Intro

Soon after OpenAI‘s DALL-E image generator was released, I found myself fiddling with prompts. Even during meetings, when someone said something that caught my ear or was “quotable”, I turned it into a prompt and allowed myself to be amused by the result.

I am an avid amateur photographer but I also like it when ideas get turned into images. Alas, I am no graphic artist or designer, but I can write and I also enjoy writing. That’s where DALL-E comes in handy as a creative outlet.

This blog will give you examples of how I use prompts to turn ideas, stories, sometimes entire poems into images. With DALL-E 3, creating image is actually a conversation where you start with an initial prompt, and then nudge your AI friend to adjust the image the way you like it. (Often the opposite happens, but you will see.)

In each post, I will include the result of one prompt. More precisely, I will show most of the conversation and the images that came out of the prompt, with some commentary: either on the story that’s depicted, or the AI quirks I observed.

This blog is inevitably a deep dive into the “psyche” of an AI, a study on how a multi-modal large language model (that’s what DALL-E, built on GPT-4, is) connects language with images or parts of images. This connection is peculiar and unlike anything that you’d find in a human mind. I wrote about this on another blog of mine: DALL-E, the perfect alien.

Let’s point out that the AI is not the artist here: I am (as much as I can call myself an artist). DALL-E will not paint anything on its own accord; it will not feel like it needs to create an image about something or other. Instead, it gives me a very wobbly and unpredictable paintbrush that I can sort of control through my words. The conversation with the AI is my creative process, showing how I am trying to find and tweak my words to get the result I want.

The header image of this site was of course generated with DALL-E 3. The prompt was: “Create a widescreen image depicting a very simple medieval watchtower on a cliff. Two more towers can be seen in the distance. There is a flock of birds flying back and forth between the towers.” The bottom of the image is cropped off to meet the style recommendations for the site template.

In real life, my name is Balázs and I run a software company that, quite predictably, integrates generative AI into its work. At the same time, I am passionate about, and trying to advocate for, the ethical development and use of AI. My LinkedIn is “kisb971” and my Instagram is @kisb971.

The real Third Tower is in the city-state of San Marino, embedded into Italy, close to the Adriatic coast. Hungarian writer Antal Szerb used it as a symbol of freedom, in a little book that he wrote about his journeys in the newly fascist Italy, in the early 1930s. I have another blog that bears this name: visit that blog (thirdtower.ink) to find out more about the connection.