Seedance 2.0 Prompt Expert

From Abstract Concepts to Executable Instructions

A common failure in AI video generation occurs when users provide vague prompts like "create a Martian scene, sci-fi feeling." The model struggles because it cannot "understand" abstract adjectives; it executes concrete, physics-based instructions. This skill bridges that gap, translating creative intent into precise "cinematography instructions" that the Seedance 2.0 model can reliably follow. It addresses the core problem of mismatched expectations between human language and AI capabilities.

Core Methodology and Workflow

The skill is built on the "Nüwa Methodology," a practical framework emphasizing structured prompting and physical constraints.

1. The Four-Layer Structure Method: It deconstructs a prompt into four mandatory layers, filled in priority order:
   • L1 Technical Foundation: Define non-negotiable parameters like aspect ratio 16:9, frame rate 24fps, ARRI Alexa camera texture.
   • L2 Spatial Setup: Define foreground, midground, and background positions, e.g., "Foreground: rocks taking up 20%; Midground: main subject; Background: architectural outlines."
   • L3 Lighting Design: Specify light sources precisely, e.g., "Key light from the left at 45 degrees, shadows on the right tinted blue."
   • L4 Motion Choreography: Describe camera and subject movement last, e.g., "Slow push-in, completed over 8 seconds."
2. Explicit Multi-Modal Binding with @ Syntax: The model does not infer the purpose of reference materials. The skill mandates explicit declarations:
   • Use @image1 as the protagonist to lock character consistency.
   • Use @video1's camera movement style to replicate motion patterns.
   • Use @audio1 at second X's beat to synchronize actions.
3. Golden Formula & Physics Validation: Before generation, it validates the prompt against the Subject + Action + Scene + Style + Emotion formula. It then filters for compliance with inherent physical laws (gravity, lighting consistency) and motion constraints (e.g., avoiding >2 simultaneous motions), significantly reducing failure rates from the outset.

Honest Boundaries and Best Practices

The skill transparently defines the model's "Honest Boundaries"—problems that cannot be solved by prompting alone.

• Capability Ceilings: Stable single-generation length is 60 seconds at 2K; complex fluid interactions (blood, lava) may clip; precise camera control is limited—using @ syntax for reference is recommended.
• Failure Stoppages: It provides clear diagnostic steps for issues like character inconsistency or chaotic motion, and a hard rule: "If it fails 2-3 times, abandon the approach." This prevents wasted effort on infeasible goals.
• Methodological Tensions: It acknowledges inherent conflicts, such as "more specificity is better" vs. "over-specification causes overload," or "cinematic feel" vs. "stability." It offers guiding principles, like being very specific for technical layers but using "fuzzy precision" for motion descriptions.