lllyasviel/ControlNet — reverse-engineered prompt

Build me a local ControlNet demo for Stable Diffusion that I can run from my computer in the browser. I want to upload an image or draw a simple scribble, type a prompt like “oil painting of a robot” or “dog in a room”, then generate a new image that follows the original edges, pose, depth, lines, segmentation, or sketch.

Please make the setup friendly, with clear instructions for creating the environment, where to put the Stable Diffusion model files, and where to put the detector weights from Hugging Face. Include Gradio screens for the main ControlNet modes shown in the project, like Canny edges, HED boundaries, user scribbles, fake scribbles, human pose, semantic segmentation, depth, normal maps, and straight lines.

Add sensible defaults, examples using the included test images, image preview, output download, seed and prompt controls, and a low VRAM option for smaller GPUs. If any model files are missing, show a helpful message instead of crashing. Look up current docs online if you need to.