Task Objective
- This Skill is used to: by analyzing viral Douyin videos, learn their viral methodology (structural framework, expression techniques, style characteristics), and help users createoriginalshort video scripts that follow viral patterns
- Capabilities include: extracting video information, deeply deconstructing viral factors, learning the methodology, and creating entirely new scripts
- Trigger conditions: the user needs to create a short video script, or has a need for Douyin video benchmarking
- Core Principles: Learn the methodology, refuse content copying
Prerequisites
- Dependency description: the dependency packages and versions required by the scripts
yt-dlp>=2024.1.0 beautifulsoup4>=4.12.0 lxml>=4.9.0 requests>=2.31.0
Complete Execution Process (Five-Step Method)
Step 1: Obtain Benchmark Video Information
Two input methods for users to choose from:
Method A: Automatic extraction via Douyin link (recommended to try)
After the user provides a Douyin video link, call the script to extract video information:
python3 /workspace/projects/viral-video-copywriting/scripts/extract_douyin_video.py <抖音视频URL>
Information the script can extract(if successful):
- Video title
- Video description
- Script content (subtitles)
- Background music information (some videos)
- Data such as likes, comments, etc. (some videos)
Important Notes：
- The Douyin platform has strict anti-crawling mechanisms, and video information extraction may be restricted
- If the script cannot extract, please ask the user to directly use Method B
Method B: Manually provide video information (100% reliable)
When automatic extraction fails or it is more convenient for the user, directly provide the following information:
- Video title: the title text of the video
- Video description: the description text posted by the creator
- Video script: the spoken text/subtitle content in the video
- Background music: description of music type, style, rhythm, etc.
- Other information(optional): number of likes, comments, creator name, etc.
The agent will conduct subsequent analysis based on this information.
Step 2: Deeply Deconstruct Viral Factors (Learn the Methodology)
Based on the video information obtained in Step 1, the agent conducts a comprehensive deconstruction and analysis from the perspective of a professional short video creator and operations expert.
Key objective: extract learnable methodology, not summarize the content itself
Analysis dimensions (refer to references/analysis-framework.md）：
1. Opening Hook Analysis
- What is the hook type (suspense, pain point, resonance, reversal, benefit, etc.)?
- What method was used in the first 3 seconds to attract attention?
- This hook method'sprinciplewhat is it?
- How strong is the hook's attraction (1-10 points)?
- Reusable hook techniqueswhat are they?
2. Script Structure Analysis
- What is the overall structure (problem-solution, comparison, story, list, progressive, etc.)?
- How is the content organized in this structure?
- What is the function of each part (attract attention → deliver value → drive action)?
- What is the duration proportion of each part?
- What is the transition rhythm of the content?
- Reusable structural frameworkWhat is it?
3. Expression Style Analysis
- Language style (down-to-earth/professional/humorous/serious/warm, etc.)?
- What is the emotional tone (curiosity/anxiety/pleasure/moved, etc.)?
- What are the sentence features (short sentences/long sentences/combination of long and short)?
- Vocabulary choice characteristics (colloquial/professional/emotional vocabulary)?
- Which expression techniques are used (rhyme, parallelism, memorable quotes, contrast, etc.)?
- Reusable expression techniquesWhat are they?
4. Duration and Rhythm Analysis
- How long do the beginning, middle, and end each take?
- What is the overall rhythm (fast/medium/slow)?
- What is the transition frequency?
- Reusable duration allocation patternsWhat are they?
5. Content Strategy Analysis
- What is the core value point?
- Who is the target audience (age, gender, interests, pain points)?
- What needs or pain points does it hit?
- What is the differentiation advantage?
6. Extraction of Viral Elements
- Which elements are key to this video becoming a hit?
- List the top 3-5 key factors in order of importance
- Reusable patternsWhat are they?
- What points need to be avoided?
Output format: The agent will generate a detailed 《Viral Methodology Breakdown Report》, focusing on extracting learnable techniques, frameworks, and patterns, rather than the content itself.
Step 3: Understand User Needs
The agent guides the user to describe their creative ideas:
Information to understand:
- Core content: What is the core content you want to express?
- Target audience: What kind of user group do you want to reach?
- Delivered value: What value or emotion do you want to convey?
- Initial copy: Do you have initial copy or keywords?
- Other requirements: Are there specific style preferences or restrictions?
The agent will proactively ask clarifying questions, ensuring the creative direction is accurate.
Step 4: Create New Viral Copy
The agent combines the viral patterns of the benchmark video and the user's original content to create new copy.
Core creation principles (most important):
The core value of this Skill islearning the methodology, not copying the content, but the copy length must strictly match the benchmark. The agent must:
-
Structure learning: Borrow the framework, not the content
- Learn the structural logic of the benchmark video (how content is organized, how rhythm is arranged)
- Refill this structural framework with new original content
- Directly copying the specific text content of the benchmark copy is prohibited
-
Style learning: Imitate techniques, not sentences
- Learn the expression style of the benchmark video (down-to-earth/professional/humorous, etc.)
- Learn the sentence structure characteristics of the benchmark video (short sentences/long sentences/combination of long and short)
- Learn the rhetorical devices of the benchmark video (contrast, parallelism, memorable quotes, etc.)
