Hardware components | ||||||
| × | 1 | ||||
 |
| × | 1 | |||
Software apps and online services | ||||||
| ||||||
 |
| |||||
It all started a few months ago when I built my first version of an AI-powered camera using an ESP32-CAM and a TFT Touch screen. The idea was simple but exciting: capture an image, ask a question about it, and get a response from GPT. While the concept worked, the reality wasn’t as smooth — the hardware was underpowered, the memory was limited, and the whole setup often crashed or froze. I knew I could do better.
Then one day, a package arrived from Makerfabs — the MaTouch ESP32-S3 2.8” Display with Camera. It had everything I needed: a better ESP32 chip (with more RAM and USB support), a sharper camera, a touchscreen, and even a microSD slot. As soon as I held it in my hands, I thought: “This could be the perfect brain and body for a new version of my AI camera.”
And so began the journey of building AI Camera V2.
I decided to take a modular approach. First, get the camera live stream working, then make it capture images on touch, and finally store them to the SD card. This time, everything worked smoother than expected. Seeing the real-time camera feed on the touchscreen was already a joy!
Next came the keyboard — because how else could I ask GPT questions? I tried using LVGL for UI, but it just didn’t play well with the camera stream. SquareLine Studio didn’t support dynamic image widgets either. After hours of frustration and digging through forums, I realized I’d have to leave LVGL behind and go full Arduino GFX — and that meant building my own keyboard from scratch.
And that’s what I did.
Drawing keys manually, mapping touch inputs, handling character input — it was a challenge, but a rewarding one. For the first time, I had a fully custom keyboard working on Arduino GFX.
Once I had image capture and text input working, it was time to bring in the real magic: GPT-4o. I converted the image to base64, added the typed prompt, and sent it to the OpenAI API. The first time I saw GPT’s response appear on the screen — accurately answering a question about the image I just clicked — I literally grinned with excitement. That’s the moment you live for as a maker.
To make it look sleek and user-friendly, I designed a 3D-printed case. Now the device looked less like a prototype and more like a mini AI-powered smartphone. Neat, clean, and compact. But more importantly — stable.
Unlike the first version, this one didn’t crash. It didn’t lag. And it delivered what it promised: a small, standalone AI camera capable of understanding what it sees. Here is the video that demonstrate the working of the project. You can start the video from 16:22 min to watch project testing part only. You can also watch the complete video if you want to see how we build it.
What’s Next?I didn’t just build this for myself. I wanted students, makers, and educators to try it out too. That’s why I’m offering a ready-to-use kit on my website — pre-programmed, cased, and tested. Just plug it in, enter your Wi-Fi and OpenAI key, and you’re ready to explore AI on the edge.
Purchase link - https://techiesms.com/product/ai-camera-v2-your-personal-gpt-4o-powered-smart-camera/
This project is proof that when you iterate, learn from your past build, and keep pushing boundaries — amazing things happen.
And I’m already thinking about AI Camera V3 😉But that’s a story for another day.
Comments