osgaa444/skills/seedance-prompt

seedance-prompt · The Director's Formula (v5.0)

This skill translates a user's creative vision into a production-ready prompt for Seedance 2.0. It uses a genre-aware, intent-driven workflow based on community-proven best practices.

The Governing Philosophy

Seedance 2.0 is an AI director. Tell it WHAT you want and HOW it should FEEL, not every micro-detail of HOW to execute it. Trust the model.

The Director's Formula

All prompts should follow this structure. The default target length is 30–100 words.

[Subject] + [Action] + [Scene] + [Camera] + [Style] + [Constraints]

---

The Workflow

1. Genre Router

First, ask the user to identify their content category. This determines the prompt's skeleton, length, and emphasis.

| Genre | Lead With | Length Target | Key Sections | | :--- | :--- | :--- | :--- | | Product/E-commerce | Subject | 30-50 words | Subject, Camera, Style | | Lifestyle/Social | Action | 40-60 words | Action, Scene, Style | | Drama/Narrative | Scene | 60-100 words | Scene, Action, Camera | | Music Video | Style | 50-80 words | Style, Action, Camera | | Landscape/Travel | Scene | 30-60 words | Scene, Camera, Style | | Commercial/Brand | Style | 40-70 words | Style, Scene, Action | | Anime/Artistic | Style | 50-90 words | Style, Action, Scene | | UGC (User-Generated) | Action | 20-40 words | Action, Scene, Camera (handheld) |

For detailed templates per genre, see [ref:genre-guides].

2. I2V vs. T2V Gate

Ask if the user has reference images. This fundamentally changes the prompt.

• If YES (I2V Mode): The prompt should only describe motion and camera movement. Do not re-describe static visual elements that the model can already see in the image.
• If NO (T2V Mode): The prompt must describe all visual elements from scratch.

For detailed I2V guidance, see [ref:i2v-guide].

3. Build with the Formula

Construct the prompt using the Subject + Action + Scene + Camera + Style + Constraints formula, tailored to the chosen genre.

4. Physics & Degree Adverbs

Instead of vague mood words, use concrete physics and intensity descriptors.

• Physics Language: tires smoke, gravel sprays, sweat flies off in slow motion, dust erupts violently, fat renders slowly, oil bubbles form.
• Degree Adverbs: slowly, dramatically, violently, gently, frantically. The model cannot infer intensity from images; it must be stated.

5. Anti-Slop Check (Expanded)

Scan the prompt for slop words that conceal a lack of specific information. This now includes action and camera slop.

• Visual Slop: beautiful, stunning, epic, amazing
• Action/Camera Slop: dynamic, energetic, cinematic camera movement, cool transition

Replace all slop with measurable, observable details.

---

Why This Works

• Aligns with Model Behavior: This workflow is based on extensive analysis of community-proven prompts that have the highest success rates.
• Reduces Failure Rate: By avoiding overspecification and using the model's preferred prompt structure, this method drastically reduces errors like jitter, morphing, and camera chaos.
• Genre-Specific: Provides tailored guidance for the most common real-world use cases, not just one expert-level workflow.

---

Maintained by Emily (@iamemily2050)