Recent work
I set out to create an endless pixel art story using ChatGPT and Stable Diffusion. The result would be streamed live to Twitch. Stable Diffusion would be used to generate the graphics and ChatGPT for the story script. The concept was a Stan the cabby driving all over New York to real street intersections and picking up passengers and having conversations with them. Through conversations an evolving story is told about an apocalyptic future New York City.
I love pixel art and I got massively excited about how to generate pixel art with stable diffusion. At the time of starting the project in 2023, Midjourney and Dal-E were not good with pixel art. The results would imitate the style, but were not pixel perfect "true" pixel art. I discovered a brilliant model for Stable Diffusion called Retro Diffusion which is trained on the work of consenting pixel artists and is capable of really impressive pixel art images.
With the announcement and massive hypetrain explosion of the Rabbit R1 I wondered, could I make a fully portable functioning ChatGPT powered voice assistant? I found the perfect hardware for this project in the M5 StickCPlus2 - it's tiny, has a built in 200mAh battery, Wifi, Bluetooth, a built in Mic and Speaker (the speaker is not really documented but I got it to work). The best bit is that it's the size of a pack of gum!
More info soon about how I went about building this including links to the code and how you can build your very own mobile personal voice assistant like Rupert.
I built and operated the first drone for BBC News as well as working to standardise drone operations across the organisation. Our cutting edge work was documented by the New York Times. , Wired Magazine and The Washinton Post.
