TunnelVision is an AI filmmaking system for directing continuous first-person journeys through imagined worlds. It is not a prompt wrapper around a video model. The filmmaker specifies a journey; a Director structures the destinations; a Cinematographer figures out how to physically shoot each traversal; Camotion prepares motion-conditioned endpoint frames; and a video model films the camera travel between canonical viewpoints.
The concept originated with filmmaker Terran Boylan, who used the original TunnelVision technique on his YouTube channel to create long continuous POV simulations—selecting successive stills so a first-person camera appears to move through a coherent environment. That technique was later adapted for the Peter Gabriel Digging in the Dirt 50:50.dev project, where we used the playful name TunnelTV. This research asks what happens if that process becomes self-directing.
It starts the way a film starts: the filmmaker imagines a story.
A red paper airplane escapes from a cell at Alcatraz, flying out through the wide steel bars of the cell window and into the open air over the bay. It heads toward the Golden Gate Bridge, but a strong gust of wind catches it and sends it banking back toward San Francisco. It crosses the bay and makes its way into Chinatown, then turns onto California Street and climbs the hill alongside the cable cars. Near the top it turns toward Nob Hill and The Masonic, approaching its monumental entrance with huge concrete columns framing the tall glass facade. The airplane descends toward the entrance and finally lands on the broad steps outside.
That is a story, not a generation prompt. The Screenwriter expands the idea into a full story. The Director breaks that story into destinations: the places the camera will actually pass through, and builds the storyboard. The Cinematographer looks at each pair and plans the move—keep following the plane, never stop, never look back at the door you just used. Camotion readies the frames for travel, then the camera rolls. If a take doesn’t hold the world, or the plane isn’t still ahead, it gets reshot. The selected takes, laid end to end, are the movie. The working loop is SPECIFY → DIRECT → CONSTRUCT → SHOOT.
TunnelVision was submitted to the Runway Hackathon 2026 and is listed on Runway Diffuse. The film below is the official submission: how a hand-built continuous-shot technique became a self-directing pipeline that plans destinations, chooses how to shoot them, routes generation through Runway, and assembles a finished journey.
Visit TunnelVision Site View Code View on Diffuse Diffuse Profile
