Turn an image the moderator shared into text the characters can read.

Group chat delivers documents as text, and a character never sees a picture. The browser posts the bytes here, a vision-capable model from llm_metadata describes them, and the description travels on the same attachment channel as extracted document text.