video-downloader

name

video-downloader

description

Download videos from social platforms without watermarks. Use when user wants to download videos from Douyin (抖音), X (Twitter), Bilibili (B站), YouTube, or Xiaohongshu (小红书). Supports MP4 video download and extracting video metadata (title, description, author). Triggers for: "下载视频", "download video", "无水印下载", "抖音视频", "B站视频", "YouTube视频", "小红书视频", "X视频", "Twitter视频", or any request involving these platforms.

Video Downloader Skill

Download videos from major social platforms in MP4 format with metadata extraction.

Supported Platforms

• Douyin (抖音)
• X / Twitter (推特)
• Bilibili (B站)
• YouTube (油管)
• Xiaohongshu (小红书)

Quick Start

Installation

Install required dependencies:

pip install yt-dlp playwright
playwright install chromium

Or use the included setup script:

bash scripts/setup.sh

Download a Video

# Basic download (saves to current directory)
python scripts/download.py "VIDEO_URL"

# Specify output directory
python scripts/download.py "VIDEO_URL" -o /path/to/save

# Download with metadata (JSON)
python scripts/download.py "VIDEO_URL" -o /path/to/save --metadata

Supported URL Formats

Platform	Example URL
Douyin	https://www.douyin.com/video/123456789 or https://v.douyin.com/xxxxx
X/Twitter	https://x.com/user/status/123456789 or https://twitter.com/user/status/123456789
Bilibili	https://www.bilibili.com/video/BVxxxxxx or https://b23.tv/xxxxx
YouTube	https://www.youtube.com/watch?v=xxxxx or https://youtu.be/xxxxx
Xiaohongshu	https://www.xiaohongshu.com/discovery/item/xxxxx

Usage Patterns

Interactive Download

When user provides a URL:

1. Detect platform from URL
2. Run appropriate downloader
3. Save video as {title}_{author}.mp4
4. Extract metadata to {title}_{author}.json

Batch Download

# Download from file (one URL per line)
python scripts/batch_download.py urls.txt -o ./downloads

Metadata Output

By default, saves metadata as JSON:

{
  "title": "视频标题",
  "description": "视频描述",
  "author": "作者昵称",
  "author_id": "作者ID",
  "upload_date": "2024-01-01",
  "duration": 120,
  "likes": 10000,
  "comments": 500,
  "shares": 100,
  "url": "原始URL",
  "video_path": "本地视频路径"
}

Scripts

Script	Purpose
scripts/download.py	Single video download with optional transcription
scripts/transcribe.py	Standalone video transcription
scripts/batch_download.py	Batch download from URL list
scripts/setup.sh	Install dependencies

Error Handling

Common errors and solutions:

Error	Solution
"No video found"	URL may be invalid or private
"Cookie required"	Some platforms need authentication
"Network error"	Check proxy settings

Platform Notes

• YouTube: Best quality, may need proxy
• Bilibili: Works well, supports series
• Douyin (抖音): iesdouyin.com 移动端 API (无需登录无需浏览器) → Playwright fallback
• Xiaohongshu (小红书): May require cookies
• X/Twitter: vxtwitter API 优先 (无需登录无需认证) → yt-dlp + 浏览器 cookies fallback → Playwright fallback

X/Twitter 下载方案（已升级）

X/Twitter 视频下载现在优先使用 vxtwitter.com API，完全不需要登录、不需要浏览器、不需要 cookies：

python scripts/download.py "https://x.com/user/status/123456789"

下载流程：

1. Step 0 (vxtwitter API)：调用 https://api.vxtwitter.com/status/xxx 获取所有视频直链，curl 直接下载
2. Step 1 (yt-dlp + 浏览器 cookies)：尝试从浏览器读取 cookie，支持年龄限制内容
3. Step 2 (Playwright 浏览器提取)：用无头浏览器提取视频 URL，curl/yt-dlp 下载
4. Step 3 (直接视频 URL)：如果 Playwright 能拿到直链，直接 curl 下载

vxtwitter API 优势：

• 无需登录、无需认证
• 不需要浏览器或 cookies
• 支持一条推文中的多个视频
• 响应快，成功率高

抖音下载 (核心方案已升级)

抖音下载现在优先使用 iesdouyin.com 移动端 API，完全不需要登录、不需要浏览器、不需要 cookies：

python scripts/download.py "https://v.douyin.com/xxxxx"
python scripts/download.py "https://www.douyin.com/video/123456789"

工作原理：

1. 解析视频 ID（自动处理短链接重定向）
2. 请求 iesdouyin.com 移动端分享页面
3. 从页面提取 play_addr 中的 video_id
4. 构造无水印播放地址：https://aweme.snssdk.com/aweme/v1/play/?video_id=xxx&ratio=1080p
5. wget 下载（自动断点续传，支持大文件）

如果 API 失败，自动降级到 Playwright 浏览器方案。

视频转录 (语音转文字)

使用 OpenAI Whisper 进行语音转文字，默认输出简体中文。

安装转录依赖

pip install openai-whisper

⚠️ 重要：模型要求

必须使用 medium 或 large 模型，tiny/base/small 模型识别效果差。

# 下载 medium 模型 (约 1.4GB)
# 首次运行时会自动下载

# 转录视频
python scripts/transcribe.py video.mp4 -m medium

# 下载+转录
python scripts/download.py "VIDEO_URL" --transcribe --model medium

Whisper 模型对比

模型	大小	准确率	推荐
tiny/base/small	~40-240MB	较低	❌ 不推荐
medium	1.4GB	较高	✅ 推荐
large	2.9GB	最高	可选

使用示例

# 单独转录
python scripts/transcribe.py ~/Downloads/video.mp4 -m medium

# 下载抖音视频并转录
python scripts/download.py "https://v.douyin.com/xxxxx" --transcribe --model medium

Implementation

This skill uses yt-dlp as the core engine with platform-specific extractors. The downloader automatically handles:

• URL parsing and validation
• Video format selection (prefers no-watermark)
• Metadata extraction
• Filename sanitization
• Error retry logic