Inspiration
We've all misplaced something and spent ages trying to remember where we left it. Sometimes the answer is sitting in hours of security camera footage, but finding that one moment means manually searching through recordings.
That's where the idea for Rewind came from.
What if you could simply ask, "Where did I leave my glasses?" and your cameras could help you find the answer?
What it does
Rewind makes your camera recordings searchable using natural language.
Instead of manually scrolling through hours of footage, you can ask questions like:
- "Where did I leave my keys?"
- "What happened to the package outside?"
- "Find the footage of someone entering the garage."
- "Show me what happened before this recording."
Rewind is designed to identify relevant moments, locate where objects were last seen, and reconstruct sequences of events across recordings.
It's useful for everyday situations, home security, and even finding specific scenes in large video archives.
How we're building it
Rewind is being designed as an AI Skill for UGREEN NAS.
The goal is to combine the NAS's existing storage and video-search capabilities with AI-powered analysis.
Rather than processing every second of footage through a large AI model, Rewind uses a more efficient approach: narrowing down relevant recordings first, then examining promising moments in greater detail.
Results are intended to include timestamps and links to the original recordings, allowing users to verify what the system found.
Challenges
The biggest challenge is balancing accuracy with efficiency.
Searching for a small object, such as a pair of glasses, across several hours of footage is considerably harder than identifying a person or a large object.
Another challenge is distinguishing where an object was last observed from where it actually is. Rewind needs to communicate uncertainty instead of presenting guesses as facts.
We're also working within the capabilities and restrictions of UGREEN's AI Skill system, which makes efficient use of existing NAS tools especially important.
What we learned
One important design insight is that finding an object isn't necessarily the same as finding footage containing it.
To answer "Where did I leave my glasses?", the system needs to understand the sequence of events, not simply recognize the glasses in an isolated image.
That distinction has shaped Rewind into something broader than a video search tool: a way to search and reconstruct moments from the physical world.
Built With
- ai
- markdown
- nas
- pro
- skills
- ugos
- ugreen