A Generative Timelapse of the Sky

For this Human-Computer Interaction research project, I explored how real-time data could generate a visual output. I wanted the project to become both a visual experiment and a way to deepen my understanding of programming.

I chose to work in Python and Processing, starting from a simple question: how can live input create an image over time?

At the beginning, I explored several directions, including neural networks, object recognition, webcam data, weather prediction, and Arduino-based sensors. Neural networks quickly became too complex for the time available, so I shifted towards a more focused webcam-based system.

The first experiments used a grid of squares filling the screen from left to right. At first, the colour of each square was controlled by the mouse. Later, I replaced that input with a seconds counter and random values to understand how time-based data could gradually build an image.

The next step was to connect the program to a webcam. After experimenting with video libraries and OpenCV, I displayed the webcam image inside the program and used the colour value of a central pixel as input.

That pixel became the source for a scanning line that moved across the canvas. Each new line translated what the webcam saw into a strip of colour, turning live sky data into a growing image.

As the project developed, I separated the webcam view from the generated output so the system could collect colour data cleanly. The scanner added thin lines from left to right, each one based on the light and colour of the sky at that moment.

After feedback, I changed the structure of the program. I placed the video image above the scanner, added automatic image exports every hour in case the program crashed, and replaced hardcoded values with variables.

I then tested the installation outdoors using a small LUCA webcam placed in my garden. The webcam was protected from rain but still exposed to changing light, clouds, wind, and atmosphere.

Further improvements made the scanner more accurate. Instead of stretching a single pixel, the program scanned a wider strip of the webcam image, creating a higher-quality result. I also added a progress indicator and increased the capture frequency to better record movement in the sky.

After 22 days of collecting data, the final result became a generative timelapse of the sky. Each image was built from live webcam input, turning small changes in light, colour, and weather into a visual record of time passing.