Seedance Prompt Designer Skill
This skill transforms a user's scattered multimodal assets (images, videos, audio) and ambiguous creative intent into a structured, executable prompt for the Seedance 2.0 video generation model. It acts as an expert prompt engineer, ensuring the highest quality output from the underlying model.
Core Workflow
This skill follows a strict three-phase workflow. The output of each phase is the input for the next.
| Phase | Goal | Key Actions | Output |
|---|
| 1. Recognition | Understand user input | Parse intent, analyze assets, tag with atomic roles | INLINECODE0 |
| 2. Mapping & Strategy |
Design an executable reference strategy | Determine optimal reference method (Text, Asset, Hybrid) |
strategy_output.json |
|
3. Construction & Assembly | Generate the final, complete prompt | Assemble text, append @-syntax, consult templates |
final_prompt.json |
Usage Example
User Request: "Make the Mona Lisa drink a Coke. I want it to feel cinematic, like a close-up shot."
User uploads monalisa.png and coke.png
Agent's Internal Process:
- 1. Recognition: Identifies
action_intent ("drink a Coke"), style_intent ("cinematic, close-up"), and tags monalisa.png as Subject Identity and coke.png as Subject Identity-Object. - Mapping & Strategy: Decides to use
@monalisa and @coke as asset references and the rest as text prompts. - Construction & Assembly: Assembles the final prompt.
Final Output:
CODEBLOCK0
Knowledge Base
This skill relies on an internal knowledge base to make informed decisions. The agent MUST consult these files during execution.
- -
/references/atomic_element_mapping.md: Core Knowledge. Contains the "Asset Type -> Atomic Element" and "Atomic Element -> Optimal Reference Method" mapping tables. Must be consulted during Phase 1 and Phase 2. /references/seedance_syntax_guide.md: Seedance 2.0 "@asset_name" syntax reference. Must be consulted during Phase 3 to ensure correct syntax generation./references/prompt_templates.md: Advanced prompt templates for different genres (e.g., Cinematic, Product Showcase, Narrative). Optional consultation during Phase 3 for stylistic enhancement.
Seedance 提示词设计师技能
本技能将用户零散的多模态资产（图像、视频、音频）和模糊的创作意图转化为结构化的、可执行的提示词，用于Seedance 2.0视频生成模型。它扮演着专业提示词工程师的角色，确保底层模型输出最高质量的结果。
核心工作流程
本技能遵循严格的三阶段工作流程。每个阶段的输出即为下一阶段的输入。
| 阶段 | 目标 | 关键操作 | 输出 |
|---|
| 1. 识别 | 理解用户输入 | 解析意图、分析资产、用原子角色进行标记 | recognitionoutput.json |
| 2. 映射与策略 |
设计可执行的参考策略 | 确定最佳参考方法（文本、资产、混合） | strategyoutput.json |
|
3. 构建与组装 | 生成最终、完整的提示词 | 组装文本、附加@语法、参考模板 | final_prompt.json |
使用示例
用户请求：让蒙娜丽莎喝可乐。我想要电影感，像特写镜头一样。
用户上传了monalisa.png和coke.png
智能体的内部处理流程：
- 1. 识别：识别出actionintent（喝可乐）、styleintent（电影感、特写），并将monalisa.png标记为主体身份，将coke.png标记为主体身份-对象。
- 映射与策略：决定使用@monalisa和@coke作为资产参考，其余部分作为文本提示词。
- 构建与组装：组装最终提示词。
最终输出：
json
{
final_prompt: 一位女性拿起一瓶可乐并喝了一口的电影感特写镜头。场景采用戏剧性的高对比度照明。使用@monalisa作为主体参考，视频中出现的对象是@coke。,
recommended_parameters: {
duration: 8,
aspect_ratio: 16:9
}
}
知识库
本技能依赖内部知识库来做出明智决策。智能体在执行过程中必须查阅以下文件。
- - /references/atomicelementmapping.md：核心知识。包含资产类型 -> 原子元素和原子元素 -> 最佳参考方法映射表。必须在阶段1和阶段2中查阅。
- /references/seedancesyntaxguide.md：Seedance 2.0 @资产名称语法参考。必须在阶段3中查阅，以确保生成正确的语法。
- /references/prompt_templates.md：针对不同风格（如电影感、产品展示、叙事）的高级提示词模板。可在阶段3中选择性查阅，用于风格增强。
