Thoughts on Diffuji
Recently a few friends and I made Diffuji, an experimental instant camera that combines the physicality of a polaroid with the dream-like reality that diffusion models can imagine. While a toy project, our camera has been incredibly fun to use and share with others.
Diffuji was built with a custom 3D printed chassis to house a Raspi Nano 2w, an Arducam (our camera), and cheap TTL thermal printer. We utilize both on-device style transfer models and more powerful API-based image generation models to quickly process images so they can be instantly printed and shared. The camera is quite simple, and is only composed of a shutter button, mode dial (rotary encoder), a power switch, and a small LCD display. It has pretty good battery life and is powered on two 18650 batteries with a 3A 5V UPS supply. Yes, in its current state it is quite chunky, but we're looking into significantly compacting the body to make it reasonably small and portable.
Normally, I am pretty averse to "AI slop", but there is this strange attachment I feel for the images Diffuji can produce. I can hold the captured moment in my hand, it feels real, but it is this perverse artifact of reality that manifests in so many intriguing and unanticipated ways. I can't help but find myself waiting in eager anticipation as the thermal printer slowly cranks as the next printed image reveals itself, as if I were watching a polaroid print slowly develop its final image into view. It doesn't try to be authentic, but rather tries to be funny, and at that it succeeds quite well.
Diffuji has been a consistent conversation starter with anyone who meets it and has since populated the walls of our (now retired) college apartment with reels upon reels of 17th century style selfies of friends and guests. Using it more, it is clear that there is something special about enshrining a memory through physically sharable media, and further its value doesn't come from the realism of the image, but rather, the memory of taking it.
Diffuji ended up winning the Neo Prize and Most Creative Prize at TreeHacks 2026. If you're interested in learning more about the hardware and software behind Diffuji, check out our Devpost!
This project was built with some amazing people: Nathan, Pranav, and Lainey.