Topaz is a LLM using Generative AI that means it uses a Probability if the outcome is not what you expect the probability is wrong Keep in mind this applies for anything you do no matter if it is copilot or Nvidia Chat with RTX from the 3 Copilot is nearly magic and in time the proper probability match will also improve eventhough subjective become acceptable for even the noisiest videos.
That sounds like the response of someone that has never tried to make their own AI for anything. Not a bad thing, just not the most informed. Sure, looking at the outcomes of AI results, they can be reduced to probabilities. But also if you run a movie in TVAI 100 times without changing any settings, it will yield the exact same result all 100 times. So the trained logic of the AI is solid and will have a 100% probability of yielding a fixed result given a fixed input—in other words, not determined by probability.
With AI, there is always the chance that it will give what we call perfect results for one movie, but always fail at another. This is because the desired results conflict with what the AI logic has been programed to do. The only way around that, is to make a new AI model.
1 Like
It is true that most people do not understand including you what generative ai is only when we understand that generative ai is not about factual truth but a probability we will understand the learning only enhances the probability but it is still a probability most including the board of directors fail to understand this
If I understand correctly how TVAI works, the app on our computers is not using AI or probablity, but the models it’s using were created that way. So the models are built on the logic that A+B has equalled C X times in the past and probably will in the future, but the app itself is using the models as rules, and so in the results we see in our enhancements, A+B will repeatably equal C.
And that’s what I tried to explain.
@annemartensa, Until someone with authority from Topaz says some, or all, of the models in TVAI are Generative AI or even Large Language Models, I will doubt that they are. (Since those usually have to run on super computers.)
1 Like
Well, on a small screen, sure. But there are still things you can notice. Like there is a scene with a forklift where it just goes over bunch of pipes like it has no depth or weight, like its just backdrop, and than when it hits some worker it bounces off as if its made out of concrete. There is a scene with puppy dogs that juts walk right trough each other. There is a scene with a dog going from one window to the other defying all laws of physics. There is a scene with a bunch of workers in the desert trying to move a plastic chair and they pick it up and than chair starts to flout in air and spin. There is a scene with a an Asian looking girl walking in a city and there is rain and reflections on the floor but only around her, while the rest of the scene, no doubt made from other sources, is not wet. Things like that. I think you can find problems with every single example they shown. Some are really bad. To me it looks like the same problems that plague all AI models are still there. So they have just thrown more processing power and more data mining at the problem, but fundamental issues remain for the time being.
Where this kind of technology could be useful is in skinning or texturing real 3D models or live footage. Matching color and tone, things like that. If we are lucky maybe also perspective. Also some of the videos in the promo they released are almost certainly just copy paste of actual videos made by real people with some smaller modifications. Its quite obvious. And that is the problem… mental incest of sorts.
“ChatGPT and similar generative AI is a statistical representation of things found on the web, which will increasingly include ITS OWN output (directly and second hand). You post something picked up from it & it will use it to reinforce its own knowledge. Progressively a self-licking lollipop. Enjoy [#AI]” - [Nassim Nicholas]
Sort of like inbreeding, and the recessive gene disorders that come with it?
2 Likes
Yes, exactly. If the governments and big corporations can steal everyone’s work with impunity and charge for it, it discourages people from actually innovating or investing in some skills that would be done manually, and learn by that method. And the AI data mining can steal or acquire more than people can produce manually, so more people will simply rely on generative AI, leading not only to shortage of new material at some point and more and more of AI generated images and videos, no doubt new AI models will be trained on derivatives from what was one human labor and its now either entirely AI generated or with the help of it. Leading to , exactly a kind of digital mental incest, trapped in its own creative or ideological echo chamber. I don’t see a solution for that being on the horizon.
Regulation won’t work because quite simply: “When buying and selling are controlled by legislation, the first things to be bought and sold are legislators.” P.J. O’Rourke
Only free and open market forces can force something with competition, but that won’t happen because governments and corporations are afraid to lose control. We already see this by them promoting “think of the children” and “for your safety” and “disinformation” campaigns.
Another big issues is inflation. Too much supply of generative images and not enough demand, leads to the same problem US goverment and central banks have when they print dollars like there is no tomorrow. They devalue the currency. And in a way for humans, images are a currency as well, for communication and culture. When we produce way to much with ease by just typing in some text into a box, we can’t consume that much, so pretty soon images themselves will lose its potency. Just like fiat currency like dollar has. Same problem. I don’t know how to solve that, I just know its a huge problem that is not getting much press about it.
1 Like
SHI* 
1 Like
I see the same problems coming and can already see them in my behavior.
1 Like
Indeed. Me as well. I try to fight it by exposing myself more to physical things, going to galleries and museums as much as possible, drawing and using physical prints as much as possible, but its hard to escape the over saturated digital world with so many images. I don’t even take smartphone with me to take photos, I use proper camera, to limit my output to only things that matter. But all that require a lot of effort, something that we didn’t have to do, simply because there was not so much being produced before. I don’t know what future holds, but its a mix bag for me. Some really exciting tech and some really scary tech.
P.S.
There is of course also the problem of AI bubble, where everyone thinks AI is solution to everything so everyone is investing in it to not be left behind, but we have seen this with bitcoin, dot.com bubble, Tulip-mania etc. Maybe after the latest AI bubble bursts we will see who are the actual AI companies that are worth anything and not just hype. Despite my gripes with Topaz I do appreciate that I can use it on my machine and have some level of control over tweaking images and videos before output. That is the kind of AI I welcome, that kind that makes repetitive, manual dull jobs easier and less costly. Also there is a lot of great old school content that can be restored with something like Topaz Video AI and that is welcomed , since just like we talked about, the originality in entertainment space is hard to come by these days, so older content restored/enhanced, is in a way new content once again. Some great BBC documentaries in SD resolution. Amazing content from late 1990’s and with Topaz Video AI its better than it was shot. That’s a good thing. They don’t make them like they used to, do they? (sorry for the cliche line)
4 Likes
Hello,
For those who want, I have created two presets especially for Star Trek DS9.
Star Trek Deep Space 9 bis.json (2,2 Ko)
Star Trek Deep Space 9.json (2,7 Ko)
Here’s a quick look at the quality on the right
The details of the garment are not bad.
the model used in the screenshot is a combination with Iris v2 and Proteus 4
6 Likes
Amazing human reply better than any AI program will ever be capable
2 Likes
I’m hoping that AI peaks and crashes quickly. Maybe then we’ll finally be able to buy GPUs at reasonable prices again.
4 Likes
What are these “QuickTime” new encoders about? Should I use them? I usually enhance real life footage or CGI.
-
Νo previews in exports.
-
No fast previews in preview mode.
-
Annoying thick stripe under preview window.
-
Complexity in enhancements.
-
No pause processing in exports.
-
No auto deleting temporary previews.
Which exactly are the improvements from version 3 ??
Instead of improve the software , constantly downgrade it
from 3.5.4 version and later. What happening?
4 Likes
Плагин для паузы уже довольно долго существует.
Файлы в формпте mkv можно просматривать в любое время.
- Зачем?
Not as an inbuilt function, but merely a „hack“ provided by another user to us users.
And it is just temporarily stopping the encoding tasks but not freezing resources (RAM/VRAM), so only partially helps.
