“A current limitation of image models like NanoBanana is their inability to generate factual and well-formed text within images for use cases like infographics or annotations.”