Here is a direct link to all the models and their scale ratios that I’ve discovered so far: The current framework that underpins the technology is fairly sophisticated at this point, but we need to build out the interface to accompany it better in all of our applications.
The goal is to allow users to optionally download all the models for their device within the application (as well as previous versions of each model), while still delivering the core-value proposition of a major speed improvement for 90% of our users in the meantime before that’s built. We have fallback models available that we ship with in case there’s no internet connection available. But it still doesn’t make sense to install every single one on a given system that will never use 80% or more of them. So in that sense it is good that some users might want to maintain copies of the models. That being said, from a preservation standpoint, people want to know they can use the software on whatever machines they want in the event that the company or servers ever disappear. Not arguing with you by any means, just trying to be as transparent as possible with our decision processes. Keep in mind that this is the same infrastructure used in Video Enhance AI, but I understand that it adds more up-front friction in Gigapixel because the expectations aren’t exactly the same. I agree that the messaging isn’t the best in the application about the downloading, but it didn’t make sense to push back the release by a few months when the core value proposition (a major increase in speed for most users) is ready to ship.
Once they’re downloaded, they’re downloaded, so it quickly becomes a non-issue for most users. It didn’t make sense for us to push back the Gigapixel AI release with the new AI engine another 2-3 months just to make sure the model downloading messaging / interface is easier to use. We did have to fundamentally change the model processing pipeline which is causing confusion and up-front frustration, but this would happen no matter what route we took. This is an intermediary step before the model repository is put in, but that takes time. Why ship with 10GB of files when most users will only use half of them? We can add that to the online installer, but then there’d be no way to support a full installer - it fundamentally changes the way we package and ship our products. The 100GB is for all supported hardware at every scale. There’s no way to just drop the models for your specific hardware during install. I find it hard to believe that 100 GB is what one needs for enlarging photos.
0 kommentar(er)
