You are a photo print on a table of paper objects that arrive as you name them; one flips over to reveal the picture you describe. A vertical 9:16 creator kits motion graphics scene. Copy it into Claude or Claude Code, change the words and colours, and render the MP4 free.
Transcribe the clip with word-level timings and keep the words as data (every caption, keyword and cue below is timed from them).
Write the cut as data before any styling, in edit.json:
source and duration.
emphasis: the 15-20 words that carry the meaning (numbers, nouns, strong verbs). Keywords and highlights only ever come from this list.
shots: one entry per camera angle in the clip, {t, face: [x, y], pip}, where your face centre sits in source pixels, so every crop and picture-in-picture keeps your head…
beats: [{t, layout, zoom?, visual?, chapter?}], one per change on screen. Something changes every 2-5 s and never more than 6 s goes by without a change.
Keep the look in two separate files, kit.json (colours, fonts, caption size and height per layout, picture-in-picture and end-card rectangles) and kit.css. Restyling later means swapping those two files, never…
Render, then watch it through once at full size: nothing may cover your face, and every caption must land on its spoken word.
Hook at frame 0: an Anton headline ([e.g. "The last page of a notebook"]) with the orange twist slammed at 0.65 s ([e.g. "changed her career."]) and a punch on…
Objects arrive when named: items deal in as you count them, a title card lands on its word, a rubber stamp on the date.
The look
Table: a newsprint surface #E9E2D0 -> #DCD3BE with paper grain. Ink #1C1A17. One accent, [sunrise orange #E2672A], for stamps, keywords and your final line. The revealed illustration is the only…
Type: [Anton] uppercase for the hook, covers and stamps; [Courier Prime] bold for captions and dates; [Caveat] for handwritten print captions in blue-black ink; [Instrument Serif] italic for the quoted…
You are a photo print: a cream border with a deep bottom lip carrying a handwritten caption ([e.g. "Lisbon, June 2019"]), tilted 1.5-6 degrees, with a real paper shadow. The…
How the prompt opens
A table of printed objects: you are a photo print, and the things you mention are dealt onto the table as you name them. For stories about a moment, an object or a memory. Turn [your talking-head clip] (you speaking to camera, 20-90 s, any shape) into a finished vertical short: 1080x1920, 30 fps, built in HyperFrames and rendered with npx hyperframes render. Your footage and your voice stay exactly as recorded. The design lives around you, and it changes with what you say.
The full prompt runs 752 words. Copy it from this page.
Copy the prompt into Claude, or pull the scene into Claude Code and render the MP4 on your machine: how it works.