The Product Video With a Moving Logo
A small outdoor-gear brand wants a 15-second social video showing a jacket in use on a mountain trail. Budget rules out a shoot. The team generates a base clip with a text-to-video model: a hiker walking a ridgeline at golden hour. Problems: (1) the jacket's chest logo morphs and flickers across frames; (2) the hiker's gait has a subtle foot-slide; (3) the brand needs the real logo composited on the jacket. The clip will run as a paid ad, so the brand's legal team requires it to be clearly labeled as AI-generated where platform rules demand it. The team has 2 days and one editor.
1. Rank the three problems by how much they threaten the ad's usability and propose a fix plan for each within the 2-day budget.
Read the worked answer
Rank by risk to the deliverable, not by how annoying each looks:
- Morphing logo - highest risk. A flickering fake logo is a brand-integrity and legal problem; it also signals 'AI slop' to viewers. Fix: do not let the model invent the logo. Generate or select a clip where the chest area is plain, then composite the real logo in post with motion tracking. If the logo region is already corrupted, regenerate that segment with 'plain unbranded jacket, no logos' in the prompt.
- Foot-slide - medium risk. It reads as uncanny but is often fixable by trimming to a segment where the gait is cleanest, slowing slightly, or cropping below the knees for part of the cut. Regenerating the full clip is the fallback if no clean segment exists.
- AI labeling - a compliance gate, not an aesthetic issue. Determine platform disclosure requirements up front and build the label into the export. It is cheap to do and expensive to miss.
Sequencing: lock the logo-composite approach first (it constrains what footage is usable), then pick the cleanest gait segment, then finalize labeling. Two days is enough only if the team stops trying to get a perfect generation and accepts post-production compositing.
2. The editor suggests generating a second clip of the same hiker 'to get more coverage' and cutting the two together. What continuity and disclosure issues arise, and what should the editor do?
Read the worked answer
Cutting two independently generated clips together creates continuity failures that viewers notice even when they can't name them: different face, different jacket color/shade, different lighting direction, different gait rhythm. For a 15-second ad, a mid-cut identity change reads as a mistake and undermines the product.
Disclosure issues: each generated clip carries its own provenance. If the team composites or trims, the credentials must still reflect that the footage is synthetic - and the platform label must cover the whole ad, not one segment.
What to do:
- Prefer one continuous clip trimmed to the strongest 15 seconds over two stitched generations.
- If a second clip is unavoidable, match on non-identifying shots (boots on rock, hands adjusting a cuff, wide silhouette) rather than faces, so continuity can't fail on identity.
- Keep provenance attached through the edit and confirm the final export carries it.
- Verify the label appears wherever platform rules require, for the full runtime.
The judgment call: generated coverage is not free - every additional clip multiplies continuity and provenance risk.