Collaborative video-annotation cataloging prototype

Reliable descriptions for moving images.

AI drafts visual descriptions for silent amateur films; people review and correct them into IIIF descriptive metadata, then use it for video search.

Create

Extract shots; Claude vision describes each scene and reads on-screen text and intertitles.

Verify

AI output is a draft. People add local knowledge (places, names) and confirm — only then is it catalog data.

Use

Search scenes by description, on-screen text, or tags and jump into the video. Limit to reviewed data.

Create → Verify → Use

Film→AI + OCR→IIIF anno.→Review→Search

Human-verified data (reviewed = true) becomes training data that improves the next round of automatic recognition.