Bjørn Karmann has developed a camera that takes “photos” – the quotation marks are important. In fact, the invention of the Danish designer breaks with the common definition of photography as the depiction of reality using an optical system. Karmann’s camera, called Paragraphica, has no optics whatsoever, i.e. no lens. That’s why she doesn’t take photos, but generates content – with the help of artificial intelligence.
Paragraphica collects GPS and weather data
Strictly speaking, Paragraphica is a mini-PC in a 3D-printed housing, the striking front of which is based on the tactile organ of a mole. The basis is a Raspberry Pi connected to a screen. The system uses GPS and a weather app to collect various data about the current location and events: what day and what time is it, what is the current weather like and what shops or special features are in the immediate vicinity?
Based on this information, Paragraphica creates a corresponding description. For example: “An evening photo taken at Karl-Wiechert-Allee 10 in Hanover. The weather is sunny with 25 degrees and it is Thursday, June 1, 2023. There is a university and a hotel nearby.”
This description is the substitute for what the lens actually captures on a normal camera. It is also the prompt for the text-image generator Stable Diffusion. Paragraphica connects to its API via a mobile Internet connection and thus presses the virtual trigger: Stable Diffusion begins to generate a corresponding image based on the description, which Karmann then displays on the screen, like with a digital camera.
Paragraphica hallucinates the world together
In order to simulate the feel of a traditional camera, there are three wheels on the top of the housing that can be used to adjust the prompts: This allows you to set the radius around the location in which data should be included in the creation and how how strongly the algorithms should adhere to the specifications of the description, i.e. how strong the focus – in a figurative sense – should be.
And the results? Of course, the “photos” that Paragraphica hallucinates sometimes have little to do with reality, even if that ultimately always depends on the actual location. The algorithms of AI image generators like Stable Diffusion, DALL-E and Midjourney are trained on billions of publicly available images on the web. In places where there is a lot of footage, the algorithm is able to create an image that is actually quite close to reality. In lesser-known and remote places, however, the AI has to “imagine” what it might look like there.
In some cases, the generated photos actually show the world as it is around him, Bjørn Karmann writes on his website: “Interestingly, the photos capture some moods and emotions of the places, albeit in an eerie way.”
He explains on Twitterthat Paragraphica is a personal art project. He neither wants to make a salable product out of the camera, nor is it about questioning traditional photography in any way. Rather, he wants to show how artificial intelligence interprets the physical world. By the way, if you want to try out Paragraphica, you don’t have to build the camera first: Karmann has also developed a version for the browser (which is currently heavily overloaded).
(jl)