AI视频剪辑Skill

全自动AI视频剪辑Skill。当用户请求以下操作时触发： - "帮我剪辑视频"、"自动剪辑"、"AI剪辑视频" - "剪辑电影素材"、"批量剪辑视频"、"自动生成视频" - "视频素材自动处理"、"从素材自动生成成片" - "制作短剧集"、"剪辑短视频"、"自动导出视频" - "素材自动导入"、"视频自动添加字幕...

Security Scan

The skill appears to implement the claimed AI video-editing features, but the package metadata omits important runtime dependencies (FFmpeg/Whisper, pip requirements) and includes some platform-specific defaults — these mismatches warrant caution before installing or running.

Purpose & Capability

Name/description match the included scripts: analysis, automatic clipping, audio processing, subtitles, effects, and export. However the skill registry claims no required binaries or primary credentials while the SKILL.md and scripts explicitly depend on FFmpeg, Whisper (or openai-whisper), and several Python packages (moviepy, Pillow, numpy, etc.). The omission of required binaries in the metadata is an inconsistency and affects install/runtime expectations. Also many examples and default paths are Windows-oriented (e.g., D:\ paths) while OS restriction is 'none'.

Instruction Scope

SKILL.md instructs the agent to run local Python scripts that scan directories, read media files, and write outputs — that behavior is consistent with the stated purpose. The scripts operate on local filesystem paths and call ffmpeg via subprocess, and the Whisper integration may download or load large models (network/disk activity) during transcription. SKILL.md mentions cloud-synced storage ('金山文档同步目录') but no code integrates with remote APIs; this is a documentation/expectation mismatch rather than hidden exfiltration. No instructions reference unrelated system credentials or external endpoints.

Install Mechanism

There is no formal install spec in the registry (instruction-only), but repository files include requirements.txt and INSTALL.md with pip commands. Dependencies are from PyPI (standard) and FFmpeg is recommended from official sources. No high-risk downloads or obscure hosts are used in the repo. The lack of a registry-level install declaration (e.g., required binaries) is confusing but the repository itself uses conventional install instructions.

Credentials

The skill declares no environment variables or credentials which is appropriate because it operates on local files. It does, however, require filesystem write access for output and substantial disk for Whisper models. The required external programs (ffmpeg) and Python packages are not declared in the registry metadata, so the agent/user could be missing preconditions at runtime. There are no requests for secrets or unrelated credentials.

Persistence & Privilege

The skill does not request always:true, does not declare elevated privileges, and does not attempt to modify other skills or system-wide agent settings. It writes outputs to user-specified/local directories only (configurable).

What to consider before installing

This package implements local AI-driven video editing and appears to be what it claims, but pay attention to these points before installing or running it: - Dependencies: You must install FFmpeg (system binary) and the Python packages listed in requirements.txt (moviepy, Pillow, numpy, openai-whisper or whisper, pydub, etc.). The registry metadata incorrectly lists no required binaries — don't assume anything is preinstalled. - Whisper model download: Transcription uses Whisper; loading certain models will download large files and require disk space and network access. Expect significant storage and potentially long downloads on first run. - File access: The scripts scan directories and read any media files they find under the provided input path and will write outputs to the configured output path. Point the skill at non-sensitive test data first. - Platform differences: Defaults and examples use Windows-style paths. Adjust paths for macOS/Linux before running. - Review code locally: The scripts call ffmpeg via subprocess.run extensively (expected for media processing). If you have security concerns, review the scripts and run them in an isolated environment (container or VM) and avoid running as an administrator. - No secrets requested: The skill does not ask for tokens/keys — a good sign — but confirm you are OK with the skill reading arbitrary files under any input directory you provide. If you want this skill to be more trustworthy for automated use, ask the author/maintainer to: (1) declare required binaries and OS constraints in the registry metadata, (2) provide an explicit install spec, and (3) document Whisper model behavior and disk requirements. If you need help verifying or sandboxing the skill, run its tools on small test folders first.

