A polished video can make a factual or rights mistake harder to notice. Separate approval of the message from generation of the visuals: agree the purpose, script, permissions and disclosure plan first, then produce bounded scenes and review the final export in the same context as the viewer.

1. Write the production brief

Define the audience, channel, duration, aspect ratio, call to action and one message the viewer should retain. List mandatory facts, prohibited claims, brand assets and accessibility requirements.

Classify the format: illustrative animation, realistic synthetic scene, avatar presentation, dubbed recording or AI-assisted edit. This classification affects consent and disclosure.

Record the origin and permitted use of scripts, images, music, footage, voices and likenesses. Obtain explicit permission before cloning a voice or creating an avatar of an identifiable person. Keep the consent record connected to the project.

Check the generator’s current commercial-use terms and the destination platform’s policies. A tool allowing an upload does not establish that you own the necessary rights.

3. Approve the script before generation

Fact-check names, numbers, quotations and product claims. Read the script aloud to test timing and pronunciation. Mark words that need on-screen evidence, a citation or a careful visual treatment.

Locking the message before generating scenes reduces wasted credits and prevents a visually polished mistake from surviving review.

4. Build a shot list

For each shot, specify purpose, duration, framing, subject, action, style and transition. Mark whether the image is illustrative or intended to represent a real person, place or event.

Generate difficult identity or product shots early. If consistency fails, change the concept before producing the whole sequence.

5. Generate in controlled batches

Keep prompts, reference assets, settings and selected outputs. Use approved brand colors and avoid adding logos, products or people whose presence could imply an endorsement.

Choose shots based on narrative function rather than novelty. A restrained visual that communicates accurately is more useful than a spectacular but misleading scene.

6. Edit outside the generation loop

Assemble the chosen footage on a timeline. Correct pacing, remove visual discontinuities, mix audio and add captions. Preserve enough separation between voice, music and visuals to make later corrections possible.

Do not rely only on automatic captions. Review spelling, timing, speaker changes and meaningful sounds.

7. Perform three reviews

Content review

Does the final video match the approved script? Are product interfaces, figures and quotations accurate?

Visual and audio review

Look for identity drift, malformed objects, unreadable text, implausible motion, lip-sync errors, abrupt audio and flashing content. Watch once with sound off and once without looking at the screen.

Trust and publication review

Could a reasonable viewer mistake a generated person, place or event for reality? Has consent been recorded? Does the destination require an altered-content or paid-promotion declaration? Are captions and disclosures visible in the final context?

8. Archive the production record

Keep the approved script, sources, asset permissions, generation record, reviewer, disclosure decision and final export. Where supported, preserve Content Credentials. C2PA explains that these credentials record provenance assertions but do not independently establish that the depicted claim is true.

The final approval is about the whole video

Do not publish until one named person watches the finished export and confirms:

  • facts match the approved evidence;
  • every identifiable voice and likeness is authorized;
  • the visuals do not imply a false event or endorsement;
  • captions and audio are usable;
  • required disclosures are enabled;
  • the final file was reviewed on the destination platform.

Sources