Seedance 2.5 Prompt Guide: 10 Prompt Patterns for 30-Second Clips
Ten Seedance 2.5 prompt patterns with copy-ready prompts, settings and fixes: how to write subject, action, camera, light and sound for clips up to 30 seconds.
Seedance 2.5 reads a prompt as five fields in order: subject, action, camera, light and sound. Write them in that order and a clip of any length from 4 to 30 seconds holds together; bury the action under adjectives and the model drifts. A 5-second run at 480p is 56 credits, so testing one prompt three times costs less than a single 1080p take.
Updated October 2026.
How Seedance 2.5 reads a prompt: subject, action, camera, light, sound
The model responds to a sequence more than to a mood. Name the subject once, say what it does, say where the camera is, describe the light, and finish with the sound you want. Each field narrows the next, so the clip has less room to invent something you did not ask for.
- Subject: who or what, with one identifying detail. “A woman in a red linen dress” is enough; a full biography is not.
- Action: a verb with a start and an end. “Walks from the doorway to the counter and turns toward the window” gives the model a path.
- Camera: one dominant move, plus framing. “Slow dolly in, medium shot” reads as intentional; three moves read as a glitch.
- Light: direction and quality. “Overcast daylight from the left” is more useful than “beautiful light”.
- Sound: what the scene should sound like, because this model generates audio with the picture. “Room tone, distant traffic, no music” is a valid instruction.
That order also maps onto the tool. The text-to-video tab takes the prompt, then resolution, aspect ratio and duration; the credit cost updates before you generate. For the model’s specs and per-tier prices, the Seedance 2.5 guide covers them in full.
Ten shot types, each with a copy-ready prompt and the settings to use
These are prompts to try, not results from a specific session. All ten run at any of the six aspect ratios; the settings column gives a starting point.
| Shot type | Prompt to try | Settings |
|---|---|---|
| Product ad | A matte black water bottle rotates slowly on a pale stone plinth; single softbox from the upper right, faint shadow beneath, slow orbit right, medium close-up, quiet room tone. | 8s, 720p, 9:16 |
| Portrait | A man in a charcoal coat stands by a rain-streaked window and looks toward the street; cool overcast light from the window, shallow depth of field, slow push in, distant traffic. | 6s, 720p, 9:16 |
| Dance | A dancer in loose white clothing moves through a four-count turn in a bare studio; hard side light, wide shot, camera locked off, footsteps and cloth movement. | 10s, 720p, 16:9 |
| Action | A cyclist rounds a gravel corner at speed, sprays a little dust, and rides out of frame; low tracking shot from the side, late afternoon sun, wind and tyre noise. | 8s, 720p, 16:9 |
| Travel | An empty coastal road curves toward a headland at sunrise; camera drifts forward at drone height, long lens compression, warm low light, gulls and sea wash. | 12s, 1080p, 21:9 |
| Food | Hands tear a warm flatbread on a wooden board, steam rising; overhead shot, soft window light from the left, slow tilt down, close-up, crisp bread sound. | 6s, 720p, 1:1 |
| Anime | A girl in a yellow raincoat waits under a bus shelter as rain falls in thin lines; painted background, cool palette, static wide shot, rain and distant thunder. | 8s, 1080p, 16:9 |
| Interview / talking | A seated woman in a knit sweater answers a question, nodding once and glancing off camera; warm lamp light from the right, medium shot on a 50mm lens, quiet room. | 15s, 720p, 16:9 |
| Transformation | A plain white sneaker assembles itself from floating panels into a finished shoe on a dark surface; rim light, slow orbit, macro detail, soft mechanical clicks. | 8s, 1080p, 9:16 |
| Loop | Ocean waves wash over dark sand and pull back to the same line they started from; locked wide shot at dusk, long exposure feel, continuous surf. | 6s, 720p, 1:1 |
Writing a 30-second shot as three beats
A long take is not a long sentence. It is three beats connected by one continuous camera move. Each beat gets a subject change, a position change or a light change, and the camera instruction covers the whole run.