- Directly copying specific sentences or paragraphs from the benchmark copy is prohibited
-
Content originality: 100% original content
- All text in the new copy must be original content created by the user
- The story structure framework may be referenced, but the story itself must be completely different
- Expression techniques may be referenced, but the specific wording must be original
-
Length benchmarking: strictly benchmark word count and duration
- Total copy word count: error within ±5%
- Opening word count: error within ±5%
- Middle word count: error within ±5%
- Ending word count: error within ±5%
- Estimated video duration: error within ±5%
- The user may proactively request length adjustments, but by default it must strictly benchmark
Creation process:
- Structure learning: Analyze the structural framework of the benchmark video (hook type, content organization method, pacing arrangement)
- Style learning: Analyze the language style and expression techniques of the benchmark video
- Content creation: Based on the user's original content, create according to the learned structure and style
- Originality check: Ensure the new copy does not directly copy any sentence or paragraph from the benchmark copy
Output format: The agent will provide 2-3 alternative copy versions in different styles, each containing:
Copy version A
Complete copy content:
[开头钩子：约X字，预计X秒]
[中间内容：约X字，预计X秒]
[结尾总结：约X字，预计X秒]
总计：约X字，预计X秒
Benchmark parameter comparison:
| Parameter | Benchmark copy | New copy | Deviation | Requirement |
|---|---|---|---|---|
| Total word count | X words | X words | X% | Within ±5% |
| Opening word count | X words | X words | X% | Within ±5% |
| Middle word count | X words | X words | X% | Within ±5% |
| Ending word count | X words | X words | X% | Within ±5% |
| Estimated duration | X seconds | X seconds | X% | Within ±5% |
| Structure type | XXX | Learn from benchmark | Consistent | Reference framework |
| Language style | XXX | Learn from benchmark | Consistent | Reference style |
| Sentence structure characteristics | XXX | Learn from benchmark | Consistent | Reference techniques |
Originality statement:
- New copy content: 100% original, no sentence or paragraph from the benchmark copy has been copied
- Referenced content: only structure framework, expression techniques, style characteristics, etc. are referencedMethodology
- Benchmarking principle: learn "how to write", not "what to write"
- Length benchmarking: Strictly benchmark word count and duration, with error controlled within ±5%
Design approach explanation:
- Opening hook: Use the _______ type (learn the hook method of the benchmark video), with original content
- Copy structure: Use the _______ structure (learn the content organization logic of the benchmark video), with original content
- Expression style: Use the _______ style (learn the language style of the benchmark video), with original specific wording
- Sentence pattern characteristics: Maintain the _______ characteristics (learn the rhythm of the benchmark video), with original specific sentences
- Length control: Strictly benchmark the word count, with the total word count controlled within the ±5% range
Highlight annotations:
- Opening hook: "_______" (attract attention)
- Punchline: "_______", (easy to spread)
- Turning point: "_______", (create a memorable point)
Benchmark element transfer:
- Benchmark copy element: "_______",
- New copy transfer: "_______",
- Transfer effect: _______
Suggested filming/performance method:
- Speaking rate suggestion:(Benchmark video:）
- Emotional tone:(Benchmark video:）
- Visual coordination: _______
Copy version B (same format as above)
Copy version C (same format as above)
Step 5: Iterative optimization (optional)
If the user provides feedback on the creation results, the agent supports multiple rounds of iterative optimization:
- Adjust the copy based on user feedback
- Optimize the opening hook
- Adjust the structure or rhythm
- Improve the expression style
Usage example
Complete process example
User input:
我创作一个知识科普类的短视频，这是对标视频的链接：
https://www.douyin.com/video/7300000000000000000
我的想法是讲"如何提高工作效率"，目标受众是职场新人
Step 1: Obtain video information
- The script attempts extraction and successfully obtains:
- Title: "3 tips to double your work efficiency"
- Description: "Essential workplace skills, recommended to save! #Workplace #Efficiency"
- Copy: "Did you know? 90% of people are wasting time. Today I'll share 3 tips..."
Step 2: In-depth breakdown analysis The agent generates an analysis report:
- Opening hook: Suspense type ("Did you know? 90% of people are wasting time")
- Copy structure: Problem-solution structure
- Expression style: Down-to-earth, colloquial
- Core value: Methods to improve work efficiency
- Viral elements: Opening suspense + specific methods + save guide
Step 3: Understand user needs
- Core content: Improve work efficiency
- Target audience: Workplace newcomers
- Delivered value: Practical methods, quick to get started
Step 4: Create new copy The agent provides 3 alternative copy versions (see above for detailed format)
Resource index
- Required script: see scripts/extract_douyin_video.py(Purpose: extract Douyin video information)
- Domain reference: see references/analysis-framework.md(When to read: during Step 2 breakdown analysis)
- Example data: see references/sample-video-data.md(Purpose: demonstrate the complete process)
Notes
- The Douyin platform has strict anti-scraping mechanisms, and video information extraction may be limited
- It is recommended to try method A (automatic extraction) first, and use method B (manual provision) if it fails
- Background music extraction depends on the level of platform openness, and not all videos can be obtained
- When analyzing and breaking down, focus on extracting reusable patterns rather than simply imitating
- When creating new copy, balance "borrowing from viral hits" and "maintaining originality"
- Make full use of the agent's language analysis and creation capabilities
- The more detailed the information provided by the user, the more precise the creation result
Test verification
Test process
Use references/sample-video-data.md The sample data in can be used to fully test the entire process:
- Use the sample data as the input for the first step
- Verify that the second step can generate a detailed analysis report
- Simulate the user providing ideas (third step)
- Verify that the fourth step can generate multiple versions of copy alternatives
- Test the iterative optimization function of the fifth step
Test command
# 测试脚本提取功能
python3 /workspace/projects/viral-video-copywriting/scripts/extract_douyin_video.py "https://www.douyin.com/video/7300000000000000000"