ControlNet Tile image upscaling
Stable Diffusion ControlNet is a neural network architecture designed to enhance diffusion models through the incorporation of additional conditions. The concept behind this model stems from a paper titled Adding Conditional Control to Text-to-Image Diffusion Models [3], authored by Lvmin Zhang and Maneesh Agrawala in 2023. For more details about ControlNet, refer to Chapter 13.
ControlNet bears similarities to the image-to-image Stable Diffusion pipeline but boasts significantly greater potency.
When utilizing the img2img pipeline, we input the initial image, along with conditional text, to generate an image similar to the starting guidance image. In contrast, ControlNet employs one or more supplementary UNet models that work alongside the Stable Diffusion model. These UNet models process both the input prompt and the image concurrently, with results being merged back in each step of the UNet up-stage. A comprehensive exploration of ControlNet...
