Comment by lcolucci

This is so obvious now that you say it (* facepalm *). We definitely need to give the LLM context on the appearance (both from the initial image as well as any /imagine updates during the call). Thanks for pointing it out!