I've just come up with an idea, let's call it the 'Infinite Canvas', which might be an app or a website. Users can zoom into any area of the pixels and enter an infinite world with new scenes. I think the coding logic is quite simple: it automatically determines the center of the area where the user zooms in, gets the outline mask of that area, asks GPT "what this shape looks like", sends the resulting prompt into stable diffusion, and draws a picture with the outline as a reference, then pastes the result back. This way, an endlessly zoomable world can be realized.<p>But now I want to run stable diffusion and the outline mask on the user's own device (to save costs), and aside from the web GPU, there seems to be no good way, and the speed would be very slow.<p>Any ideas on this?
It's sort of novel, but its not very engaging. It's like zooming-in on fractals. After about 2-minutes you are ready for something else.<p>Here's 10 hours of zooming, see how long you last, <a href="https://www.youtube.com/watch?v=zowLNSKyfI0">https://www.youtube.com/watch?v=zowLNSKyfI0</a>
How large database do you expect for this canvas? Seems like a rare example when a blockchain might have a good use for this, all you need for this approach is to implement Byzantine fault-tolerance mining on this framework. But of course, you need to serve some txs in that case, aside from your creative content.