GENREG / REAL-TIME VIDEO GENERATION
GENREG is Payton's live video-generation lab. A scene is composed in real time from small trained entity models: objects, shadows, ground, sky, and clouds each render their own visible contribution.
This is not a recorded clip or a sequence of memorized positions. A declared scene script supplies motion, parenting, time of day, and camera state; the active models generate the image frame by frame. Better models hot-swap into the running scene without stopping it.
Watch an action happen live. Timestamped generation recordings will be published here for validation.
Scripts define where entities are, how they move, what they are parented to, and how the camera observes the scene. The system renders the visible result rather than simulating a game world.
Objects, halos, shadows, terrain, sky, and clouds are separate trainable components. An upgrade to one entity can improve the live output without rebuilding everything.
Live-generation MP4s are timestamped and will be archived on GENREG. Viewers will be able to trigger an action on Twitch, then inspect the corresponding recording to validate what the system generated.
The Live Composer runs continuously while the underlying models improve. Current work is focused on wider camera coverage, clean shadows between objects, smoother edges, higher frame rate, and richer multi-object composition.
GENREG is documenting the scene as it is generated and upgraded, not presenting a polished offline render as a live system. The streamer, controls, entity updates, and recording timeline make the work inspectable while it changes.
Join the Twitch stream to see and influence the scene live. Timestamped recordings will provide a public trail for checking the generation afterward once the archive is published.