[throw310822]

I wonder if at this point you could just ask the agent to iteratively refine the image in smaller portions.