Post pages were emitting only the author's avatar thumbnail as og:image with twitter:card=summary, so iMessage and other rich-link previewers rendered a tiny avatar next to the post text instead of the post's photo (the way bsky.app does). Now generateMetadata pulls the first image, video thumbnail, or external thumb out of the post embed, includes its aspect ratio + alt, and switches the Twitter card to summary_large_image when media is present. Text-only and quote-only posts still fall back to the avatar.