LLMs write rendering code. Existing engines produce the image. 100x cheaper than diffusion.Alexey Vidanov · July 2026