A single prompt is turning GPT Image 2 into a weapon against polished AI aesthetics. The instruction: "Redraw the attached image in the most clumsy, scribbly, and utterly pathetic way possible. Use a white background, and make it look like it was drawn in MS Paint with a mouse." The results look exactly as described - wobbly lines, off-register colors, something that wouldn't pass a first-grade art class - applied to whatever photo you feed in.
The appeal is obvious. Most AI image generation optimizes for impressive output: crisp lighting, perfect proportions, photorealistic detail. This prompt flips that entirely, asking ChatGPT's image model to deliberately produce bad art. The results are consistently funny in a way that polished AI images almost never are.
GPT Image 2 is OpenAI's image generation model built into ChatGPT, handling editing, style transfers, and generation from text descriptions. The model is notably precise at following visual instructions, which is exactly what makes this prompt land - it applies that same precision to the goal of producing something terrible.
There are practical uses here beyond the joke. Rough, deliberately imperfect sketches work well for wireframes, storyboards, or brand assets that need a hand-drawn feel. The bigger takeaway is that these models don't have a fixed aesthetic - the output is entirely shaped by how you ask.