The Seedance 2.5 prompt guide
Most weak results come from weak prompts, not the model. Here is the formula that gets sharp, controllable clips out of Seedance 2.5 — with examples you can copy.
The one-line formula
Subject → action → setting → camera → mood, written as one natural sentence. Seedance 2.5 is trained on described scenes, so a sentence beats a pile of tags every time. Start with the most important thing, say what it does, then place it and shoot it.
Example: "A lone hiker stands on a rocky ridge at golden hour, wind moving her jacket, slow dolly-in on an 85mm lens, shallow depth of field, 24fps cinematic."
1. Lead with the subject
Put the subject first and keep it concrete. "A red vintage motorcycle" gives the model an anchor; "something cool" does not. If identity matters — a specific person or product — use image to video and upload the real frame instead of describing it.
2. Name the motion
This is the single biggest fix for "my video doesn't move." Use strong motion verbs and say what should animate: drifting, surging, spinning, blinking, steam rising, fabric rippling. A still description produces a near-still clip.
3. Add camera language
Camera direction steers the look hard. Borrow film vocabulary:
- Movement: slow dolly-in, orbit, handheld, crane up, locked-off.
- Lens & depth: 85mm, wide angle, shallow depth of field, bokeh.
- Feel: 24fps cinematic, anamorphic, film grain, golden hour.
4. Set the mood and light
Light and tone do a lot of quiet work: "soft window light," "neon reflections on wet asphalt," "overcast and moody." One or two phrases is enough — do not stack ten.
5. Pick length and aspect ratio deliberately
On the Seedance 2.5 model, cost is flat across clip length — so if the motion suits it, a longer clip is better value per second. Choose aspect ratio up front: 9:16 for shorts and reels, 16:9 for landscape.
Copy-paste starting points
- Product: "A glass perfume bottle on wet stone, slow orbit, droplets catching light, soft studio key light, 24fps, shallow depth of field."
- Character: "A young woman in a rain-soaked city street, neon signs behind her, she turns to camera as rain falls, slow push-in, anamorphic, cinematic."
- Nature: "A humpback whale glides through deep blue water, shafts of sunlight from above, slow tracking shot, particles drifting, 24fps."
Troubleshooting
- Too static? Add motion verbs + a camera move.
- Face drifting? Use image-to-video with a sharp, well-lit source photo.
- Looks generic? Add specific light + lens language; cut vague adjectives.
- Wrong vibe? Adjust the light and lens language before rewriting the whole prompt.
Prompting questions
How long should a Seedance 2.5 prompt be?+
Aim for roughly 30–80 words: one clear subject, the action, the setting, and a short camera note. Too short is vague; a wall of comma-separated tags confuses the model.
Should I use keywords or full sentences?+
Full natural-language sentences. Seedance 2.5 reads a described scene far better than a list of tags. Write it like a shot description, not like a search query.
Why does my video barely move?+
Add explicit motion verbs and a camera move. "A portrait" sits still; "slow push-in as her hair drifts in the wind" tells the model what should actually animate.
Image-to-video or text-to-video?+
Use image-to-video when you have a photo to animate and want faces kept faithful; use text-to-video when you have no image and want a clip straight from a prompt.