← Back to Skills Marketplace
roseryztzhoutong

Video Frame Analyzer

by roseryztzhoutong · GitHub ↗ · v0.1.0 · MIT-0
cross-platform ⚠ suspicious
85
Downloads
1
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install video-frame-analyzer
Description
AI漫剧/短剧视频智能帧分析工具。自动提取关键帧 → 多模态逐帧解读 → 输出完整分析报告(剧情大纲、镜头语言、角色设定、台词整理、商业改编建议)。适用于抖音/视频号短剧拆解、竞品分析、改编策划。
README (SKILL.md)

视频帧智能分析 Skill

快速开始

使用前提:先切换到多模态模型(支持图片输入),如 GLM-5v-Turbo 或 GPT-4o。 在 WorkBuddy 对话框输入 /model 切换模型。

  1. 上传视频 或直接告诉 AI 视频路径
  2. 发送指令:"分析这个视频" 或 "拆解这条短剧"
  3. AI 自动完成全流程 → 输出完整报告

适用场景

  • 拆解竞品短剧/AI漫剧的叙事结构和镜头语言
  • 分析视频剧情、角色、台词,输出改编建议
  • 批量处理多个视频,建立素材库分析报告

工作流程

第一步:确认视频来源

询问用户视频路径,或让用户上传视频文件到工作目录。 支持 mp4 / mov / avi 等常见视频格式。

第二步:智能提取关键帧

重要:必须先切换到多模态模型(支持图片输入)再开始!

用 Bash 工具执行帧提取脚本:

# 获取 skill 目录路径(自动解析)
SKILL_DIR="$(dirname "$(find ~/.workbuddy/skills -name 'smart_extract.py' 2>/dev/null | head -1)")"
# 执行提取(自动用当前工作目录作为输出)
python "$SKILL_DIR/smart_extract.py" "你的视频路径.mp4" "视频名_frames/" 15

参数说明

  • 视频路径:支持绝对路径或相对于当前工作目录的路径
  • 输出目录:帧图片存放位置,默认为 {视频名}_frames/
  • 最大帧数:默认15帧,覆盖单集短剧完整剧情

如遇 OpenCV 依赖问题,改用保底方案:

python "$SKILL_DIR/extract_frames.py" "你的视频路径.mp4" "视频名_frames/" 3

第三步:逐帧分析(分批,避免上下文溢出)

⚠️ 核心原则:每批最多分析 3-4 帧,分析完立即写入文件,再继续下一批。 严禁一次性将所有帧图发给模型,会导致 400 input length too long 错误。

操作循环

  1. Read 工具读取 3-4 张帧图片
  2. 按以下维度分析每帧:画面内容、字幕/台词、镜头类型、叙事功能、情绪基调
  3. Write 工具将分析结果追加到 {视频名}_分析_partX.md
  4. 清空上下文,继续读取下一批帧,重复步骤1-3

每帧分析模板

## 帧N [时间] 文件名
**场景**:
**画面内容**:
**镜头类型**:(特写/近景/中景/全景/航拍)
**叙事功能**:
**情绪基调**:
**字幕/台词**:

第四步:整合完整报告

所有帧分析完成后,生成 {视频名}_完整分析报告.md,包含以下章节:

  1. 视频基本信息(时长、帧数、AI生成标识)
  2. 逐帧详细分析(表格形式)
  3. 剧情大纲(分幕整理故事线)
  4. 镜头语言分析(镜头类型分布、叙事节奏)
  5. 角色设定(角色表,标注一致性问题)
  6. 字幕/台词整理(时间轴表格)
  7. 商业分析与改编建议(亮点、问题、改编方案)
  8. 技术备注(提取参数、文件路径)

注意事项

模型切换(必须!)

/model GLM-5v-Turbo
# 或
/model GPT-4o
  • 必须使用多模态模型(支持图片输入)
  • 纯文本模型(如 MiMo-V2.5-Pro)无法分析图片,会报错
  • 切换后需重新开始本流程

上下文管理

  • 每批 3-4 帧 → 写文件 → 清空上下文 → 下一批
  • 这是避免 400 input length too long 的核心策略
  • 写文件比清空上下文更可靠,确保中间结果不丢失

AI生成视频的特殊问题

分析时特别注意以下几点(在报告中标注):

  • 角色一致性:性别/外貌是否在场景间突变(如女孩→男孩)
  • AI水印:画面是否有"内容由AI生成"等标识
  • 表情自然度:人物表情是否僵硬、不自然
  • 场景割裂:故事线之间是否有合理过渡(常见拼接问题)

