Dan DeGeest
ABOUT
TunnelVision is an agentic generative video system for creating continuous cinematic journeys through changing environments. It grew out of the manual workflows developed for Peter Gabriel’s Digging in the Dirt, automating and expanding the process into a system that plans journeys, generates shots across multiple models, evaluates the results, and adapts as it works.
SELECTED FRAMES — 8 OF 8
PROCESS
Built with Cursor around Runway Dev and multiple image and video model backends. TunnelVision represents a journey as a sequence of pristine canonical frames, then derives camera-motion-conditioned shooting frames for each traversal. An agent plans the route, generates each shot, and uses a separate evaluation layer to score set consistency and traversal confidence. When a shot falls below target thresholds, the system can adjust its approach and reshoot rather than simply accepting the generation. Successful outputs are carried forward into subsequent shots, creating a feedback loop that maintains continuity while still allowing the journey to evolve. The system grew from the largely manual process used on Digging in the Dirt, turning those techniques into a repeatable, adaptive generative-video pipeline.
Credits
More Experimental Work
Stupid 2026TV Episode #2
Demon Ego
Stupid 2026TV Episode #1
Demon Ego
Norway, as a Product
John-Ivar Eidsmo