The second example looks better text wise, but I think because my original source was slightly lower in quality the rest of the image ended up processed more heavily and looks crisper (at least to my eyes).
is this really ever going to desktop? You can either sell software or sell a service, and its well established now that even with periodic upgrades, selling a service is a lot more cost effective for the company, guarantees a much more robust revenue stream over time, avoid all sorts of support headaches ( you patch a couple of central servers vs thousands of differently configured end users workstations).
Microsoft figured that out. Adobe figured that out. Topaz moving entirely to a delivered service model, like O365 is inevitable. The current desktop software offerings seem to me, just to be the loss leaders to build a customer base for the inevitable switch.
Diffusion video enhancement will eventually end up in someone’s desktop app. Either the models will become more efficient, desktop hardware will become more powerful, or both. This is already starting to happen for still image enhancement.
So to maintain a marketable service product, the tech will need to grow more advanced. Like a generative AI model in which the user can give the app text instructions for modifying video as well as enhancing it (changing the season of the backgrounds, replacing entire backgrounds or changing characteristics of subjects). We can already see this happening with still images as well.
There will most likely always be something that can be done with images and videos that will be so resource intensive that it exceeds the capability of desktop hardware.
If Topaz goes the cloud rendering only way, they loose most of existing users. We will not be able to afford the high prices for our private videos. We don’t make money with our upscales
What then remains are companies, professionals e.g. the film industry, who have their own tools to remaster existing films, SDR to HDR and UHD. They do the lossless path (codecs) and they had to upload terabytes of data to your cloud, does this happen? The AI model would have to be outstanding then.
If you let die the base product and the cloud project does not generate the necessary profit, Topaz goes a very danegrous way this is a high risk agenda.
Mainfocus on mass rather than exclusive customers
The cloud should be used to further develop models and explore limit areas with the aim of breaking them down to local models with the awareness that the cloud will not be a replacement for the masses.
Ensure greater distribution of TVAI to attract more customers, todays awareness of the program is not very high. You could offer free version (without support), where some functions for example SDR to HDR and frame interpolation is pay to unlock, maybe video lenght is limited to 30min
But the free version must be full usable without watermark, it must be massive used and becomes indispensable. You have a great product, show this to the people. Then with the free version at some point they reach limits and are willing to pay, that’s the trick ![]()
I’ll add my views here too… I do not want to upload any of my videos to any cloud server, period.
I’m more than happy to keep paying the annual fee to use the software on my own machines only.
THIS!
You’re not the first to mention it. Don’t worry. I was mentioned by someone the same day it got released and that person is still here commenting.
compared a clip from lq 720x480 enhanced with Dione and Iris to 1080p; the Starlight result appeared to me much better and realistic, and the way it handled trees, hair, clothes looked very nice, faces too. But, echoing some of the above comments, definitely waiting for the local versions of this exciting new model (I don’t mind if upscales are slow, mine already take about an hour per 10 seconds anyway.) … some of my family films with toddlers are not something I would feel comfortable uploading to cloud. But I am glad to read about these new AI models and look forward to upgrading to them in the future.
I’ve noticed that for several versions of the software, it always leaves a trace in the Windows registry of the previous version. I always have to delete the entire registry folder of the old version.
Yeah, I feel like once we get a lighter weight Starlight model locally, it will be icing on the cake.
I want nsfw videos are allowed in Starlight?
free version uses videos for training, I dont want to see porn images on wedding videos…
use normal video ai for nsfw content, horny boi.
Cry in another place, married boi. Its not porn, its fashion Topaz Compare
By the way, devs, here we have problem, the manicure on girls nails is gone (see the last frame)
When Adobe went to a mandatory subscription based model, they lost a vast majority of hobbyists and home users, who could not justify paying rent for software. This is why I’ve been on CS6 for almost 10-years, and this is with an Adobe employee discount through my sister. Even a discounted subscription though my sister is not worth it. Adobe mainly caters to corporations and governments, and that’s their bread-and-butter now. They lost their home users and hobbyists forever.
Does TVAI… is it supposed to alter the colors? I’ve notice when using proteus that the color isn’t the same. That being said, does it do anything to HDR? In my case because HDR doesn’t play well with the application of grain at all on anything white-ish

I tried exporting a job that included HDR… I then took that exported file and ran it through to do proteus and grain… The colors get muted… Is there a work around?
Acrobat Pro XI here. In addition to the subscription pricing, I haven’t seen a single new feature added since then that interests me.
My wife uses the free Acrobat Reader, and the first thing we did when we downloaded it was turn off most of its options. Classic case of over design.
NSFW is pretty loose categorization. A lot of content considered not suitable for your work actually is someone else’s work.
i do have a feeling that this diffusion model size is going to be larger the a terabyte
If you don’t want to, don’t look. And I want to!
It would be interesting to know how large the model file for Starlight is compared to a desktop model like Rhea. Wouldn’t tell us anything useful, but still interesting,
