Imagine an AI model that takes an image or a series of video frames and translates that data into a text file containing generation instructions for Stable Diffusion... Assuming the text file is much smaller than the original file, that might be a huge boon for file compression. Images could be saved as text files. Opening one of these files would have Stable Diffusion generate it and put it into a temporary directory, memory, a cache, or whatever. This could introduce interesting tradeoffs. One could either decide to store image based data on hard drives for fast access or use a GPU & SD to reconstruct the data. The GPU way would probably be a slower way of accessing your data but it would save storage space. The other thing to account for would be the size of the "translation model" and the Stable Diffusion model. This is all just a fantasy though. I certainly haven't thought it all the way through. Someone could already be working on something similar. It also might not be possible. Maybe the idea itself is too inelegant and a little convoluted for something as simple as storing images. But the idea itself gets me thinking. 🤔