zhouxiaoka / autoclip
zhouxiaoka / autoclip
AutoClip turns long videos into shareable highlights. AutoClip 把长视频变成值得分享的精彩片段。
简体中文 · English · 日本語 · 한국어 · Español · Português · Русский · Français Simplified Chinese · English · Japanese · Korean · Spanish · Portuguese · Russian · French
项目网站 · 反馈问题 Project Website · Report Issues
桌面安装包: macOS · Apple Silicon · Windows · x64 Desktop Installers: macOS · Apple Silicon · Windows · x64
安装与第一次出片 · 完整排错指南 Installation & First Export · Full Troubleshooting Guide
自 v1.3.1 起，产品界面、官网和 README 均支持中、英、日、韩、西、葡、俄、法。顶栏可切换界面语言或跟随系统；用户素材和生成内容保留原文。 Since v1.3.1, the product interface, official website, and README support Chinese, English, Japanese, Korean, Spanish, Portuguese, Russian, and French. The top bar allows switching the interface language or following the system settings; user materials and generated content remain in their original language.
AutoClip 用 AI 分析视频字幕、定位高光、生成标题，并自动剪出片段与合集。适合访谈、播客、课程和直播回放，提供桌面应用、Docker Web 界面和 CLI / MCP 三种使用方式。 AutoClip uses AI to analyze video subtitles, locate highlights, generate titles, and automatically clip segments and compilations. It is suitable for interviews, podcasts, courses, and live stream replays, offering three usage modes: Desktop App, Docker Web Interface, and CLI / MCP.
界面预览 v1.3.0 真实 Web 界面：在文件导入区添加本地视频，可同时提供 SRT 字幕。 Interface Preview v1.3.0 Real Web Interface: Add local videos in the file import area, with support for simultaneous SRT subtitle provision.
社区成就 以下徽章由 Trendshift 提供，点击可查看 AutoClip 的上榜记录。GitHub Trending 与 Trendshift 是不同榜单；徽章展示平台记录的成就，不代表当前实时排名。 Community Achievements: The following badges are provided by Trendshift; click to view AutoClip’s ranking history. GitHub Trending and Trendshift are different lists; the badges display achievements recorded by the platform and do not represent current real-time rankings.
你可以做什么 / What You Can Do
| 能力 / Capability | 说明 / Description |
|---|---|
| 导入素材 / Import | 支持本地视频、YouTube 与 B 站链接，可附带 SRT 字幕。 / Supports local videos, YouTube, and Bilibili links, with optional SRT subtitles. |
| 发现高光 / Highlight Discovery | 从字幕提取大纲、话题时间线、精彩度评分和片段标题。 / Extracts outlines, topic timelines, highlight scores, and segment titles from subtitles. |
| 剪辑与合集 / Clipping & Compilations | 自动生成视频切片，组合推荐合集，也可手动调整顺序。 / Automatically generates video clips, suggests compilations, and allows manual reordering. |
| 发布导出 / Export | 提供抖音、小红书、YouTube Shorts 和 B 站导出预设，支持烧录字幕与标题卡。 / Provides export presets for Douyin, Xiaohongshu, YouTube Shorts, and Bilibili, supporting burned-in subtitles and title cards. |
| 自由选择模型 / Model Flexibility | 支持通义千问、OpenAI 兼容接口、Gemini、硅基流动，以及 Ollama / LM Studio 本地模型。 / Supports Qwen, OpenAI-compatible APIs, Gemini, SiliconFlow, and local models via Ollama / LM Studio. |
| 接入自动化 / Automation | 通过 CLI 批量编排，或让支持 MCP 的客户端调用同一条处理流水线。 / Batch orchestration via CLI, or call the same processing pipeline via MCP-supported clients. |
处理流程 / Workflow: 导入视频 → 准备字幕 / 语音转写 → AI 分析与评分 → 生成切片与合集 → 导出成片 Import Video → Prepare Subtitles / Transcription → AI Analysis & Scoring → Generate Clips & Compilations → Export Final Video
快速开始 / Quick Start
1. 桌面版 / Desktop Version
从 GitHub Releases 下载适合你的安装包： Download the appropriate installer from GitHub Releases:
| 平台 / Platform | 安装方式 / Installation |
|---|---|
| macOS · Apple Silicon | .dmg |
| Windows 10 / 11 · x64 | -setup.exe |
Intel Mac / Linux 使用下方 Docker 或 CLI。桌面安装包内置 Python 和 FFmpeg。实际支持的平台与首次启动说明以对应 Release 为准。安装后先在设置中选择模型提供商、测试连接并保存，再导入视频。 For Intel Mac / Linux, use Docker or CLI below. Desktop installers include Python and FFmpeg. Refer to the specific Release for actual supported platforms and first-launch instructions. After installation, select your model provider in settings, test the connection, save, and then import your video.