输出文件命名规范

文件类型 命名格式
逐帧分析(分批) {视频名}_分析_part{1,2,3}.md
完整报告 {视频名}_完整分析报告.md
帧图片目录 {视频名}_frames/

常见错误处理

错误 原因 解决方案
ImportError: No module named 'cv2' OpenCV 未安装 pip install opencv-python
ImportError: No module named 'moviepy' moviepy 未安装 pip install moviepy Pillow
400 input length too long 一次性发图太多 改为每批 3-4 帧
无法打开视频 视频路径含中文/空格 用绝对路径或加引号
模型无法识别图片 当前为纯文本模型 先切换到多模态模型
400 Param Incorrect 模型切换后参数不匹配 重新执行读图命令

依赖环境

pip install opencv-python          # 智能帧提取(推荐)
pip install moviepy Pillow           # 保底帧提取(备选)

Python 版本:3.8+


示例触发词

当用户说以下内容时,激活本 Skill:

  • "分析这个视频"
  • "拆解这条短剧"
  • "帮我看看这个AI漫剧的叙事结构"
  • "视频帧分析"
  • "帮我分析竞品视频"
  • "帮我看看这个短剧的剧情"
Usage Guidance
Before installing, be aware that the skill runs local Python scripts, may require installing video-processing packages, writes extracted frames and reports to disk, and sends selected frames to the multimodal model you choose. It appears purpose-aligned and benign, but only analyze videos you are allowed and comfortable to process.
Capability Analysis
Type: OpenClaw Skill Name: video-frame-analyzer Version: 0.1.0 The skill provides legitimate video analysis functionality but employs high-risk behaviors, specifically the execution of shell commands and Python scripts to process user-provided files. The instructions in SKILL.md direct the AI agent to run bash commands that interpolate user-provided video paths (e.g., `python "$SKILL_DIR/smart_extract.py" "你的视频路径.mp4"`), which presents a shell injection vulnerability if the agent does not properly sanitize the input. While the included scripts (smart_extract.py and extract_frames.py) appear benign and focused on frame extraction, the reliance on shell execution and broad file system access for writing frames and reports meets the criteria for a suspicious classification.
Capability Assessment
Purpose & Capability
The stated purpose, instructions, and included Python files are coherent: they extract frames from a user-provided video and guide the agent to create an analysis report.
Instruction Scope
The workflow asks the agent to use Bash, read generated image files, and write reports. These actions are purpose-aligned, but users should confirm the video path and output directory.
Install Mechanism
There is no install spec or declared required binary, but the documentation asks users to install Python packages such as opencv-python, moviepy, and Pillow. This is user-directed and expected for the skill, but it is under-declared in metadata.
Credentials
The skill processes local video files and reads extracted frames into a multimodal model. That is necessary for the feature, but videos may contain private or copyrighted content.
Persistence & Privilege
No credentials, elevated privileges, background workers, or persistent autonomous behavior are requested. The only persistence shown is generated frame images and analysis report files.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install video-frame-analyzer
  3. After installation, invoke the skill by name or use /video-frame-analyzer
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v0.1.0
Initial release of video-frame-analyzer skill. - Automatically extracts key frames from AI short-drama videos. - Performs multimodal frame-by-frame analysis (visual, dialogue, shot type, narrative, emotion). - Outputs a complete structured report: plot summary, cinematography, character design, dialogue, business adaptation suggestions. - Workflow optimized to handle context limits: analyzes frames in small batches, writes results incrementally. - Supports popular video formats (mp4, mov, avi); works with multimodal models like GLM-5v-Turbo or GPT-4o. - Useful for short-video deconstruction, competitor analysis, and content adaptation planning.
Metadata
Slug video-frame-analyzer
Version 0.1.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is Video Frame Analyzer?

AI漫剧/短剧视频智能帧分析工具。自动提取关键帧 → 多模态逐帧解读 → 输出完整分析报告(剧情大纲、镜头语言、角色设定、台词整理、商业改编建议)。适用于抖音/视频号短剧拆解、竞品分析、改编策划。 It is an AI Agent Skill for Claude Code / OpenClaw, with 85 downloads so far.

How do I install Video Frame Analyzer?

Run "/install video-frame-analyzer" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Video Frame Analyzer free?

Yes, Video Frame Analyzer is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Video Frame Analyzer support?

Video Frame Analyzer is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Video Frame Analyzer?

It is built and maintained by roseryztzhoutong (@roseryztzhoutong); the current version is v0.1.0.

💬 Comments