Sift Video

Paid APISelf-hostedWeb

What Sift Video is

Sift Video is an open-source project for finding relevant video moments through natural-language searches of audio and visual content. It processes videos offline, turns audio and frames into semantic embeddings, and indexes them in Qdrant. Results include video names, timestamps, match types, transcript excerpts, and relevance scores, with an option to open the video at the matching moment. Searches can cover audio, visual content, or both, and can be limited to one selected video. The web interface accepts MP4, WebM, MKV, MOV, AVI, and OGV uploads up to 1 GB. The project can also download yt-dlp-compatible URLs and stream ingestion progress through server-sent events. Its REST API documents search, downloads, uploads, video listing and streaming, deletion, and health endpoints. Docker Compose instructions cover CPU deployment, with a separate NVIDIA GPU configuration; GPU use requires NVIDIA Container Toolkit on the host. The documented stack includes Whisper, MobileCLIP2, FFmpeg, and Qdrant. The repository states that the project is licensed under GNU GPLv3 and invites issue and pull-request contributions.

Who it suits

Sift Video suits people who can follow Docker and Git setup instructions and want to search video audio and visuals locally. Users with an NVIDIA GPU can follow the separate GPU setup instructions, which require NVIDIA Container Toolkit.

What it does well

Searches audio and visual content with natural-language queries.
Results include timestamps, excerpts, and relevance scores.
Supports uploads up to 1 GB.
REST API documents search and video-management endpoints.
Licensed under GNU GPLv3.

What to know first

Web uploads are limited to 1 GB.
NVIDIA GPU setup requires NVIDIA Container Toolkit.
Deployment instructions require Docker and Git.

Limits

Web uploads accept MP4, WebM, MKV, MOV, AVI, and OGV files up to 1 GB. NVIDIA acceleration requires an NVIDIA host with the NVIDIA Container Toolkit.

Getting started

Open the project website on GitHub
Install Git and Docker
Choose Docker Compose CPU deployment or the NVIDIA GPU configuration
Prepare local video and output directories with the Qdrant service
Open the web interface or use the REST API
Upload a supported file or provide a yt-dlp-compatible URL
Search with a natural-language query

Verdict

Sift Video offers timestamped semantic search across video audio and visuals, with a web interface and REST API. Its local Docker deployment and setup requirements make it most suitable for users comfortable with Docker and Git.

Questions

Is Sift Video free?

The project is free and open source under the GNU GPLv3 license. It does not list a separate free plan.

Which platforms can run it?

Sift Video provides web and API access and can be self-hosted. Docker Compose instructions cover CPU deployment, with a separate NVIDIA GPU configuration.

Can it search audio and images?

Yes. Searches can cover all matches, audio only, or visual content only.

What can the API do?

Documented REST endpoints cover search, URL downloads, file uploads, listing and streaming videos, deletion, and health status.

What video files can I upload?

The web UI accepts MP4, WebM, MKV, MOV, AVI, and OGV files up to 1 GB.

Facts from Sift Video's own pages

API: The REST API includes search, URL download progress, file upload, video listing and streaming, deletion, and a health endpoint. · github.com, 8 Oct 2026
Audience: The README provides Docker, Git, and optional NVIDIA GPU setup instructions, and invites users to contribute issues or pull requests. · github.com, 7 Oct 2026
Audio and visual search: The search interface supports all-match, audio-only, and visual-only filtering. · github.com, 8 Oct 2026
Deployment: The project runs through Docker Compose and serves its web interface at localhost:8080. · github.com, 8 Oct 2026
File uploads: The web UI accepts MP4, WebM, MKV, MOV, AVI, and OGV files up to 1 GB. · github.com, 7 Oct 2026
GPU support: Separate NVIDIA GPU builds can use CUDA for Whisper transcription and ONNX Runtime inference, with NVIDIA Container Toolkit required on the host. · github.com, 8 Oct 2026
License: The repository states that the project is licensed under GNU GPLv3. · github.com, 8 Oct 2026
Live ingestion: The system accepts yt-dlp-compatible URLs and streams ingestion progress while downloading, extracting frames, transcribing with Whisper, and embedding content. · github.com, 8 Oct 2026
Local data handling: The README describes a local Docker deployment with video and output directories and a Qdrant service; it does not state a security certification or compliance status. · github.com, 7 Oct 2026
Maker: The GitHub profile identifies the maker as Sourav Kumar. · github.com, 8 Oct 2026
Offline indexing: Videos are processed offline, with audio and frames converted into semantic embeddings and indexed in Qdrant. · github.com, 8 Oct 2026
Processing: Videos are processed offline, with audio and frames converted into semantic embeddings and indexed in Qdrant. · github.com, 7 Oct 2026
Purpose: SIFT-Video searches video audio and visual content with natural-language queries and returns relevant timestamps. · github.com, 7 Oct 2026
Search filters: Users can filter searches to audio, visual, or all matches and scope searches to one selected video. · github.com, 7 Oct 2026
Search results: Results show video names, timestamps, match types, transcript excerpts, and relevance scores, and can open the built-in player at the matching moment. · github.com, 8 Oct 2026
Support and contributions: The README welcomes suggestions, fixes, and improvements through issues or pull requests. · github.com, 8 Oct 2026
Technology: The documented stack includes Whisper small.en for timestamped speech transcription and MobileCLIP2-S3-OpenCLIP for text and image embeddings. · github.com, 8 Oct 2026
Upload limits: Local uploads support mp4, webm, mkv, mov, avi, and ogv files up to 1 GB. · github.com, 8 Oct 2026
URL ingestion: The project can download yt-dlp-compatible URLs and stream ingestion progress through server-sent events. · github.com, 7 Oct 2026
What it does: SIFT-Video is a multimodal semantic video search engine that searches video audio and visual content with natural-language queries and returns relevant timestamps. · github.com, 8 Oct 2026

Where Sift Video stands on Freedom251

AI Video Search Tools#23 of 24