Seedance 2.0 Video Prompt Writing Guide

Problem: Prompts Are Too Vague

Seedance 2.0 accepts images, videos, audio, and text, but unstable outputs are often caused by missing engineering constraints: unclear asset roles, insufficient camera and pacing breakdown, and weak audio direction. This guide turns video generation into a checkable prompt structure: asset roles, timeline, camera language, style, and audio constraints.

How It Works

• @ reference system: assign a role to each asset, such as @image1 as first frame or reference @video1 for camera work.
• Time-based structure: for videos over 8 seconds, describe subject, action, and camera by time segment instead of stacking scenes.
• Camera and framing: use terms like slow push, orbit, dolly zoom, low angle, and close-up to make shot intent explicit.
• Audio sync: add narration tone, footsteps, and transition beats rather than describing visuals only.

Boundaries and Notes

It fits ad clips, short drama, science explainers, product showcases, and beat-synced edits. Respect limits: no more than 12 files, per-file size and duration constraints, and no realistic human face assets. Avoid contradictory instructions, such as requiring a fixed camera and an orbit shot in the same segment.