全自动化AI视频剪辑解决方案，实现从素材导入到成片导出的全流程无人干预。

核心能力

1. 素材智能处理

• 自动识别：解析视频(mp4/mov/avi/mkv)、音频(mp3/wav)、图片(jpg/png)格式
• 内容分析：识别场景、人物、动作、文字、音频类型
• 智能筛选：自动剔除黑屏、模糊、杂音等无效素材

2. 剪辑逻辑自主决策

• 风格匹配：根据内容自动确定剪辑风格（卡点/叙事/快节奏/慢节奏）
• 片段选取：自动提取高光片段，智能排序确保流畅
• 转场添加：根据场景切换自动匹配淡入淡出/叠化/闪白等转场
• 时长控制：精准匹配预设时长（15秒/30秒/1分钟/5分钟等）

3. 音频智能优化

• 人声增强：自动降噪、增强人声清晰度
• 背景音乐：根据视频节奏自动匹配、卡点同步
• 音效添加：识别关键动作自动添加对应音效

4. 字幕自动生成

• 语音转文字：自动识别台词，支持中英双语
• 智能排版：自动调整位置、字体、颜色
• 错误修正：自动识别并修正错别字

5. 特效与优化

• 滤镜匹配：根据场景（风景/人物/夜景）自动添加
• 画面优化：自动调整亮度、对比度、饱和度
• 特效添加：关键片段自动添加光晕/粒子/缩放特效

6. 全自动导出

• 智能参数：根据素材自动设置分辨率/帧率/比特率
• 自动存储：按"日期+主题+时长"命名，存储至预设路径
• 异常处理：导出失败自动重试或记录日志

使用方式

基础剪辑命令

用户："帮我剪辑D:\电影下的搞笑片段，生成3-5分钟的搞笑集锦"

执行流程

Step 1: 素材准备

确认素材来源路径，使用 scripts/analyze_media.py 分析素材内容：

python scripts/analyze_media.py --input "D:\电影" --output "D:\AI视频剪辑\素材分析"

Step 2: 配置剪辑参数

根据用户需求设置：

• 目标时长：15秒/30秒/1分钟/3分钟/5分钟
• 输出格式：mp4（推荐）/mov
• 分辨率：保持原分辨率或指定（如1080p）
• 存储路径：金山文档同步目录或本地指定路径

Step 3: 执行自动剪辑

使用 scripts/auto_clip.py 执行全自动剪辑：

python scripts/auto_clip.py --config "配置文件路径"

Step 4: 字幕与特效

使用 scripts/add_subtitles.py 生成字幕：

python scripts/add_subtitles.py --input "成片路径" --output "带字幕版本"

Step 5: 最终导出

使用 scripts/export_final.py 导出成片：

python scripts/export_final.py --input "待导出视频" --preset "高质量/标准/压缩"

脚本说明

脚本	功能	输入	输出
analyze_media.py	素材分析	素材目录	分析报告JSON
auto_clip.py	自动剪辑	分析报告+配置	中间视频文件
audio_process.py	音频处理	视频文件	处理后音频
add_subtitles.py	字幕生成	视频文件	带字幕视频
add_effects.py	特效添加	视频文件	添加特效后视频
export_final.py	最终导出	处理后视频	成片文件

配置模板

默认配置 (references/default_config.yaml)

output:
  format: mp4
  resolution: "1920x1080"
  frame_rate: 30
  bitrate: "8M"
  
subtitle:
  enabled: true
  language: "zh-CN"
  position: "bottom_center"
  font_size: 36
  color: "white"
  
audio:
  bgm_volume: 0.3
  voice_volume: 1.0
  enhance_voice: true
  
style:
  transition: "fade"  # fade/ dissolve/ cut/ flash
  transition_duration: 0.5
  filter_preset: "auto"  # auto/ vivid/ vintage/ cool/ warm
  
export:
  storage_path: "D:\AI视频剪辑\成品"
  naming: "{date}_{theme}_{duration}"

适用场景

场景	推荐配置	预期时长
搞笑集锦	快节奏卡点+音效	3-5分钟
电影解说	叙事风格+字幕	5-15分钟
短视频	高潮片段+滤镜	15-60秒
Vlog剪辑	自然过渡+BGM	3-10分钟
教程视频	清晰叙事+标注	5-30分钟

技术依赖

• FFmpeg：视频处理核心引擎
• Whisper：语音识别与字幕生成
• MoviePy：Python视频编辑库
• Pillow：图像处理
• NumPy：数值计算

注意事项

1. 首次使用：需要配置素材路径、输出路径等基础参数
2. 素材要求：建议使用清晰、无严重抖动的高质量素材
3. 性能：1-5分钟视频剪辑约需2-3分钟完成
4. 存储：确保输出路径有足够空间

踩坑经验

• 素材路径包含中文时，FFmpeg命令需要使用UTF-8编码
• 字幕生成依赖Whisper模型，首次使用需下载模型文件
• 批量剪辑时建议使用队列管理，避免内存溢出