The prompt that produces the pier video has two shots in it, not one. Shot 1 recreates your uploaded photo as a still-to-life moment. Hard cut. Shot 2 puts the same person on a wooden pier over the sea at golden hour, medium close-up, singing softly into the lens while the camera pushes in slowly. Then you add the format constraints: 9:16, about 8 seconds, keep the face and outfit consistent.
What do I type to get the Raindance pier video?
Here is a complete prompt you can paste into an image-to-video model. Replace the bracketed parts.
Two-shot sequence, cinematic, photoreal.
SHOT 1 (0.0-2.0s): Recreate the uploaded photo exactly as the opening
frame — same person, same outfit, same framing, same background. Add only
subtle life: a slow blink, slight head movement, gentle breathing. Camera
locked off.
HARD CUT.
SHOT 2 (2.0-8.0s): The same person now stands on a weathered wooden pier
over calm open sea. Low golden-hour sun behind and slightly to the side,
warm backlight rimming the hair, glittering highlights on the water.
Medium close-up, chest-up framing, eye level. The person looks directly
into the lens and sings softly — gentle, sincere, small mouth movements,
a faint smile, eyes steady on camera. Slow, continuous push-in toward the
face across the whole shot. Light breeze in hair and clothing.
CONSISTENCY: identical face, hair, skin tone and outfit in both shots.
No change of identity. No extra people. Hands relaxed and out of frame.
FORMAT: 9:16 vertical, 8 seconds, 24fps, shallow depth of field, warm
filmic grade, film grain minimal, no text, no captions, no logos,
no on-screen lyrics.
NEGATIVE: no dancing, no fast cuts, no camera shake, no exaggerated
mouth movements, no distorted hands, no duplicate faces, no watermark.
Three parts do the heavy lifting:
- The hard cut. Without it, models tween between your photo and the pier, and you get a smeary morph instead of the snap everyone recognises.
- "Slow, continuous push-in." The trend's signature move. Leave the camera out of the prompt and most models drift or orbit.
- "Sings softly … small mouth movements." Ask for singing and you get a mouth flapping wildly. Ask for soft singing with small movements and it reads as sincere, which is what the bridge needs.
Why does my prompt need a negative list?
Because the trend's name contains the word "dance". Models see it and start choreographing. An explicit "no dancing" is the single highest-value line in the whole prompt. Second most useful: "no duplicate faces" — image-to-video models are prone to spawning a second head during a cut.
Worth adding if you keep hitting the same problem: lips too busy → no lip-sync exaggeration, mouth movements minimal; face drifting off-model → strict identity preservation, do not restyle the face; pier looking like a city dock → open sea horizon, no buildings, no boats, no crowd.
How do I change the scene?
Keep shot 1, the cut, the push-in and the format block identical. Swap only the scene sentence in shot 2.
Rain on the pier (blue hour):
SHOT 2: The same person on the same wooden pier during light rain at blue
hour. Cool blue-grey light, wet planks reflecting, fine rain visible
against a dark sea, hair and shoulders damp. Medium close-up, eye level,
singing softly into the lens, slow push-in. Melancholy, intimate mood.
Sunset beach:
SHOT 2: The same person standing on wet sand at the water's edge at
sunset, shallow waves behind, warm orange sky, long soft shadows.
Medium close-up, eye level, singing softly into the lens, slow push-in.
Reception lights (evening):
SHOT 2: The same person at an outdoor evening reception under warm
string lights, soft bokeh lanterns behind, dark blue sky above.
Medium close-up, eye level, singing softly into the lens, slow push-in.
These are the scenes the trend settled on, mirroring the real music video — which opens on a pier at sunset and later moves to a formal dinner and a night beach party under string lights.
What is the duet prompt?
For two people, the changes are framing and separation. Models fail duets by merging two faces, so rule it out explicitly.
SHOT 1: Recreate the two uploaded photos as a side-by-side opening
frame — person A on the left, person B on the right, each matching
their own source photo.
HARD CUT.
SHOT 2: Both people stand side by side on a weathered wooden pier over
open sea at golden hour, shoulders almost touching, facing camera.
Two-shot, waist-up, eye level. Both look into the lens and sing softly
together — gentle, sincere, small mouth movements. Slow push-in on the
pair. Warm backlight from the low sun, glittering water behind.
CONSISTENCY: person A keeps the face, hair and outfit of the first
source photo; person B keeps the face, hair and outfit of the second.
Do not blend the two faces. Do not swap their positions. Exactly two
people in frame.
Add no third person, no duplicated limbs to the negative list. More at duet mode and the couple duet guide. For a dog-and-cat version, use the same structure with Pets mode plus two animals, one dog and one cat, normal animal anatomy, no anthropomorphism.
Does the prompt change per model?
Somewhat. These are general tendencies, not guarantees — model families ship new versions and change behaviour, so adjust from what you actually get back.
| Model family | What tends to help |
|---|---|
| Seedance | Responds well to explicit shot-by-shot structure with timings, as written above. Keep the identity-preservation line. |
| Kling | Tends to prefer shorter, denser prompts. Try one paragraph per shot and lean on the negative list. |
| Hailuo | Often good with camera language — "slow push-in", "locked off" — so keep those words rather than describing the movement in prose. |
| Veo | Generally comfortable with cinematic, natural-language direction. A shot list with lens and light notes usually reads better than keyword stacking. |
Two habits help on any of them: generate the same prompt two or three times rather than rewriting after one bad result, and change one line at a time so you know what fixed it.
Do I need to write a prompt at all?
No — not if you use our generator. The prompt above is essentially what we apply for you. Upload one photo (or two for a duet), pick a scene, and we build the two-shot structure, the hard cut, the push-in, the identity constraints and the negative list, then render on Seedance models. Output is MP4, 9:16 by default, 8 seconds at 720p, up to 15 seconds and 1080p on higher models.
The honest limits are the same as for any prompt you write yourself: the clip is silent — no song, no AI vocals, because the model cannot hear music, so the mouth movement is a soft singing performance rather than true lip-sync. You add the real track in TikTok, which is also what keeps the audio licensed. See adding the sound. One free 5-second 480p watermarked teaser per account (details), then clips from $4.99 — pricing.
Writing your own prompt is worth it if you want a scene we do not offer, or already pay for a model subscription. If you just want the pier clip, skipping the prompt is faster: the full walkthrough takes about five minutes, and your photo matters far more than your wording — see best photo.