SceneSense AI, Video Metadata Junior Editor
A lightweight tool that extracts structured scene metadata (shot type, camera motion, mood, lighting, audio) from uploaded video without GPU-heavy processing, built and shipped in 4 days.
- Base44
- React
- HuggingFace API
- Supabase
- Tailwind
Video editors spend hours manually tagging footage, a repetitive task that slows creativity. SceneSenseAI generates structured scene metadata from uploaded video using a 5-prompt cinematic-analysis workflow. Designed to slot into Premiere, Frame.io, and Firefly Video workflows.
Problem
Video editors spend hours manually tagging footage. There is no fast, lightweight tool that auto-generates useful metadata from videos without GPU-heavy local processing.
Objective
Create a lightweight AI tool that extracts scene metadata without manual tagging or heavy ML infrastructure, showing how LLMs can assist editorial workflows at the seams, not at the timeline.
How it was built
Base44 for rapid development with built-in hosting and analytics. A 5-prompt workflow that simulates cinematic analysis. HuggingFace API for structured metadata generation. A custom JSON schema keeps output consistent enough to be useful downstream.
Supabase for uploads, React and TailwindCSS for the interface, a progress UI with explicit upload, processing, result, and retry states.
Output schema
Every clip is tagged with shot type, camera motion, scene mood, lighting style, subject action, audio type, and a one-sentence metadata summary. The schema is the product, not the UI, because that is what plugs into the next tool in the chain.
Where it lands
Demonstrates GenAI video editing assistance without heavy ML. Saves up to 35% of footage organization time on real cuts. Designed to integrate with Adobe Premiere, Frame.io, and Firefly Video.