Run your own video
Start with the included Colab notebook or install locally with Python and FFmpeg. Generated images and enrichment use optional paid model stages.
Open the notebook in Colab ↗A video-to-wardrobe pipeline
Every garment in your closet, catalogued from one video
ClosetScan extracts garments from a phone video, groups their views, and adds generated flat-lays, attributes, and narration. Explore a real run below.
From recording to structured data
- 01Capture & extract
Video → candidate garment frames
- 02Group & enrich
Distinct pieces, front/back images, attributes
- 03Browse & query
Interactive catalogue + local MCP server
Python pipeline · Colab notebook · Agent skill
Interactive example
One recording. 34 garments.
Precomputed results · No API key needed to explore
Input Phone walkthrough
5:49Select a garment, then a timestamp to jump to its source in the recording.
68candidate rows
→34grouped garments
These results were generated from this recording using DINOv2 and the paid AI stages.
Output Generated catalogue
Ordered by first appearance in the video.
Flat-lays are AI-generated reconstructions and may invent details. Open a garment to inspect its attributes, narration, and original frames.
Use the project
Take it beyond the demo.
Process your own recording, inspect the outputs, or give an agent access to the catalogue.
Connect an agent
The local MCP server exposes garment search, details, narration, and wear logs. The included wardrobe skill shows agents how to use it.
View MCP configuration ↗Inspect. Adapt. Build.
Read the setup instructions, pipeline stages, and known limits. ClosetScan is MIT licensed; contributions are welcome.
Read the project README ↗