Independent visual research / Public editionReviewed 05 Sep 2026
Motion fieldnotesOpenAI film study
About this study

One composed original

Fifty seconds.
One continuous subject.

A small exhibition carries the same request and geometric motif through eight editing beats.

The edit list

This sequence combines eight studies around an original exhibition request and one geometric motif. It is not an edit list recovered from an OpenAI film.

Output interval Composition Continuity and hold
0–5 s TitleCut Complete premise before the 2.8 s cut; 2.2 s formed-surface hold
5–10 s CameraPush Same original artwork; brief push, action label, 2.8 s quiet target
10–17 s FocusType Cut to input; target before text; complete request held for 2.8 s
17–24 s SplitResult Same request; blank pane before content; 3.2 s completed-result hold
24–30 s ChapterReturn Same task interrupted by comparison title; return at 28 s; 2 s hold
30–36 s MotifMontage Three treatments of one silhouette; direct cuts, then a longer final view
36–42 s ObjectContext Same request and artwork move into device context by a cut, not a morph
42–50 s QuietEnding Return to the established proof; cut to ending at 43 s; original mark settles at 45 s and holds for 5 s

The broader sequence deliberately uses a result-first opening before explaining the request. A change from phone back to desktop proof at 42 s is an editorial return, not an animated device transformation or continuous product capture.

Comparison with the studied sources

Dimension Source observation Original sequence / discrepancy
Title-to-proof cut Atlas, 16.5–20.5 s: direct title-to-browser cut at 19.900 s; browser then continues enlarging Direct cut at output 2.8 s; browser-equivalent window remains stationary. Original phrase and longer chosen title hold
Camera Codex, 3.4–5.8 s: intermediate magnifications and blur before a selector menu Shared-layer camera, 0.5 s chosen move; no motion blur. The full-window heading leaves the shot so the selected artifact and action can be read
Request order Codex, 6.5–10.5 s: input is established before request growth Aim → type → complete request hold. Shorter original sentence; no submit or latency claim
Result order Prototype film, 51–56 s: narrow conversation, empty preview, then running calendar Same container-before-content order. Static original geometry instead of a running prototype; source camera continuation omitted
Chapter continuity Workspace risk film, 68–76 s: setup, chapter, same task context Exact original snapshot before/after the card. Source context continuity does not establish pixel-identical backend state
Montage Chameleon film, 49–50 s: recurring subject, several roughly 0.1 s portraits Three original variants with 0.8 s cadence, then a long last hold. Fixed registration is our choice; source portraits vary crop and size
Container handoff Gift film, 7.5–11.5 s: cut to phone at 8.133333 s, request returns afterward Same cut/brief empty-device order. Added original artwork and slower four-frame gap; no claimed geometric morph
Ending Codex, 127.85–132.61 s: identity motion precedes quiet final samples Original bars, not branded letter deletion or knot animation. Five-second motionless tail is a chosen research value, not a real talking-head buffer

Adapt and verify

The complete handoff documents every input, the pinned project, preview/render commands and a 52-second adaptation. extraHold extends only the final tail. Individual scene durations and phase frames are asserted before rendering. Use the source audit for date and coverage limits.

These durations do not claim pixel-identical reproduction or measured viewer comprehension. Text has been kept short, and the quiet result states are intentional reading opportunities. Final presentation pacing must be rebuilt against the real recorded speech and reviewed with the actual playback setup.

Reference frame

Enlarged reference frame

Watch this moment on the official channel ↗