Hallucination refers here to the phenomenon in which an AI or generative model produces an image, place, or object that appears familiar and convincing, even though it does not exist in reality. In the provided context, a model trained in part on photographs taken by the artist in Los Angeles and Berlin generates still images extracted from synthesized video. What emerges is described as something “hallucinated” from archival records of those two cities: a hybrid location that feels recognizable but cannot be found anywhere. Hallucination is therefore not simply an error of perception, but a mode of image production in which the gap between image and reality becomes visible.
The context also links this phenomenon to the bodily limits of AI perception. Because AI does not possess human physical experience, its relation to movement, sound, and viewpoint may differ from human sensory framing. As a result, the position of the observer can become unstable or blurred, and the work suggests a kind of disembodied or drifting perspective. In this sense, hallucination names a condition in which generative systems recombine traces of the real into something that looks true while remaining unreal. It is a key term for understanding how AI-generated images can feel evidential, archival, and fictional at the same time.