Updated
Case evidence: Read the source breakdown report · View case frames and shot timings
Based on a published tool report. Production classifications are analytical judgments; suggested steps are not the original project or a verified recreation.
Use timing to understand pacing
The unboxing report records a 1:11 vertical video with 69 shots (about 1.0 second each) and 68 cuts. Its first six shots last approximately 0.6, 0.6, 0.6, 0.5, 0.7 and 0.8 seconds: five quick product teasers in the first 3.1 seconds, then a flat-lay of every item. Those short edits are useful pacing evidence, but they do not tell you the duration of the original takes. The report's shooting plan suggests leaving about one extra second before and after each action so you can choose the cut later.
Separate the edit list from the shooting order
Create one sheet in story order and another grouped by camera setup. The report groups the reference into 3 scenes and 7 camera setups; most shots come from one carpet filmed from a phone directly overhead, and its edit note describes each item as tear the parcel → unwrap → reveal → show details. For your own product, a suggested sequence is unopened packaging, opening action, contents laid out, useful detail and a final overview. This is a planning example, not a claim about the exact order in the reference. Group overhead actions together, then record the closer views without repeatedly moving the camera back and forth.
Write observable actions into each row
Replace a vague instruction such as “show the product nicely” with a starting state, an action and an ending state. For example: the closed box is centered; two hands lift the lid; hold briefly with the contents visible. Add framing, where the hands enter and space reserved for labels. These are directions for a new shoot, not extracted original camera settings.
Plan text and sound as separate layers
The report lists 28 subtitle lines, 2 titles and 2 emoji stickers, and found no added sound effects; the background music runs at about 92 BPM and its source is unconfirmed. These are tool-output counts, not a target for your edit. In this reference each item gets one short line of text that stands in for a voice-over. Start with labels that clarify what the viewer is seeing. If you add effects, assign each to an event, such as a label appearing or an object being placed, and move it when you change the timing. More overlays do not automatically make the explanation clearer.
Review the classification before planning AI
The report reads 68 shots as live action and one, shot #68 (a 1.0-second flat-lay photo), as a still image with pan/zoom, with an overall reliability of 79%. That reading still needs review: platform compression can obscure visual clues, and a classification is not proof of how a creator worked. The still shot can be rebuilt from a single photo and the measured zoom and pan. If the actions can be filmed with your product, film them. The useful deliverable is a shot list, an asset list and an editing plan you can execute, whether or not AI generation is part of it.
A reusable analysis template
Shot / purpose / reference time Camera setup and framing Start state → action → end state Product and props needed Caption / sound event / edit duration Shoot order / takes captured / review notes
Practice with a real report
Compare the first six shot durations on the case page, then open the full report for the rest of the timeline and the overlay analysis. Keep report observations separate from the choices you make for your own shoot.
Open the case and download its report →Continue your workflow
- Video to prompt: translate observations into instructions
- How to describe a video with AI: beyond a summary
- Camera movement prompts: separate the camera from the action
- Video to storyboard: build a shot list from a reference
- How to reverse engineer a video, shot by shot
- Get started in Claude Code or Codex