The Paragraphica is the weirdest AI camera so far

On the topic of weird AI cameras, the Paragraphica has to be the weirdest I’ve seen so far. Like the Black Box Camera we featured a couple of months ago, this one creates an image using AI. Unlike the Black Box Camera, which takes a picture of the scene, uses AI to determine what’s in the scene and then uses that to generate a prompt for the AI to make an image, this one uses GPS data.

There’s no lens at all on this “camera”. It figures out where you are based on the GPS data. From there, it determines the time of day, weather conditions and what your surroundings look like. This generates the prompt that the Paragraphica uses to create its output image. Not surprisingly, it’s powered by a Raspberry Pi.

Introducing – Paragraphica! 📡📷
A camera that takes photos using location data. It describes the place you are at and then converts it into an AI-generated "photo".

See more here: https://t.co/Oh2BZuhRcf
or try to take your own photo here: https://t.co/w9UFjckiF2 pic.twitter.com/23kR2QGzpa

— Bjørn Karmann (@BjoernKarmann) May 30, 2023

Paragraphica was created by Bjørn Karmann, who describes it as a “context-to-image camera”. It uses location data based on where the user (Bjørn) is standing with the camera along with some AI magic to generate a “photo” of that place at that moment. Obviously, however, it can’t actually “see” anything. The image it creates is entirely computer-generated.

On the front of the camera, where we would normally see a lens, we instead see a strange-looking bright red object. This design was inspired by the snout of the star-nosed mole. It’s a metaphor for the project as a whole. On the Paragraphica, it’s simply decorative, serving no practical function, but for the mole, it’s a different story. The star-nosed mole lives and hunts underground, where there is no light, using its snout to find its way and build a picture of its environment.

The camera is based on the Raspberry Pi 4 and features several knobs and buttons on top along with a touchscreen LCD on the back. The display on the back shows you the current prompt that will be used to generate the image. The three dials on top of the Paragraphica let you control the data and AI parameters to adjust the appearance of the image it generates. Once you hit the red “shutter” button, an AI-generated image appears on the display a short time later.

GPS data from the location where it’s being used allows the camera to find information to generate the prompt. Starting, obviously, with the location. It then determines other factors, such as the current weather conditions, the time of day and any local points of interest. All of these are then included in a short paragraph that becomes the prompt.

You can try it for yourself, virtually

The camera was built using some custom Python scripts created specifically for this project, along with the Stable Diffusion API to create the final image. Noodl was used to create an app that communicates between the camera and multiple APIs to generate both the location-based prompt and the final image itself.

It appears that the camera itself is only a one-off, though. Its code and schematics haven’t been released and it doesn’t look like Bjørn plans to put it up for sale. It’s his own, solely unique, camera. But don’t worry. You can still play with it virtually. Bjørn has produced an online web-based version of the camera for us all to play with and see what it would generate in our local surroundings.

It appears, though, that recent media exposure has caused some issues to the virtual camera.

The server has officially crashed! 🔥

There are a lot of people trying to generate photos at the same time, so in case it won't load or generate a photo atm this is why 😅 pic.twitter.com/SfMZvSmIKi

— Bjørn Karmann (@BjoernKarmann) May 30, 2023

Bjørn has created a mirror of the virtual camera on another host that appears to be up and running, although I’ve yet to see an image from it at my location. Perhaps you’ll have better luck! He also has an excellent write-up of the project on his website. It’s well worth a read if you want to learn more about it or even figure out how to make one of your own.

[via Raspberry Pi]