The only video in this collection. Ten seconds, cut into five shots, and the prompt writes the shot list with timecodes.
It hit the numbers exactly. The file is 10.20 seconds at 24 fps — the prompt asked for 10 seconds and 24 fps, and those are the two things you would expect a text prompt to have no purchase on at all. It did not get 8K; it got 480×624, which is roughly 3:4.
All five shots landed, in order. Sampling a frame every two seconds gives you the shot list back: front selfie smiling at the lens; extreme close-up with a hand coming up to move the hair; three-quarter turn toward the bedside lamp, which is visibly in frame; an over-the-shoulder glance back toward the photo wall; and a final close-up with a wink. Five instructions, five blocks, no drift and nothing swapped.
That is the same result as the nine-panel collage, and it is becoming the clearest pattern in this whole collection. When a prompt breaks a request into an explicit numbered list — nine cells with nine colours, five time ranges with five actions — the mapping holds. When the same amount of detail is written as prose, individual instructions lose fights with their neighbours. Structure is doing something that wording cannot.
One thing worth catching: the poster on the wall is mirrored. The prompt asks only for photo prints behind her. The model added a framed print, and its text reads backwards. Front cameras mirror their preview, so a selfie video that comes out flipped is arguably correct — but it means every piece of text in the frame is reversed, which is a fast way to spot the flip if you ever need the footage the right way round. It also joins the running list of printed things nobody asked for: a seal, a route number, a date stamp, a garment badge, and now a poster.
