GPT-3 generating prompts for SD img2img based on last image generated. Produces a beautiful animation where each frame is a new idea, but still looks like the one before it. #stablediffusion