz Video Enhance rounded up to 23.98 and therefore the video sequence is longer than the sound sequence.
third, on Dione filters, would be to correct the ghost trace of the deinterlace.
And finally in fourth: 2nd very important point. The possibility to choose the default location of the filters. Because it takes more and more space on the “c” disc.

i did some benches with my RTX3070 and RTX3090

open that link:

short version:

today i would buy the RTX3080 since the RTX3090 gives me only 10% advantage,
but costs at least 50% more

and VEAI does not use the full RTX3090 power, yet.
not even on 3 parallel instances i get 100% GPU load

and VEAI only uses up to 10GB Vram … (on 3 instances)
the 24gb Vram on the RTX3090 what makes the heavy price difference, it not even used

Yes, I’m dreading the day they screw GaiaHQ and I know it’s coming, because they simply cannot stop tinkering with models. I too made the leap to the 3000 series too early as it’s barely any faster than the 2080ti. The only good thing is running two versions of VEAI at the same time, which I’m not sure the 2000 series could handle.

The ghosting/flashing/flickering with Dione is ferocious. Can any beta tester confirm it is improved in newer versions? It’s like having double vision or seeing strobing, sometimes.

Do you know which model you were using in your benchmarks? That’s the sort of speed I might hope to get with Artemis on my 3070, but never with the Gaia models.

A feature I would like the most is “bulk image sequence processing to video”. Like, when u import 50,000 png images, u have to manually select the upscaling setting and model for every…single…image. That is terrible. We need a way to set 1 model and 1 upscaled resolution for all those 50,000 png images at once. I am NOT gonna take years selecting 1 model n 1 resolution for each of those 50,000 png images. That’s insane.

N since Topaz Video Enhance allows image sequence imports, this should honestly be the top priority for images…

Can we please get some videos, blogs, or white paper that describes how AI Models reacts with the presets?

Like one reply said upscaling to 1440p is using the 200% denoise/deblock and then downscales the video to 1440p. Utter pointless to do then and then you should consider sticking with 1080p or go to 4K.

I personally wasted a day to figure out that the red box is a crop box and since I was trying to clean up an old TV show that is 4:3 ratio. I needed to change the preset to 100% denoise/deblock.

I don’t know how much upscaling old TV shows is part of your business, but some direction on how best to upscale with your software or fastest ways to upscale your videos would be nice.

You know, Like how to upscale from 480p to 8k videos, How to handle/prepare interlaced video before upscaling, Should you upscale straight to 4k or not, or Should you Denoise your videos before upscaling or not.

Plus I would love a paper or something saying out up upscaled Deep Space 9. StarTrek shows have both computer generated and people being flimed, so does that mean I need to chop up the show and use different AI Models or not.

These are the kind of questions I personally been having and would like some help please.

2 Likes

This kind of info is vital. It doesn’t take much effort to maintain a supported video card page that doesn’t even have to say each video card, but supported architectures, like Maxwell, Pascal…

Then list know issues link to a community tab. Like my issue with my GTX 1060 with 6GB could run the software, but not with you do much of anything else at the same time and I have an AMD 5900X processor with 64GB of ram. Since I wanted to be able to play a game or watch some movies on my one computer at some point. I needed to get a new graphics card.

Now after 1 1/2 months after purchasing the software. I’m rocking a Quadro P4000 as a secondary card. A little funky to get both working at the same time and I still need to upgrade the power supply, but I’m doing my dream of upscaling my old shows I loved to share with my kid and enjoy.

This is just an example of something that might have changed my mind on getting started on this path because I always thought I needed some really expensive computer and some expensive programs to do upscaling as well as taking days per video. If I could have easily found success stories and actual timelines. I might have tempered my expectations and pulled out an old computer that could have upscaled a few videos a day. Instead of me spending $500 on a surplus video card a business never used and me wanting a tensor graphics card would have made me need to buy a new power supply, so can’t blame Topaz for that one.

So can we get a little peek behind the curtain and set up more community managed and driven support? I can only say for the Enhance AI software, but I think you got a real quality program here. I’m looking forward to learning how to use it.

Thank you for your time.

1 Like

Alright, this may sound like a retarded question, but how the heck do I send you a PM??? My trustlevel (or “badge” level) should be high enough, but I cannot find a link or icon to send you a PM… :thinking:

well i cant remember what models i used during my benchmarks.

but i use all models, before i upscale i do lots of tests to have an idea of the final result.
nothing is more disappointing than rendering a video for 5+ hours and the result is way too sharp or too soft (wax skin) = useless. The 30frames preview i not always helpful. Too short or too small…
for many videos you need a real ~1min fullscreen sample to choose the right model.

so for sure i used Artemis AND Gaia maybe 50:50
and to be honest i never noticed a big difference of speed/spf

i will keep that in mind and see if there is more than a 5% difference during the next upscales.

Thanks! I contacted them and turns out, I was running out of VRAM. I can lower it settings and have no issues enhancing any video now. :slight_smile:

1 Like

Great :slight_smile: !

Just keep the current model. I think they learned their lesson about not changing the numbers. At least I hope so.

With this product, you take what you get, develop your own process, hold on to every version and model that works for you, and mostly fend for yourself, because no logic or help is coming from the source.

Just click on the icon/user-name and the profile of that user should come up. On the right side there is a field “message” with a little envelope. Click on that and start typing. :wink:

I can get TeslaV100, Tesla T4, RTX 2080TI, RTX 3090
Which card will be faster/good for VEAI?