AI startup idea: use stable diffusion to generate images of code that are then OCR'd and executed. The code could literally do anything. The sky's the limit.

I see no potential technical hurdles.