You're watching a reel. A movie plays in the background. You don't know the name. The creator says "comment and I'll DM you." You comment. They never reply.
That exact moment made us build YouTube Spotter: paste any YouTube link, get the movie, series, anime, game or tool that appears on screen but was never named.
The real problem was cost
The obvious build: have a multimodal AI model watch the video and identify what it sees. It works — but the capable model cost enough per video that a free hobby tool would bleed money at any real usage.
The fix wasn't a bigger model
Viewers almost always name the thing in the comments. So instead of making an expensive model watch the whole video, we feed the top comments to the cheapest available model — it only has to confirm what the crowd already hinted at.
The cost dropped by roughly 95%. Same user experience, a fraction of the spend.
Why we're telling you this
This is the difference between "we added AI" and "we engineered an AI system." When we build AI features for clients, this cost-thinking is built in from day one — because an AI feature that's too expensive to run isn't a feature, it's a liability.
