Appearance
Moments
A moment is one meaningful beat inside a video — the product being revealed, the title card, the person talking to camera, the closing sign-off. A video is one thing you can point at; a moment is one of the several things that happen inside it. Moments exist because a creative team does not want "that Nike film" — they want the bit where the shoe turns in the light. Moments are how the library becomes searchable at the grain people actually think in.
Every moment belongs to exactly one video. There is no such thing as a free-floating moment.
How moments connect to everything else
Read it as sentences: a video is cut into moments; each moment sits on a shelf that says what the beat is doing; each moment also carries a few side descriptions; and the video itself belongs to a brand.
Where moments come from
Moments are found automatically. Software watches the video and detects where one shot ends and the next begins, and each of those becomes a candidate moment. A model then looks at the whole video and describes each beat. A person can correct any of it afterwards, and we record which of the two said what — the required behavior is that a human correction survives the next automatic pass. That protection still needs complete verification; describing the rule does not certify every current writer.
Underneath the moments sits a rawer layer: the mechanical cut list, before anything has been said about what a beat means.
The three levels
Moments are described at three levels of increasing detail. The top two are fixed lists; the bottom one is deliberately open.
Level one — the browse groups. These exist for navigation: the headings you would see down the side of a page. Nothing is ever labelled with a group; a group is worked out from the shelf a moment sits on.
Level two — the shelves. This is the real classification: what the beat is doing. Each shelf belongs to one group.
| Group | Shelves under it |
|---|---|
| Product | Product reveal · Feature highlight · Product demo · Product in use |
| Interfaces | Interface |
| Typography and titles | Title card · End card |
| Brand in motion | Logo animations |
| Design | Data visualisation |
| Editorial | Transitions |
| Effects | (none yet — the group exists, no shelf sits under it) |
| Shot types | Dialogue · Establishing |
Level three — the detail. Underneath each shelf sits a free level for the specific craft: a full-page title, a logo morph, a chat window, a chip reveal. This level is open on purpose, so a genuinely new technique can be recorded rather than forced into an existing word. Recurring terms are reviewed for promotion into a curated vocabulary with written definitions, and spelling variants are brought together. The reviewer sees a proposed promotion list. The vocabulary stays stable between planned versions; accepting new detail does not silently change the governed shelf list.
Two rules that govern shelves
A moment may sit on several shelves, and exactly one is its main shelf. A closing card that also contains an animated logo genuinely belongs to both, and forcing a single choice would lose real information. The main shelf is the one used to title the moment and sort it.
Browse visibility is separate from classification. Featured selection does not remove a matching moment from its chosen browsing path. The approved Shots entrance includes matching production shots without requiring a hidden catalogue switch. The library serves people looking for production shots, such as an establishing shot, and people looking for narrative or design beats, such as a product reveal or logo animation. Both are useful ways to find references.
The side descriptions
Alongside its shelves, a moment can describe what is visible, what happens, how the view is composed and what the beat does in the story. These questions may have overlapping answers.
| Description | What it answers | Illustrative boundary |
|---|---|---|
| Subject | What is literally shown? | A person or a product interface |
| Camera framing | How is the view composed? | A close-up can show a reaction |
| Activity or presentation | What happens in the scene? | A chat exchange, interview or demonstration |
| Story job | What does the beat do in the sequence? | Introduces a problem, offers proof or requests an action |
The 5 October decision separates these questions. The older Framing list mixes them and asks for one answer. That list is not the approved final vocabulary. Exact choices, compatible combinations and conversion of existing assignments remain to review. A new name alone will not repair the classification rule.
Chat Interaction describes an exchange of messages. Whether AI participates needs separate evidence. Data Analysis describes examining data for findings or comparisons; AI involvement is a separate fact. Both can coexist without becoming synonyms. The chat and analysis decisions preserve these boundaries.
An observed activity may demonstrate a product feature. Keep its product connection and supported name separate from reusable descriptions. A comparison interface may show prices for the products being compared; those are not automatically the price of the app demonstrating the feature.
Calls to action, prices and offers may occur together or separately, inside or outside an end card. Keep their wording, subject and source location. The same observations apply to images and pages through their own content relationships; those parts do not become video moments. See Content.
A real example, measured on 2026-10-03 18:41:42 UTC
OpenAI has the most recorded moments among companies linked as the primary company on videos that have not been deleted. Its 192 videos have 2,644 recorded moments. The most frequent recorded main shelves are: Dialogue (851), Product Demo (699), Title Card (216), Product in Use (191), UI (172).
These shelf counts describe main assignments only. They do not include additional shelves or establish that a classification is correct. A shelf's frequency does not tell us whether a shot is useful as a reference or how it should appear in browsing.
Finding the beat without mislabeling the film
A brief logo animation is a description of the moment, not of the entire film. The required browsing behavior is to open the matching interval and keep a link to its parent video. That behavior still needs end-to-end verification. The parent can therefore be discovered through the match without being presented as a whole logo-animation piece.
Filters use maintained choices with written definitions. Rich descriptions explain what makes the reference useful and can suggest unfamiliar terms for review. Writing a new phrase does not silently add a canonical filter. The browsing decision confirms this separation; the exact vocabulary still needs its own review.
How a product appears and what it does
How a product appears answers whether we see an interface, a physical object or another supported form of presentation. Its role answers whether it is the focus, actively used or incidental. Both facts can be true of the same moment and remain separate.
The useful-differences decision settles that separation. The exact choices still need definitions and examples. Keeping the questions separate does not certify existing classifications, and unexpected stored words need investigation before any conversion.
Hands-on Interaction describes deliberate physical interaction with a product using the hands: operating controls, adjusting it or actively handling it. A posed hand merely holding the product is insufficient. Keep the observed action separate from whether the product is being used for a task: use can happen without touch, and adjustment does not necessarily establish use. Hands-on Interaction can coexist with UI Interaction when both are supported. The definition decision does not merge older labels or change the whole Product in Use shelf.
Who contributed to a moment
A contribution can concern a particular moment, a whole piece or a grouping. Keep the contributor, their role, the scope of their work and supporting evidence connected. A studio may animate a sequence inside a video while another company produces the live-action scenes. Recognizing animation does not identify its maker, and a campaign credit does not automatically apply to every interval.
The contribution decision establishes this requirement. Exact role names, storage and the complete collection-to-browsing path still need verification. Preserve uncertainty and human corrections when intervals or source claims change.
Where to look next
- Every value you can actually choose — the full vocabulary: the subject, the shelves, the story job and the framing.
- How many moments we hold, and how they are spread — the state of the data.
- Where the live system disagrees with this page — the conformance report.
- What is still undecided about moments — the open questions.
- The exact tables and columns — the data reference.
Last reviewed 5 October 2026.