The project proposes a framework to ensure AI-generated visual media accurately follows user instructions. By representing both visual content and user prompts as programs, the system can automatically verify and iteratively refine the output. This approach aims to empower creators – such as journalists and designers – to produce precise, personalized visual content that aligns with specific needs and preferences.
Jiaju Ma, Yunzhi Zhang, and Vishnu Sarukkai, PhD Candidates in CS (Stanford); Advised by Professors Jiajun Wu and Kayvon Fatahalian (Stanford)