Large language models can't generate images. They can only write text or code. So I asked them to recreate famous paintings as SVG files, using nothing but their memory of what each artwork looks like.
No reference images. No examples. Just the model's training data and whatever it learned about art, composition, and color from text descriptions and code examples.
I tested this across 15+ models released between 2022 and 2025, including OpenAI's GPT series, Anthropic's Claude, Google's Gemini, and open source models from DeepSeek and others.