It was probably only a matter of time before the first AI-only camera appeared on the scene. Curiously, however, it does not come from one of the established camera manufacturers but from Bjørn Karmann, a Danish experience and interaction designer who has combined Python programming, a Raspberry Pi 4 and the Stable Diffusion API in a 3D-printed camera housing.
The Paragraphica camera collects data from its location via open APIs. It uses address, weather, time of day and nearby locations. From all these data points, Paragraphica compiles a paragraph (text prompt) detailing the current location and moment.
Using a text-image AI (stable diffusion), the camera then converts this paragraph into a "photo".
According to Karmann, the resulting "photo" should then be "not just a snapshot, but a complex and nuanced reflection of where you are, and perhaps how the AI model &sees& that place".
The photos "thus also capture moods and emotions that are reminiscent of the place, but in an uncanny way, because the photos never look exactly like the place I&m in" - Karmann continues.
On his website, Karmann shows in some detail how his Paragraphica AI camera works: For example, this schematic illustration makes it easy to see what data goes into creating the images and how the placeholders of the AI prompt are populated with environmental data:
Bjørn Karmann refers here primarily to the exciting artistic context of the images - but Karmann&s AI workflow can also be read as a reference to the future use of AI-generated images: To the "real" photo or video that one takes of a certain motif at a certain place, the "image creator" then always receives an AI-rendered option in addition. Which one is then used as the "image document of the place" is decided by the camera user.
In fact, this rather interesting experiment only demonstrates a little more clearly what has long been reality: all our digitally created image worlds of "reality" also contain artificially added elements. "AI" has therefore been part of our visual worlds for much longer than the current hype about stable diffusion and the like would suggest.
By the way, if you want to try out Paragraphica yourself, you can do so completely automatically on this Karmann&s website ...