VideoSeek

Project Overview

VideoSeek is a local video-material retrieval platform: find shots in your own library with a sentence, a screenshot, or a line of dialogue. Indexing and search run entirely on your machine (ONNX + Lance + FFmpeg) without uploading media, and hits can be previewed and exported straight into editing tools.

Turn digging through folders into searching shots by meaning, with media staying local.

Key Features

• Local library indexing: add folders, sync to extract frames and build embeddings
• Text semantic search: describe a scene and get time ranges, per frame or aggregated chunks
• Image search: find similar shots from a reference or cropped frame, with best-shot and deep-search modes
• Hard-subtitle search: OCR on-screen subtitles and search by dialogue, exact or fuzzy with highlighting
• Preview and export: timeline preview, mp4 clips, Jianying drafts, Premiere XML and DaVinci FCPXML
• Video understanding: optional per-chunk captions, whole-video summaries, recap cuts and local CAM++ speaker clustering
• Tag search: filter shots by combinations like blue hair after understanding runs
• Swappable vision models: openai-clip, chinese-clip, siglip2 in base and large packs
• LAN sharing: one machine indexes as server, others search as clients
• Agent API: localhost HTTP endpoints for semantic, subtitle and tag search

Target Audience

• Short-video and mashup creators
• Owners of large local or NAS media libraries
• Editors prepping rough cuts
• Developers wiring agents into retrieval

Quick Start

1. Get the installer from the site or GitHub, or run from source with pip install -r requirements.txt and python main.py
2. Import runtime assets (models and FFmpeg) when prompted on first launch
3. Add a video folder under Local Library, select files and hit Sync selected
4. Search by text or image in the sidebar, then preview and export

Core Advantages

• Private by design: indexing and search stay on your machine
• Semantic shot search: find by content, not filenames
• Editing-ready: export drafts for Jianying, Premiere and DaVinci
• Free and open source: AGPL-3.0, installer and source share the same UI

System Requirements

Item	Requirement
OS	Windows 10/11 recommended; Linux/macOS need ONNX Runtime adjustments
Dependencies	Python, ONNX Runtime, FFmpeg
License	AGPL-3.0

Notes

• Subtitle search needs the OCR pack and extraction under the Subtitles tab
• Switching vision models requires re-syncing the library
• Video understanding and related features are desktop-only
• Get it from GitHub or the official site; beware paid install-service copies