How to Make AI Videos for Reels and TikTok
Build a clear vertical story without promising reach or retention.
Step by step
Write a one-sentence concept
State the subject, change and payoff. If the sentence contains several scenes, split it into separate shots.
Compose for a vertical frame
Select a portrait option only if currently offered. Keep critical faces, products and text away from interface overlays and crop edges.
Prompt one clear shot
Specify subject action and camera motion. Avoid asking the model to create captions; add exact text during editing.
Verify current duration and cost
Use the current signed-in controls. This guide does not repeat previous unsupported duration ranges or free-video claims.
Edit, disclose and publish
Add accurate captions and properly licensed audio. Disclose materially synthetic media where platform rules or viewer context require it.
Pro tips for better results
- Preview with interface-safe overlays.
- Never promise that a hook will improve retention without an actual experiment.
- Keep exact product text out of generation and add it in an editor.
- Review platform policies immediately before publishing.
Vertical safe-area review
Preview the clip with typical top, bottom and side interface overlays. Keep essential content near the visual center and add titles in an editor where placement remains controllable.
Measure honestly
Define one metric before posting, compare similar audiences and retain losing variants. A single post cannot isolate whether generation method caused an outcome.
Original vertical crop protocol (not executed)
Test input: Portrait composition of a ceramic cup centered on a small table; steam rises while the camera makes a gentle push in; clean background; no text.
Settings: Nearest portrait ratio, shortest current standard duration, no generated captions.
- Subject survives UI-safe crop
- Cup geometry
- Steam continuity
- Space for editor-added caption
Status: This separate protocol was not executed. The visible executed sample below uses a different disclosed input and must not be treated as its result.
Landscape sample exposes crop risk
Actual application handler, isolated environment, real provider; not the production browser/UI.
Prompt: A gentle slow camera push toward the orange mug. Keep the mug shape and handle stable. No new objects.
Recorded settings: 1280×720 landscape output; no vertical or social-platform publishing test.
Observed result and limitations: The teal circle leaves the frame during the push-in, and the opening frames have black side areas.
Why it matters here: This is evidence to preview crop-safe areas, not evidence of 9:16 support, retention, reach or social performance.
This is one of the same three image edits and one video reused across these 12 pages. It is a small illustrative sample and cannot be generalized. Read the evidence manifest.