Clef-omni, watching
A decision model watches your camera in 4-second video clips and answers nine fixed questions, each with a probability.
Camera off. Start it, or pick a photo.
Your camera records one 4-second clip at a time, video only. Not stored. Stops after about 2 minutes.
- clips
- 0
- last call
- —
- tokens
- 0
- cost
- $0.000000
What
Clef-omni is a decision model, so it doesn't write text. It answers fixed questions about a video clip, each with a probability, in a fraction of a second. Here it watches your camera.
How
- model
- Clef-omni
- id
@cf/cloudflare/clef-omni- runs on
- Cloudflare Workers AI, through this site's Worker
- costs
- about $0.0002 a clip
- Your browser records a 4-second clip, video only.
- This site's Worker sends it to Clef-omni with nine standing questions. Clef-omni samples the clip at 2 frames a second.
- Each answer comes back with a probability. A bar lights up at 0.6 or more.
Why
A camera sends 30 frames a second, and no one can afford a language model call on every frame. A decision model can ask "what changed?" across whole clips of video, and call the big model only when it matters. That is the first layer of the router in Inference That Knows Where to Run. See it place work in Where should it run?
Try
- Start the camera, then wave, smile or hold something up; "something is moving" climbs.
- Turn away, or cover the lens, and watch "a person in frame" drop.
- No camera? Pick a photo of a person; it answers the eight picture questions.
Limits
- One 4-second clip at a time here, to keep the cost low. A real watcher would read every frame on the device.
- The questions are fixed. You cannot ask your own.
- A probability is not a fact. Dim light lowers it.
Write to me about the problem you want to fix.
- emailEmail contact@meirdick.com
- linkedinLinkedIn in/mdick
- githubGitHub meirdick
- substackSubstack aistrategyconsultant
- rssRSS /rss.xml