FEED

Concept

Social media augments reality not through a headset but through the feed. Bot-generated posts steer public conversation at scale: manufacturing consensus, inflating outrage, flattening nuance into engagement metrics. The augmentation is already here. You just don’t notice it.

FEED makes this augmentation literal and visible. A live camera captures the viewer. AI-generated tweets produced in real time from whatever topics are actually trending scroll across the frame, accumulating like grime over the camera image. Post by post, the feed buries the real person underneath. Each synthetic voice is designed to pass as human: shitposters, concerned parents, bots, pundits. The content is fabricated. The topics are real. This is how narrative gets bent, not through a single lie, but through a thousand plausible voices saying the same thing until it becomes the room tone of reality.

How It Works

FEED is a real-time browser-based installation using cam feed, face detection, AI-generated text, and generative audio. In a gallery setting, it runs fullscreen on a single screen with a camera feed.

Visuals — A live webcam feed is pixelated into a mosaic grid, evoking raw surveillance footage. AI-generated tweets scroll across this grid, gradually painting over the camera image in layers of dark, accumulating strokes. When a face is detected, the paint erodes around it, exposing the pixelated video beneath. Detected faces trigger a surveillance overlay: glitchy color-tinted tiles, bounding box crosshairs, and “SUBJECT IDENTIFIED” labels.

Audio — The tweet grid itself is the musical score: a sequencer reads it column by column, mapping cell colors to notes in a pentatonic scale and rows to octaves. With no faces present, the piece is silent. Each detected face unlocks another row of the chord and increases the tempo, so the machine grows louder and faster as more people are watched. Blips and thuds punctuate each new tweet entering the grid, giving the production line its mechanical pulse.

Content — Tweets are generated live by an AI language model prompted with real trending topics from Google Trends. The voices are varied and realistic: shitposters, bots, concerned parents, industry insiders. The content refreshes continuously, so the feed always reflects what the internet is actually talking about.

Video