2. Docker / Web
需要 Docker 和 Docker Compose v2。以下命令在仓库根目录执行： Requires Docker and Docker Compose v2. Execute the following commands in the repository root:
git clone https://github.com/zhouxiaoka/autoclip.git
cd autoclip
cp env.example .env启动前编辑 .env：选择 LLM_PROVIDER，填写对应服务的 API Key 和模型名；也可以启动后在设置页配置。
Edit .env before starting: select LLM_PROVIDER, fill in the API Key and model name; or configure it on the settings page after startup.
mkdir -p data logs uploads
docker compose up -d --build打开 Web 界面；API 文档在后端启动后可用。部署细节见 Docker 指南（中文）。 Open the Web interface; API documentation is available after the backend starts. See the Docker Guide (Chinese) for deployment details.
Linux 上若绑定目录出现权限错误，先执行以下命令修正项目数据目录的归属，再重新启动服务： If you encounter permission errors on Linux regarding bound directories, run the following commands to fix ownership before restarting:
docker compose run --rm --no-deps --user root --entrypoint sh autoclip -c 'chown -R autoclip:autoclip /app/data /app/logs /app/uploads'
docker compose up -d3. CLI / MCP
需要 Python 3.10+（建议 3.11）和 PATH 中可用的 FFmpeg。以下安装示例使用 macOS / Linux shell；Windows PowerShell 用 venv\Scripts\Activate.ps1 激活虚拟环境。CLI 本地处理不需要 Redis。
Requires Python 3.10+ (3.11 recommended) and FFmpeg in your PATH. The following example uses macOS / Linux shell; for Windows PowerShell, use venv\Scripts\Activate.ps1 to activate the virtual environment. CLI local processing does not require Redis.
git clone https://github.com/zhouxiaoka/autoclip.git
cd autoclip
python3 -m venv venv
source venv/bin/activate
python -m pip install -r requirements.txt
python -m pip install -e .本地模型示例：先安装并启动 Ollama，再下载模型。无字幕视频需要 faster-whisper，首次转写会下载语音模型；已有字幕可用 --srt talk.srt。
Local model example: Install and start Ollama, then download the model. Videos without subtitles require faster-whisper; the first transcription will download the speech model. If you already have subtitles, use --srt talk.srt.
ollama pull qwen2.5:7b
python -m pip install faster-whisper
autoclip doctor --provider ollama
autoclip run talk.mp4 --provider ollama --json把 PROJECT_ID 替换为处理结果中的项目 ID，即可导出 Shorts 格式；用 autoclip mcp 启动 stdio MCP 服务：
Replace PROJECT_ID with the ID from the processing results to export in Shorts format; use autoclip mcp to start the stdio MCP service:
autoclip export PROJECT_ID --preset shorts
autoclip mcp在 MCP 客户端中将 command 设为虚拟环境里 autoclip 的绝对路径，args 设为 ["mcp"]。详见 CLI / MCP 指南（中文）和 Agent skill（中文）。
In your MCP client, set the command to the absolute path of autoclip in your virtual environment, and args to ["mcp"]. See the CLI / MCP Guide (Chinese) and Agent skill (Chinese) for details.
