Interactive installation in Adaptér creative technology knowledge center
Interactive installation in Adaptér creative technology knowledge center
Interactive installation in Adaptér creative technology knowledge center
2025
2025
2025
Fragmentation is an interactive installation created for the exhibition Machines as humans as machines... at the Adaptér Creative Technology Knowledge Center in Budapest. The exhibition explores innovative and experimental applications of artificial intelligence across various fields.
Fragmentation demonstrates how AI is trying to imitate humans – in this case, literally shaping itself in the form of the person viewing it.
Development
For this installation I built a particle system consisting of simple geometric shapes for a large 4K portrait display. The visitors are tracked by a depth camera, which is using AI to extract the human silhouette in front of the screen. The particles are always flowing in the direction of the silhouette, influenced by different additional forces to simulate a more organic motion.
The continuous flowing movement towards the silhouette is tricky – for a still image I would have created an SDF flow map without the need for immediate texture generation. However for processing a video capture stream that is updating at ~30 fps, I needed GPU calculations in real time. Which is why I used Unreal's Niagara again, but this time for the computations only.
I used the Jump Flooding Alghorithm (JFA) transcribed into Niagara's custom HLSL nodes and modules. With this method, an approximated SDF (signed distance field) can be generated within milliseconds, which is exactly what I needed to continuously produce the updated flowmap for the incoming stream.
This flowmap tells each particle which direction to move in order to reach the outlines of the silhouette. This technique opened up a wide range of creative possibilities – so naturally, I could not resist experimenting with various visual effects.
When no visitor is nearby, the particles organize themselves into a grid that is essentially the 3D pixelated version of the camera feed. Check out the animations for that:
