Seedance 2.5 Prompt Guide: 12 Prompts That Actually Work

Seedance 2.5 rewards a different prompting style than 2.0: think in beats with timestamps, name one camera move per beat, and describe sound as part of the shot. These 12 patterns cover the most common jobs. New to the model? Start with How to Use Seedance 2.5.

30-second storytelling (timestamp beats)

1. 0-8s: dawn wide shot of an empty surf beach, slow push-in toward the waves; 8-18s: a surfer paddles out, tracking shot at water level; 18-30s: she catches a wave, camera follows the ride. Ambient swell and gull cries throughout.

2. 0-6s: macro shot of flour hitting a wooden board; 6-16s: hands knead dough, gentle top light; 16-24s: the loaf enters the oven, warm glow; 24-30s: steam rises from the sliced bread. Kitchen sounds, no music.

Seconds-level beats are 2.5’s signature control — use them for any clip past 15 seconds. Keep each beat to one action plus one camera move.

Dialogue and performance

3. A fitness coach faces camera in a bright studio: "Three moves, ten minutes, no equipment." Energetic but natural delivery, room tone, subtle echo.

4. Night market in Taipei: a vendor calls out in Mandarin, then switches to English for a tourist; sizzling wok, crowd murmur, handheld follow shot.

2.5 handles 10+ languages with matched lip tempo — write the language switch directly into the prompt for localization work.

Product and brand

5. A matte-black wireless speaker on a concrete plinth, slow 180-degree orbit, soft studio key light, dust particles in the beam; a deep bass note lands as the orbit completes.

6. 0-10s: condensation slides down a chilled soda can, macro; 10-20s: the can opens with a sharp crack and fizz; 20-30s: pull back to a rooftop party at golden hour.

Camera grammar

7. Single continuous tracking shot: follow a cyclist through a narrow old-town alley, camera at handlebar height, morning light slicing between buildings.

8. Slow crane reveal: start on a single lantern, rise to expose a full night festival street, crowds in motion, layered chatter and distant drums.

Name the move (push-in, track, orbit, crane) and its speed. 2.5 keeps perspective stable through moves where 2.0 used to drift.

Reference-driven recipes

9. [3 character reference images + 1 location image] The referenced woman walks through the referenced train station at rush hour, medium tracking shot, announcements overhead.

10. [white-model blockout of a car assembly + 1 photo of a finished car] Render this assembly sequence photorealistically: robotic arms fit the doors, factory ambience, sparks at the weld points.

11. [10s reference video of a camera move + 2 product photos] Reproduce this exact camera path, replacing the subject with the referenced product on a marble counter.

12. [1 audio track: acoustic guitar loop] A street musician plays this exact tune on a rainy evening corner; match finger movement to the audio, neon reflections on wet pavement.

Label every reference’s role in the prompt. The model understands image, video, audio, white-model and green-screen inputs — mixing modalities in one job is where 2.5 pulls ahead. More in the multimodal reference guide.

What to avoid

• Multiple actions inside one beat — split them into separate timecodes.
• Abstract mood words without concrete visuals — anchor every adjective to something the camera can see.
• Text overlays in-frame — add titles in post, not in the prompt.