模型配置 / Model Configuration
| 方式 / Method | 配置 / Configuration |
|---|---|
| 云端模型 / Cloud Models | 在设置中选择通义千问、OpenAI 兼容接口、Gemini 或硅基流动，填写 API Key；兼容接口可配置 Base URL。 / Select Qwen, OpenAI-compatible, Gemini, or SiliconFlow in settings; fill in API Key; Base URL is configurable for compatible interfaces. |
| Ollama | 服务地址默认为 http://localhost:11434/v1，默认模型 qwen2.5:7b，无需 API Key。 / Default service address: http://localhost:11434/v1, default model: qwen2.5:7b, no API Key required. |
| LM Studio | 加载模型并启动 Local Server，默认地址 http://localhost:1234/v1，选择服务实际提供的模型。 / Load model and start Local Server, default address: http://localhost:1234/v1, select the model provided by the service. |
Docker 访问宿主机模型服务时，localhost 指向容器自身；需配置容器能访问的宿主机地址。详见 CLI / MCP 指南。视频剪辑在本地进行，云端模型分析会向所选服务发送字幕文本；下载视频与模型仍需要网络。 When Docker accesses host model services, localhost points to the container itself; you must configure the host address accessible to the container. See the CLI / MCP Guide. Video clipping is performed locally; cloud model analysis sends subtitle text to the selected service; downloading videos and models still requires an internet connection.
常见问题 / FAQ
需要付费或 API Key 吗？ AutoClip 本身免费、开源（MIT）。云端模型由所选服务商计费，需要自己的 API Key；Ollama / LM Studio 本地预设无需云端 Key，但需要模型和相应硬件。 Do I need to pay or provide an API Key? AutoClip is free and open-source (MIT). Cloud models are billed by the provider and require your own API Key; Ollama / LM Studio local presets do not require a cloud key but require models and appropriate hardware.
我的视频会上传吗？ 本地剪辑在你的设备上完成；使用云端模型时，字幕文本会发送给该服务商。主动使用发布上传功能时，视频会发送到目标平台。统计与错误报告取决于版本和设置，详见隐私说明。 Will my videos be uploaded? Local clipping is done on your device; when using cloud models, subtitle text is sent to the provider. When using the publishing feature, videos are sent to the target platform. Statistics and error reporting depend on the version and settings; see the Privacy Policy.
没有字幕也能使用吗？ 可以，需要先准备本地 Whisper 组件和语音模型。已有字幕时可同时导入 SRT；准确字幕通常能减少转写等待和识别错误。 Can I use it without subtitles? Yes, you need to prepare the local Whisper component and speech model. If you already have subtitles, you can import SRT files; accurate subtitles usually reduce transcription wait times and recognition errors.
为什么没有生成片段？ 先检查失败阶段：字幕是否为空、模型连接是否成功、评分阈值是否过高，以及 FFmpeg 和磁盘是否正常。可以尝试把评分阈值从 0.7 降到 0.5，但不保证一定有片段。 Why are no clips generated? Check the failure stage: are subtitles empty? Is the model connection successful? Is the score threshold too high? Are FFmpeg and disk space normal? You can try lowering the score threshold from 0.7 to 0.5, though this does not guarantee clips will be generated.
什么视频更适合？处理要多久？ 当前分析主要基于字幕，适合访谈、播客、课程和口播。纯视觉动作或音乐类视频效果可能有限。耗时取决于时长、硬件、模型与导出设置，建议先用短样本验证。 What videos are suitable? How long does it take? Analysis is primarily based on subtitles, making it suitable for interviews, podcasts, courses, and talking-head videos. Purely visual action or music videos may have limited results. Processing time depends on duration, hardware, model, and export settings; it is recommended to test with a short sample first.
参与贡献与联系 / Contribution & Contact
欢迎提交修复、使用反馈和翻译改进。报告问题时请附上系统、版本、所选模型、复现步骤及已脱敏的错误日志。 Contributions, feedback, and translation improvements are welcome. When reporting issues, please include your system, version, selected model, reproduction steps, and sanitized error logs.
个人业余维护，回复时间不固定，不提供即时客服或一对一部署服务。联系前请先查看常见问题与已知问题。 Independently maintained in spare time; response times vary. No instant customer support or one-on-one deployment services provided. Please check the FAQ and Known Issues before contacting.
邮箱: christine_zhouye@163.com Email: christine_zhouye@163.com
感谢 FastAPI、React、Tauri、FFmpeg、yt-dlp、Whisper，以及所有贡献者。项目采用 MIT License。如果 AutoClip 帮到了你，欢迎给项目一个 Star。 Thanks to FastAPI, React, Tauri, FFmpeg, yt-dlp, Whisper, and all contributors. The project is licensed under the MIT License. If AutoClip has helped you, please consider giving the project a Star.