Gazing into the Loop

• 

  Gazing into the loop_cover

Inspiration

The project began with a curiosity about what an AI system “sees” when it generates video. Instead of treating AI as a tool, we wanted to use the video medium itself to reveal the hidden mechanics behind prediction, repetition, and human projection. The goal was to let the machine’s internal rhythm become visible on screen.

What it does

The work is an experimental AI-generated video that reimagines token prediction as visual motion. Simple human actions are translated into repeating loops, while fragmented characters appear and disappear as if assembling meaning frame by frame. The video acts as a window into the machine’s internal process, not as a narrative but as a sensory experience.

How we built it

We used generative video models to create looped sequences and paired them with a visual language that removes text completely. Through iterative prompting and refinement, we shaped a video that mirrors the mechanical logic described in the research. Every motion and transition is designed to echo how tokens are selected, repeated, or abandoned inside the model.

Challenges we ran into

The biggest challenge was letting the medium speak for itself. Generative models tend to mirror human aesthetics, so removing unintended emotion or storytelling required many iterations. We also had to find a form that remained engaging while staying true to the machine’s perspective.

Accomplishments that we're proud of

We created a video that does not explain AI but lets viewers feel its structure. The final loop maintains both visual simplicity and conceptual depth, and the medium of AI video becomes part of the message rather than a separate layer of technology.

What we learned

Working directly with AI video exposed how much human intention is embedded in every model output. Even when we try to minimize intervention, the system reflects our patterns and assumptions. This reinforced the idea that human-machine interaction is always co-constructed.

What's next for Gazing into the Loop

We plan to extend the video into an interactive format where real-time inputs modify the loop’s behavior. We are also exploring multi-channel installations that allow the AI’s predictive rhythm to unfold across several screens, creating an immersive environment built entirely from machine-generated motion.

Built With

• python
• veo2