01 · BRIEF
Speech to text · en-GBTell your pundit what you want.
Press to record your brief
Who watches, what they should notice, how it should feel.
Pundit watches every five seconds, picks what is worth talking about, researches it, outlines it frame by frame, and your viewers tap it inside YouTube.
Pulled in by video ID, frames and audio.
NVIDIA Cosmos3-Reason watches every 5 second clip: scenes, objects, what is said.
Serverless inference runs the brief, director, research and design agents.
Direct, research and write what people explore, then design the look.
Pixel-accurate masks for every object worth tapping.
One file per video ID.
Choices on the Director, Overlays and Design steps go into the downloaded .pundit file.