Turns LLM coding capabilities into image composition through a fine-tuned model and virtual canvas. Developers describe layouts in code, which the system then renders into generated images.