Techniques architect Simon Aubury has developed a Raspberry Pi-powered digital camera with a special: it does not seize what you level it at, however a recreation filtered by two generative synthetic intelligence techniques.
“Generative AI (GenAI) is a sort of Synthetic Intelligence that may create all kinds of photos, video and textual content,” Aubury explains by means of background. “To speed up the robotic rebellion I chained two GenAI fashions collectively to construct a digital camera which describes the present scene in phrases, after which makes use of a second mannequin to create a brand new generated stylized picture. Let me introduce GenPiCam — a RaspberryPi primarily based digital camera that re-imagines the world with GenAI.”
When it comes to {hardware}, the undertaking is fairly easy: A Raspberry Pi 4 Mannequin B single-board laptop is positioned in a customized housing with a show on the rear and a push-button swap wired into its general-purpose enter/output (GPIO) port. A Raspberry Pi Digital camera Module, situated on the entrance, captures a nonetheless each time the button is pushed — however somewhat than merely saving the ensuing picture, begins a strategy of AI-based transformation.
In step one, the captured photograph is distributed by Midjourney’s Describe AI system — which inspects a picture after which generates a textual description which can or could not precisely match the contents of the image. This description is then fed again by a second Midjourney AI, Think about, which turns the outline again into a picture — however one which has by no means existed in the true world.
The digital camera features a “filter” management which provides model directions to the textual content immediate. (📷: Simon Aubury)
To supply a point of inventive management, Aubury added a rotary swap which permits for a spread of kinds to be included within the image-generation immediate — from “pop artwork” to “anime.” A Python program operating on the Raspberry Pi then takes the generated picture and collages it with the unique seize and the textual content immediate it impressed.
“I had a lot enjoyable constructing the GenPiCam digital camera — and this was an fascinating path for exploring immediate engineering for Generative AI. The higher pictures have been those which had a easy composition — basically photos that have been straightforward to place phrases to. The GenPiCam has been a enjoyable option to discover Generative AI, reworking pictures into stylized (and someday shocking) photos.”
Aubury’s full undertaking write-up is offered on Medium.
