Turn the script into a visual rule set
Before generating scenes, define the visual world in concrete terms: subject, location, period, palette, light direction, texture, framing, and what must remain consistent. Add exclusions for details that would misrepresent the story. A history video may need period-correct objects; a product explanation may require exact shape and labels; a news video may need clearly illustrative rather than documentary-looking images.
Use one or two reference frames to evaluate continuity, not as evidence that every scene must be identical. The visual system should support the story’s changes in place, scale, and emotion.
Use motion only when it has a job
A restrained pan can reveal a relationship, a push can direct attention, and a crop can adapt a wide composition to vertical framing. Movement becomes noise when it competes with captions, creates false physical action, or repeats on every scene regardless of meaning.
Inspect edges, hands, text, reflections, architecture, and recurring characters at full resolution. A smooth transition cannot repair a fabricated label or inconsistent identity. Replace misleading imagery even when it is visually impressive.
- Reveal: motion introduces a previously hidden detail.
- Direct: motion guides attention to the narrated subject.
- Connect: a transition explains a change in time, scale, or place.
- Hold: a static frame provides time to read or compare.
Grade the sequence, not isolated frames
Review the full video for abrupt changes in palette, contrast, camera height, subject scale, and caption placement. Create a shot list with the purpose and acceptance criteria for each scene, then judge whether the sequence feels intentional at normal playback speed.
Export and inspect on a phone. Fine details that look cinematic on a desktop can disappear in a vertical feed, while high-frequency motion can make captions harder to read.
VidiPrompt workflow example: a documentary-style lighthouse Short
Visual direction: cold coastal dawn, restrained blue-gray palette, consistent stone lighthouse, and diagrams only where they clarify the lens mechanism.
- 1Build a scene list that separates setting, mechanism, historical context, and conclusion.
- 2Generate the draft in VidiPrompt with concrete period and composition language.
- 3Reject scenes with inconsistent lighthouse geometry, unreadable labels, or false historical details.
- 4Use motion for reveals and scale changes; hold diagrams long enough to understand.
- 5Review the complete sequence on desktop and phone before export approval.
Review questions
- Can a viewer identify the same principal object across scenes?
- Does each movement clarify rather than decorate?
- Are illustrative or reconstructed scenes represented honestly?
Pre-publish checklist
- Concrete visual rules and factual exclusions
- Scene purpose and acceptance criteria
- Full-resolution artifact and text review
- Motion justified by meaning
- Sequence-level phone playback check
Common failure modes
- Style adjectives replace concrete subject and continuity requirements.
- Every scene moves even when the viewer needs time to read.
- A polished synthetic image is mistaken for documentary evidence.
- Individual frames look strong but conflict when played together.
Primary sources
- Disclosing use of altered or synthetic content
YouTube Help — Current YouTube disclosure requirements for realistic altered or synthetic media.
- AI-generated content
TikTok Support — Current TikTok guidance for labeling and handling AI-generated or significantly edited content.