PICKVEDIO / 项目文档PICKVEDIO / DOCUMENTATION
操作指南User guide
首次运行Getting started
桌面软件面向 Windows 10 / 11，建议使用 64 位 Python 3.11 或 3.12。先安装 FFmpeg，加入 PATH 或在软件设置中填写 ffmpeg.exe 路径。The desktop app targets Windows 10 / 11 with 64-bit Python 3.11 or 3.12. Install FFmpeg and add it to PATH, or set the ffmpeg.exe path in the app.
本次公开仓库是介绍网页源码。桌面项目目前为本地准备的 0.1.0 源码候选版，尚未提供新的桌面程序下载。以下命令仅用于完整桌面项目，不可在网页仓库中执行。The public repository contains this presentation website. The desktop app is a locally prepared 0.1.0 source candidate; a new desktop download is not available yet. The commands below apply to the complete desktop project, not this website repository.
获取完整桌面源码后，在包含 app.py 与 requirements.txt 的项目根目录执行：Once you have the complete desktop source, run these commands in the directory containing app.py and requirements.txt:
powershell -ExecutionPolicy Bypass -File .\scripts\setup_dev.ps1
.\.venv\Scripts\python.exe .\app.py配置模型Configure models
在“设置 → 本地处理”选择 Whisper 模型、设备、计算类型和语言。首次使用模型名称可能联网下载权重，也可指定已有 faster-whisper 模型目录。In Settings → Local processing, choose the Whisper model, device, compute type, and language. A model name may download weights on first use; you can instead provide an existing faster-whisper model directory.
在“设置 → AI 与提示词”选择本机 Ollama 或云端 API。Ollama 模型需提前安装，地址通常为 http://127.0.0.1:11434；选择适合本机内存和显存的模型。In Settings → AI and prompts, choose local Ollama or a cloud API. Install your Ollama model first; the usual address is http://127.0.0.1:11434. Choose a model that fits your RAM and VRAM.
使用云端时填写你自己的 Key，扫描或手填实际可用的模型，并测试连接。API Key 存在系统凭据库，不会写入 settings.json。For a cloud service, provide your own key, discover or enter an available model, and test the connection. API keys are stored in the system credential store, not settings.json.
处理与阅读视频Process and read videos
粘贴 B 站或 YouTube 视频链接或分享文案，可先读取元数据检查标题和时长，然后点击“开始一键总结”。Paste a Bilibili or YouTube video link or shared text. You can inspect metadata first, then start summarization.
批量队列一行一个视频，支持剪贴板和 TXT / Markdown 链接列表。按顺序处理，一条失败后继续下一条；每条任务独立导出。Use one video per queue row, or import links from the clipboard or TXT / Markdown. Tasks run sequentially and export separately; one failure does not stop the next item.
总结导出为 Markdown、TXT、JSON，逐字稿导出为 TXT、JSON、SRT、VTT。当前没有本地视频文件、字幕或逐字稿直接导入入口。Summaries export to Markdown, TXT, and JSON. Transcripts export to TXT, JSON, SRT, and VTT. Direct local video, subtitle, and transcript import is not currently available.
阅读器可切换总结、逐字稿、任务说明和 AI 追问。追问依据总结资料，不保证覆盖完整逐字稿；重要结论应回到原话核对。The reader offers summary, transcript, task details, and AI follow-up tabs. Follow-up uses summary material and may not cover the full transcript. Verify important conclusions against the original words.
当前保留下载媒体和中间文件。删除历史条目只删除数据库索引，不删除输出文件。Downloaded media and intermediate files are retained. Deleting a history item removes its database record, not its output files.
Chrome 浏览器扩展Chrome extension
先在桌面版保存模型配置，再在完整桌面项目根目录运行本机桥接：Save the desktop model settings first, then start the local bridge from the complete desktop project root:
.\.venv\Scripts\python.exe .\bridge.py在 chrome://extensions 开启开发者模式，加载桌面源码中的 browser_extension/ 文件夹。扩展通过 127.0.0.1:8765 把当前视频提交到本机队列。Enable developer mode in chrome://extensions and load browser_extension/ from the desktop source. The extension submits the current video to the local queue at 127.0.0.1:8765.
修改桌面设置后重启桥接。关闭弹窗不会取消已提交任务；关闭桥接会停止处理。不要通过端口转发或隧道把桥接暴露到外网。Restart the bridge after changing desktop settings. Closing the popup does not cancel submitted tasks; closing the bridge stops processing. Do not expose it through port forwarding or a tunnel.
常见问题与排错Troubleshooting
找不到 FFmpeg：检查 PATH 或设置里的真实路径。Qt DLL 错误：使用项目独立环境与固定的 PySide6 6.8.3，避免混用 Conda 的 Qt。FFmpeg missing: check PATH or the configured path. Qt DLL errors: use the project environment and pinned PySide6 6.8.3; avoid mixing Conda Qt libraries.
视频访问失败：检查权限、链接和网络；必要时选择已经合法登录的 Cookie 浏览器。只处理有权下载、转写和总结的内容。Video access failures: check permissions, the link, and connectivity. If needed, choose a browser you have legitimately signed into. Process only content you have permission to download, transcribe, and summarize.
Ollama 5xx：检查模型、服务与可用内存，或改用较小模型。API 401 / 404：核对 Key、端点、模型和账号权限。Ollama 5xx: check the model, service, and free memory, or use a smaller model. API 401 / 404: verify the key, endpoint, model, and account permissions.
取消可能等待当前网络或模型调用返回。当前没有自动断点续跑；重跑会创建新任务并重新执行完整流程。Cancellation may wait for the current network or model call to finish. Automatic checkpoint resume is not implemented; rerunning creates a new task and repeats the full pipeline.