Prompt writing guide for Seedance 2.0

Plain-English Prompting: Get the Shot You Want with Seedance 2.0

Strong Seedance 2.0 prompts read like a shot brief, not a keyword pile. The model can work from text, image, video, and audio references, so the fastest path is to describe the shot in plain production language and add only the control signals that remove ambiguity.

The four-line prompt structure

Use one line for subject identity, one for action, one for scene context, and one for camera behavior. This keeps the model focused on visible facts. If you need a deeper template library, use the Seedance 2.0 prompt playbook after this guide.

Reusable template

[SUBJECT] Street singer with blue bob haircut, round glasses, denim jacket.
[ACTION] Walking forward while singing, right hand tapping the beat.
[SCENE] Night pedestrian street after rain, neon signs reflected in puddles.
[CAMERA] 50mm close-up, handheld, slight side-to-side sway, natural motion blur.

What to write, and what to leave out

Start with concrete nouns and verbs: person, product, location, direction, lens, camera path, lighting, and duration. Avoid vague quality words such as "amazing", "viral", or "high-end" unless you also explain what those words mean visually. A prompt like "cinematic product ad" is weak; "macro shot of a matte black smartwatch rotating on wet slate, thin rim light, slow clockwise dolly, shallow depth of field" gives Seedance 2.0 useful instructions.

Keep negative instructions short. Instead of listing ten things you do not want, rewrite the shot as a positive constraint: "single subject only", "fixed camera", "no text on screen", or "hands remain outside the frame." If a reference image matters more than the text, say so directly: "preserve the product shape and color from the image; only animate camera movement and reflections." For reference-heavy shots, read the omni reference workflow.

A fast iteration loop

  1. Render a short draft first. Use a 4-6 second clip before spending credits on longer versions.
  2. Fix one variable per retry. Change the subject, action, scene, or camera, but do not rewrite everything at once.
  3. Save the winning prompt beside the output. A small prompt log makes future batches easier to scale through the API workflow.
  4. Only increase resolution after motion and composition are correct. Upscaling a bad shot only makes the mistake clearer.

Example rewrite

Weak prompt: "Make a cool AI fashion video." Better prompt: "Young model in a silver raincoat turns slowly under a transparent umbrella, Tokyo crosswalk at night, soft reflections, 35mm waist-up tracking shot, camera moves backward at walking speed, no text, no logo." The second version gives Seedance 2.0 a visible subject, action, environment, lens, camera path, and boundaries.

Test one clean prompt

Start with the four-line format, render a short draft, then scale only when the shot direction is stable.

Start creating
Plain-English Prompting for Seedance 2.0