Diffusers adds ControlNet support via StableDiffusionControlNetPipeline and ControlNetModel

AI Impact Summary

Diffusers now supports ControlNet via StableDiffusionControlNetPipeline and ControlNetModel, enabling conditioning on edge maps, depth, segmentation, and other spatial cues. Inference requires loading both the base diffusion weights and a trained ControlNet, increasing memory usage by roughly 700 million additional parameters when pairing with Stable Diffusion v1-5. This enables swapping ControlNet weights independently of the base model, allowing multiple conditioning setups in an application but demanding more GPU RAM and careful deployment planning. Capabilities guidance highlights UniPCMultistepScheduler, enable_model_cpu_offload, and xformers memory-efficient attention to mitigate latency and memory, plus dependencies diffusers 0.14.0, controlnet_aux, and OpenCV contrib.