Originally published at openvideos.ai.
If you generate a 16:9 clip and crop it for TikTok / Reels / Shorts, the model composed for a wide stage. Heads get clipped and the payoff lands in the wrong third of a phone screen.
I write vertical shorts as a timed beat sheet instead: one subject, one location, and costume/camera changes on a clock.
The skeleton
[Title / vibe]
0s–Xs, [Shot size], [Look A + one action + light]
Xs–Ys, [Camera move], [Transition / look B]
Ys–Zs, [Pull back], [Finale pose + light accent]
Rules that matter in 9:16:
- One identity across beats (the tall frame magnifies face drift)
- Explicit shot sizes: medium → close-up → medium
- One camera verb per beat
- Name the second when the outfit/light changes
A verified 10s run
On OpenVideos I ran Seedance 2.0 (seedance-2-t2v) as text-to-video:
| Setting | Value |
|---|---|
| Duration | 10s |
| Aspect | 9:16 |
| Resolution | 480p |
| Audio | on |
| Credits (registry) | 520 (= 10 × 52 at 480p) |
Prompt (abridged structure — full text on the canonical post):
- 0–3s medium / eye level — plain ruqun, warm light, small hem lift
- 3–7s push to close-up — tear/splash into cheongsam, hard contrast
- 7–10s pull back — embroidered battle look, cold metal accents, pose that loops
That “look A → transition hit → finale pose” template also remaps to streetwear, beauty, or product unbox. Keep the clock; swap the world.
Compose in the prompt, not only the picker
Setting aspect_ratio: 9:16 is necessary but not enough. Prefer medium/close-up, put the face in the upper third, leave a quiet band for captions, and favor vertical motion over long lateral pans.
Full annotated prompt, failure table, and the playable result live on the original guide:
Top comments (0)