Reading the page.
The wait. After the cook picks a photo, AI vision does the slow work — reading cursive on grandma's pink paper, transcribing the recipe verbatim. Centered on cream: a small typographic 'page' — five Ochre hairlines at varying widths that draw left-to-right and fade in sequence (each animation-delay offset so the wave never empties), suggesting the AI reading line by line down the page. Italic Lora '— reading the page —' beneath, in the system's own voice. No spinner, no progress bar, no percentage — those are SaaS chrome. No fake-photo graphic either — a paper-warmth placeholder where the cook's actual photo isn't would read as cosplay. Honors prefers-reduced-motion (lines settle into their drawn state, no animation). No tap target — the screen auto-advances to the transcription gate when vision completes. Cancel is the only out.