Stable Diffusion 1.x is trained for images of size 512x512.
Stable Diffusion 2.x is trained for images images of size 768x768.
Therefore, I propose we default all models based on SD 2.x to 768x768. This will result in higher quality images for these models.
From a personal perspective, I also think this is important because there is a popular narrative that SD 2.x is worse quality than 1.x. Generating images of the wrong size doesn't help with that 😆
Important detail: only logged in users can request images of this size, so we should prompt for login when selecting one of these models.
Stable Diffusion 1.x is trained for images of size 512x512. Stable Diffusion 2.x is trained for images images of size 768x768.
Therefore, I propose we default all models based on SD 2.x to 768x768. This will result in higher quality images for these models.
From a personal perspective, I also think this is important because there is a popular narrative that SD 2.x is worse quality than 1.x. Generating images of the wrong size doesn't help with that 😆
Important detail: only logged in users can request images of this size, so we should prompt for login when selecting one of these models.