R&D project

Live vs. Dead

Some moments on camera feel alive. Others feel staged, flat, or dead on arrival — even when nothing is technically wrong with them. We're studying what actually separates the two.

Our working idea: liveliness isn't about being unscripted, and it isn't about rigid control either. It shows up when someone stays genuinely open to how a moment unfolds while still operating inside real structure — a scene, a rule, a constraint they can't talk their way out of. Constraint without that openness reads as dead. Openness without any constraint reads as noise.

Right now this is early-stage research: we're collecting human judgments on real clips to see whether people even agree on what "alive" looks like before we try to teach a model to recognize it. If that agreement holds up, the next phase is building an automated pipeline. That doesn't exist yet — this page is the public front for the project while the groundwork gets laid.

How the pipeline works

  1. 1. MonitorFinds public videos and adds them to a review queue.
  2. 2. PopularityFilters out clips nobody watches before the costly step. It never judges live versus dead.
  3. 3. Feature detectorScores each clip against every feature in the taxonomy.
  4. 4. Feature discoveryProposes new features from clips the current taxonomy explains poorly, for a human to approve.