This doesn't strictly have to do with stable diffusion, but I really really hope that someone eventually creates a service that hosts live interfaces for all of the legacy image generation models. The turn around on these things is wild, and I worry that unique image generation techniques will become antiquated and forgotten.
In case with SD, you can download every version yourself. The main code is in git (you can go back to any point), and weights are versioned, I have every version saved, for example.
Sure. It is probably hard to parse because I do not have solid knowledge regarding image generation AI stuff.
To restate my previous sentiment, I am quite attached to image generation with GPT-2, which is now legacy. Even though GPT-2 is not as advanced in many ways as more modern alternatives, I am artistically attached to its' shortcomings/artifacts. I really hope older image generation methods like gpt-2 remain available as a service to the general public.
