Skip to content
6v17Public

About

Offline video material retrieval platform for semantic search, subtitle search, Agent API, and AI-assisted editing.

Topics

Resources

Stars

91 stars

Watchers

0 watching

Forks

Repository files navigation

VideoSeek

中文说明 | English

Search a local video library by a sentence, tags, a screenshot, or a line of dialogue. Preview the hit, then export.
Indexing and retrieval run on your machine (ONNX + Lance + FFmpeg); your media files are not uploaded.

Personal open-source utility. Windows-first; installer and from-source builds share the same UI.
Free forever. Don’t pay for resale or “paid install” copies — get it from GitHub / the site.

Features

Core (available after you sync a library)

Feature Description
Local libraries Add folders, sync to extract frames and build embeddings
Text search Describe a scene; get time ranges in indexed videos, per frame or aggregated into chunks
Image / screenshot search Find similar shots from a reference or cropped frame. One of: frame (default), best per video, deep search, chunk
Hard-subtitle search OCR on-screen subtitles, then keyword search by time; exact match, or fuzzy (complete query spans first, then scatter hit-rate, with highlighting)
Scope & presets All libraries, selected libraries/videos; Compose search can save text plus reference images as a preset
Preview & export Timeline preview and mp4 clips. A shot basket can also export a Jianying draft, Premiere XML, or DaVinci FCPXML. The installer includes the Jianying export dependency

Optional

Feature Description
Video understanding Desktop-only optional: per-chunk captions + whole-video summary; recap cuts (plan → match shots → captions); local CAM++ speaker clustering (resources/asr/campplus.*)
Tag search After understanding has tagged shots, filter by combinations such as “blue hair” or “red-and-white outfit”
Vision models openai-clip, chinese-clip, or siglip2, each in a base and a large pack; switching models means re-syncing that library
LAN sharing One machine indexes as the server; other machines search as clients. Media stays on the server
Localhost Agent API HTTP on 127.0.0.1: semantic, subtitle, and tag search; list libraries/videos; export clips or a timeline (see docs/for-agents.md; no understanding endpoints)
Video download Resolve page links, download into a local folder, then sync like a normal library
Optional plugins Load extension pages before startup via VIDEOSEEK_PLUGINS or profile/plugins.json (nav / package kinds / i18n). The open-source tree ships no third-party plugin code.

Who is it for?

  • Large local or NAS media folders — find semantically similar shots, not filenames
  • Rough-cut prep before editing in a NLE
  • Local models plus optional HTTP access for Cursor / other agents

Screenshots

搜索界面

Download

Prefer not to manage Python? Get the installer from lv17.top. On first launch, import runtime assets (models, FFmpeg) as prompted.

Minimal workflow

  1. Runtime assets — First-launch prompts, or Settings → import models and FFmpeg.
  2. Add a library — Sidebar Local Library → Videos tab → add a folder.
  3. Sync selected — Check videos → Sync selected; wait for indexing.
  4. Search — Sidebar Search → text or image query.

Video understanding, subtitle extraction, and the Agent API are optional — sync a library first (subtitle search also needs the OCR pack and extraction under the Subtitles tab).

Illustrated walkthrough (Chinese): User guide (Feishu).

Community

  • User guide (ZH): Feishu doc
  • QQ group: 1033551438
  • GitHub Issues — especially for source builds and the Agent API

From source

Windows 10/11 recommended. Linux / macOS: adjust ONNX runtime (see docs/quickstart.md).

pip install -r requirements.txt
python main.py

Or Conda: conda env create -f environment.yml → conda activate VideoSeek.

Missing runtime assets (import in-app for both installer and source):

Missing Fix
Models, FFmpeg In-app prompts, or 123 cloud zip → Import and Parse; or download model zips and ffmpeg.exe from GitHub Releases — models, then import
VLC (Windows source only) Download vlc_lib.zip from GitHub Releases — vlc_lib and extract at project root

Fallback: if the model zip and ffmpeg.exe are already on disk, an agent can close VideoSeek, download import_runtime_resources.py, and run this. No git checkout is required; it writes the installed app's %LOCALAPPDATA%\VideoSeek:

python import_runtime_resources.py openai-clip.zip ffmpeg.exe

See docs/quickstart.md for troubleshooting and tests.

Docs

Doc Topic
User guide (Feishu, ZH) Illustrated setup for installer users
docs/quickstart.md From-source setup, troubleshooting, tests
docs/for-agents.md Localhost Agent API
docs/architecture.md Architecture (developers)

License

Copyright (c) 2026 6v17

Licensed under AGPL-3.0 (GNU Affero General Public License v3.0).

Thanks for your support

Sincere thanks to everyone who tipped VideoSeek — your encouragement keeps the project going. Listed by date; amounts are not shown.

To use your own avatar, email 2627538472@qq.com with your WeChat nickname and an image.

Avatar Nickname Message Date
热心用户 很好用!感谢!😋 2026-07-13
岁岁平安 软件很牛逼,你也很牛逼 2026-07-17
榨一杯橙汁 默默支持! 2026-07-29
😊 默默支持! 2026-07-29

About

Offline video material retrieval platform for semantic search, subtitle search, Agent API, and AI-assisted editing.

Topics

Resources

Stars

91 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages