engbergandreas/GauGAN ? reverse-engineered prompt

Build me a simple web app where I can draw semantic doodles on a canvas and have them turn into photorealistic images in real time.

I want it to feel like a GauGAN style demo, with a left panel for picking different material or object labels, a drawing area in the middle, and the image updating automatically as I sketch. It should let me move shapes around with a transform tool, and also give me a switch to turn live updates on and off if it gets too slow. Use the pretrained model files if they are there, and make the app work even if generation is a bit slow on CPU, but it should use GPU if available.

Keep the interface clean and easy to use, and if you need to check any current library docs while wiring up the drawing canvas or app framework, go ahead and look them up online.