XingangPan/DragGAN — reverse-engineered prompt

Build me an interactive AI image editing demo like DragGAN. I want to generate an image from a pretrained StyleGAN model, then click points on the image and drag them to new positions so the face, animal, car, landscape, or other generated subject changes naturally instead of just smearing pixels.

Please set it up as a simple Python app with a visual interface, preferably Gradio, where I can load the pretrained weights, pick or generate an image, add handle points and target points, run the drag edit, and save the result. Include the model download script and make it work with CUDA when available, but don’t break completely on CPU or Apple Silicon if possible.

Also include clear run instructions for local use and Docker, since this will probably need specific dependencies. Keep the watermark or “AI Generated” notice because the project requires that. Look up current PyTorch and Gradio docs online if needed.