AI Short Video Prompts for 9:16: A Timed Beat Sheet That Works

Vertical shorts need a different prompt, not a cropped widescreen one

If you generate a 16:9 clip and crop it for TikTok, Reels, or YouTube Shorts, the model composed for a wide stage. Heads get clipped, hands leave the frame, and the “hero” action sits in the wrong third of a phone screen.

A stronger default for short-form is to write and generate natively at 9:16. On OpenVideos that means setting aspect_ratio to 9:16 in Create—not only mentioning “vertical” in the prompt text.

This guide is for creators who want one short clip with clear pacing: a hook, a payoff beat, and a pose that reads on a phone. It uses a verified Seedance 2.0 text-to-video run (Chinese fashion transformation, 10 seconds, 9:16) as a complete example. It is not a multi-model benchmark.

For general prompt anatomy (subject, action, camera, light), see How to Write AI Video Prompts. For the broader cinematic workflow, see How to Make Cinematic AI Video. Product entry for short-form generation: /ai-short-video-generator.

The timed beat sheet (for 6–15s clips)

Most weak short prompts try to narrate an entire music video in one paragraph. A timed beat sheet keeps one continuous subject while changing what the camera and costume do on a clock.

Use this skeleton:

[Title / vibe line]
0s–Xs, [Shot size + eye line], [Look A + one action + light]
Xs–Ys, [Camera move], [Transition / look B + contrast]
Ys–Zs, [Pull back or hold], [Finale pose + light accent]

Rules that keep the clip readable on a phone:

You can remix the aesthetic (streetwear, beauty, product unbox). Keep the clock + shot + change structure.

A verified 10s example (Seedance 2.0 · 9:16)

We ran the following on OpenVideos with Seedance 2.0 (seedance-2-t2v) as text-to-video.

Verified output (9:16, 10s):

Full prompt used:

Chinese Style Internet Celebrity Beauty 10s Transformation
0s-3s, Medium Shot Eye Level, A beauty wearing a plain-colored ruqun stands in front of a red wall, her wrist gently turning to lift the skirt hem slightly, warm light slanting to outline her silhouette, ancient poetic style
3s-7s, Camera quickly pushes to close-up, The instant the beauty turns and flings her sleeves, the plain skirt tears and splashes out red silk, in flashing strong light the cheongsam clings tightly to her figure revealing itself, strong light and shadow contrast, Chinese trendy mix-and-match style
7s-10s, Camera pulls back to medium shot, The beauty bends one knee and raises her hand to tousle her hair, the cheongsam transforms into a short embroidered battle outfit, waist tassels sway with the movement, cold light accents metal decorations, dashing Chinese style

That is the whole job: three beats, one courtyard, one woman, escalating costume and light. Credits for this model are calculated in the registry as duration × (52 if 480p else 70). Always confirm the live credit total in Create before you submit—pricing packs live on /pricing.

Beat-by-beat: what each line is doing

Editorial stills below illustrate the three looks. They are teaching frames for this guide, not frames pulled from the video file. Use the player above for motion and timing.

Beat 1 — 0s–3s: establish identity in a tall frame

Beat 2 — 3s–7s: the scroll-stop transition

This is the beat Western feeds reward: a clear before/after inside one clip, without needing a second take.

Beat 3 — 7s–10s: end on a pose that loops

Ending on a still-friendly pose also helps when you grab a thumbnail for the post.

Compose for 9:16 inside the prompt

Aspect ratio alone does not fix composition. Add spatial language the model can follow:

• Prefer medium or close-up over wide landscapes unless the subject is huge in frame.
• Put the face in the upper third; leave a quiet band at the bottom for captions and UI.
• Favor vertical motion (push-in, pull-back, hair/tassel sway) over long lateral pans.
• Keep backgrounds simple (a wall, a doorway, a single practical light)—busy streets fight the tall crop.
• Do not ask the model to render readable captions or logos; add text in CapCut / your editor.

If you are adapting a landscape idea, rewrite the shot sizes first, then set 9:16. Do not only append “vertical video” to an old widescreen prompt.

Failure patterns (and the smallest fix)

Remix the beat sheet without copying the whole aesthetic

The Chinese transformation is one viral pattern. The reusable template is:

1. Look A (calm, readable) for ~30% of the clip
2. Transition (camera push + costume/light hit) for ~40%
3. Look B / pose for the final ~30%

Swap in: coffee-run → night-out outfit; gym kit → stage look; product box → unboxed hero hold. Keep the red-wall courtyard only if you want that world—otherwise change the setting in all three beats together.

Limits

• This article documents one successful Seedance 2.0 job. It does not rank models or claim the same prompt behaves identically elsewhere.
• Wall-clock generation time was not logged for this write-up; do not treat any third-party “minutes” figure as ours.
• generate_audio: true was used on this job; treat audio quality as model-dependent, not guaranteed for every prompt.

When you are ready to try the verified prompt (or your remixed beat sheet), open Create in text-to-video, choose Seedance 2.0 if available, set 9:16, and confirm credits before generate.