Generating an image for a lesson now passes along the text of the page it sits on, so the picture reflects what the page actually teaches rather than the search term alone. This applies whether the image comes from the AI agent building a lesson or from the regenerate button in the editor.
Art direction adapts to the image's role. A background behind lesson text is directed to stay quiet and low-contrast so the words stay readable, while a standalone image aims for a single clear subject. Question tiles are directed not to give away the answer, and every prompt now discourages rendered text, logos, and interface chrome inside the image.
Your organization's theme colors are part of the prompt too, so generated images sit closer to the palette of the lesson around them.