Review the edit as a chain of promises
Continuity is not only whether a sleeve changes color. Each shot makes a promise about place, time, body, object, physics and attention. The next shot either carries that promise forward or deliberately breaks it. A contact sheet can help select individual clips, but it cannot replace an editor asking what the previous cut made the viewer expect.
Start with a silent, full-speed screening. Mark every moment where a reviewer involuntarily reorients: where is this? whose hand is that? did the product change? why did the light jump? Then screen again with sound, because music can hide a break during review and leave a message-level break in a silent autoplay placement.
Use five passes, each with one job
Pass one is screen direction and geography: subject movement, eyeline, entrances, exits and the source of light. Pass two is identity: face, hair, wardrobe, product, packaging, prop and markings. Pass three is physical behavior: contact, weight, liquid, shadow, reflection and cause-and-effect. Pass four is brand behavior: color, typography, product orientation, legal copy and pack visibility. Pass five is editorial behavior: pace, emotional escalation, sound bridge and the exact moment the viewer understands the offer.
Give a distinct reviewer the chance to stop each pass. The director may own performance continuity; the product lead owns package accuracy; the editor owns edit logic; the producer owns whether a fix can be acquired. One person can hold multiple roles on a small job, but the check should not vanish because its specialist was not at the first review.
Classify the fix before you ask for another generation
A continuity flag needs a type and an owner. ‘Continuity feels off’ is a note that creates more options than it resolves. Try: C1, product label swaps between shots, fix with locked plate; C2, character motion changes direction, fix with alternate take or crop; C3, emotional beat lands before the offer, fix in edit. This lets production compare an edit solution, practical pickup and regeneration instead of assuming the model must repair every issue.
Our recommended rule is that a generated shot is provisional until it has played in context at target duration and aspect ratio. Do not approve it at a paused full-resolution frame only. A good still can create a bad cut when motion, compression or a social crop removes the cue that connected it to the previous shot.
Leave an edit test record
For every locked sequence, retain the review export, timecoded notes, shot version IDs and the final decision for each critical flag. This is the least glamorous production artifact and one of the most valuable on versioned campaigns. When a regional cut, platform crop or late legal change reopens the film, the editor can see why the original transition was selected.
The next job benefits too. A small team can categorize recurring break types—hands, product contact, wide-to-close geography, lip sync, pack labels—and use that local pattern to plan better plates or simpler edits. It is a production learning loop, not a claim that a tool is reliable everywhere.
Sources & evidence limits
Source-backed facts are distinguished from the editorial workflow proposed here. Brand and agency accounts document their own work, not independent proof of performance. Read the linked source for its scope.
- How Lipton and Critical Mass crushed advertising targets with generative AI
Adobe states that generated visual references preceded a campaign with real models. Page displays 5 May 2026, while embedded card metadata says 2 June 2025; the displayed date is recorded here, not a verified campaign launch date.
Checked 2026-09-19 · Source published 2026-05-05