Some “art”

This gallery is a collection of graphics made with various AI image generation software. I use a variety of online services, namely MidJourney, DALL-E, Stable Diffusion, NightCafe, and Snowpixel, as well as Google Colab to run VQGAN+CLIP and CLIP-guided Diffusion notebooks. Some of the first images sets (pigeonGAN & butt drawings) were created from models I created with RunwayML.

Eric Decker

is an Engineering Manager at Squarespace and previously the VP of Technology for Firstborn NYC. He lives in New Jersey with his wife, daughter, and too many cats. When he’s not wasting time generating silly images, he enjoys cooking and watching the opossums that visit the backyard at night. His favorite animal is the humble pigeon.

Starting with Style Transfer

I got started with creating AI-based images at the start of 2020 when I discovered NightCafe Studio, which at the time only offered style transfer as an option. I quickly became fascinated with it, and trying to use it for purposes it my have not originally been intended for. Instead of applying painting styles to photos, I gravitated toward textures – using raw meat for The Girl with a Pearl Earring, pie for The Scream, farmland ariels for American Gothic, or classified documents for UFO images.

Text to Image Generation

After a while, NightCafe released the first iteration of their text to image tool. For whatever reason, one of my first prompts was “a sexy muffin dripping with butter” - the result was more disturbing than I could have ever imagined, and needless to say, I was hooked.

Shortly after, I also branched out to running various notebooks on Google Colab. The great part of experimenting with these notebooks was greater control, but the downsize was that image size was constrained due to memory. Some of my first and favorite generations were trying to emulatethe style of various sci-fi artist, namely one of my favorites, John Harris.

The desire for print-quality sized images then lead me to Snowpixel, which was able to generate higher resolution and overall cleaner looking images.