CoF-T2I: Video Models as Pure Visual Reasoners for Text-to-Image Generation | ScienceToStartup | ScienceToStartup