Between 2021 and 2022, I avidly tested every image-generating algorithm that would fall in my hands. This is a small selection of everything I asked to generate during that time. Note that most of them are edited through Photoshop, Affinity Photo and Gigapixel apps, there aren't any raw outputs.
At the beginning there was Dall-e (ancestor of the DAll•e algorithm that we know today) and Dall-e was... weird. It didn't produce anything closely related to what you asked it to generate. But it gave many original looking pictures though, these are some upscaled generations from that era. They look like plastic-organic hybrids from a biopunk parallel universe. Cool.
As these generations were pretty rough and difficult to work with, here are some more compositions that I created from its outputs through mirroring and Photoshop editing. My objective has always been creating something more complete and closer to my vision than the algorithm is capable of generating.
Eventually, I figured out how I could mix different pictures into one, here is an imaginary record cover made with two overlaid images (by combining DALL-E Pytorch, Real-ESRGAN and Photoshop). Swimming in the troubled waters of love...
Later, I discovered Disco Diffusion, which worked through Google Collab, and which allowed for a greater creative freedom. These two, that I called "entomology grotto" and "lonely mother" are the generations from that era that started to show me the capabilities of this medium.
Around that time, one of the questions that started to arise for me was how these images could be used to tell stories. These pictures represent an industrial accident provoked by an imaginary company called Deep Sea Kawaii Industries.
In 2022 I continued using Disco Diffusion and I started getting better at writing prompts, here is an abandoned clown hospital (I used references from several war photographers to create its grim look).
Disco Diffusion was also getting more detailed and better at producing complex outputs, here are some colorful worms partying in an abyss.
Another algorithm from back then that could be used to generate realistic landscapes was Gaugan2, here with loads of Photoshop editions to make the picture more lifelike.
Back to Disco Diffusion, I liked to ask it to generate "difficult" prompts, like this landscape that mixes dystopia (barren arid land, polluted and depleted of all of its resources) and utopia (lush green areas thriving with life and water), between aerial photography and macro photography.
Just to finish, an illustration generated from several outputs from Disco Diffusion mashed together through a technique called "photo bashing" (yes, seriously). There are even some of my photos in the mix, while some elements that were painted by hand. This is what I think is better to do with Ai-generated images, instead of just using the raw outputs you can combine them into something more particular and unique. Anyway, here's a glimpse of everything ai-image-related that I did during those two years.
🡠 Back ⌂