Example skeleton: “A woman opens a bakery door and steps inside (beat one); she crosses to the counter and lifts a loaf from the tray (beat two); she turns to the window, breaks the crust and looks out at the street (beat three). One slow dolly follows her from door to window, warm morning light, ambient shop sound and no music.” That is about 55 words and covers 30 seconds comfortably. Two beats feels rushed at that length; five tends to blur.
Keep the subject description to the first clause and refer to her as “she” afterwards. Re-describing the wardrobe mid-prompt is one of the most reliable ways to make the model change it halfway through.
Image-to-video prompts: what to leave out when the photo already shows it
When you start from a still, the image already carries the subject, wardrobe, framing and light. Repeating those details in the prompt gives the model two sources of truth and invites drift. Describe only the change over time: what moves, where the camera goes, and what sounds.
For a product photo: “Camera pushes in slowly while a soft highlight sweeps across the metal; the surface stays still; quiet room tone.” For a portrait: “She blinks, turns her head slightly toward the window, and smiles once; camera holds; soft wind outside.” The still is the first frame, so anything you describe as already present is redundant. If you want the shot to end somewhere specific, supply a last frame as well and describe the motion between the two.
Camera words that work and ones that get ignored
Concrete camera language changes output. Vague camera language tends to be skipped.
| Works | Ignored or ambiguous |
|---|---|
| slow dolly in, push in, pull back | cinematic movement |
| orbit left, orbit right, arc shot | dynamic angle |
| locked off, static wide shot | professional framing |
| handheld follow, tracking shot from the side | cool camera work |
| tilt up, tilt down, crane up | epic reveal |
| macro close-up, medium shot, wide shot | nice composition |
One move per clip is the rule that matters most. Our guide to camera movement prompts goes deeper, and the Seedance prompt library has more starting points.
Draft cheap, then render the keeper at full resolution
Seedance 2.5 costs 11.2 credits per second at 480p, 25.2 at 720p and 63.2 at 1080p. A 5-second draft is 56 credits at 480p and 316 at 1080p, so the cheap pass is where the prompt work belongs. Run the same prompt at 5 seconds and 480p until the subject, framing and motion are right, then extend the beat list and render once at delivery quality. Seedance 2 Fast is an even cheaper place to find the composition, at 24 credits for a 5-second 480p clip.
The exact price for your settings is shown before you generate, and failed steps are refunded. If you are buying credits, one-time packs never expire and there is no subscription needed; the current options are on the pricing page.
Common failures and the one-line fixes
- Subject changes mid-clip: describe them once and use a pronoun afterwards.
- Motion repeats or resets: give the action an end state, not just a start.
- Camera drifts: remove every move except one, and name the framing.
- Scene looks flat: add a light direction and a time of day instead of a quality adjective.
- Audio clashes with the scene: state the sound bed explicitly, including “no music” when you want none.
- Clip wanders at 20 seconds: add a third beat and a position change at the midpoint.
For a broader list of what goes wrong, see AI video prompt mistakes.
Frequently asked questions
How long can one Seedance 2.5 prompt run?
From 4 to 30 seconds in whole seconds. Past about 15 seconds, plan three beats and one continuous camera move.
Which resolutions and aspect ratios does it support?
480p, 720p and 1080p, at six ratios: 16:9, 4:3, 1:1, 3:4, 9:16 and 21:9.
How much does a test render cost?
A 5-second 480p run is 56 credits, so three tests cost 168 credits while one 5-second 1080p render is 316. The exact price is shown before every generation.
Should I write the sound in the prompt?
Yes when it matters. This model generates audio with the picture, and a short sound instruction, including “no music”, keeps the result closer to the scene.
Can I use an image as the first frame?
Yes. Then describe only the motion, camera and sound, because the photo already supplies the subject, wardrobe, framing and light.
Open text to video to run these prompts, or image to video to animate a still. New accounts start with 10 free credits and no card, enough to see how the model reads a short prompt before you spend on a full take. One-time credits never expire, and there is no subscription needed.
Ready to try it?
Create your first AI video in minutes — no credit card required.
Start Creating Free →