Generative AI is artificial intelligence that generates something like text or art. I have had an interest in artificial intelligence since studying several AI subjects as part of my BSc(Computing) back in the 1990s. However, nothing much came of it back then. Advances in recent years have astonished me and I have been doing my best to keep up. When ChatGPT was released around early December 2022 I jumped on the wagon immediately, got my account and started playing. But the other thing that fascinated me was AI that generated art. This post is about a series of experiments with generative AI and art at the beginning of 2023 (the results originally posted on Twitter). I will simply show you the prompt and the output and leave you to make up your own minds.

The idea was to give the same prompt to a number of different generative AI products and see what they gave me. I chose to use as a prompt a moment in scene from the novel that I am currently writing. So, I tried to describe that moment using references and images that I hoped would be understood. Here it is: ‘A steampunk scene with a horror feel. Night. Slums. A mix of 17th century London and Venetian architecture. Lots of rubbish. Rain. Cobblestones. A few very old lamps hanging from walls. Raining. Puddles. Moss. Lichen. A lean but strong dark-skinned gunfighter in a heavy black cape and a black broad-brimmed hat. His eyes are like fire. A sharp-faced woman following, dressed in black with brimmed hat and cloak. A shadowy figure in a doorway behind them.”

Fotor

Fotor is a long-time photo editor, dating from around 2012. I found its outputs a bit disappointing as there was not much of a steampunk feel to it. Basically, it was just old laneways.

Midjourney

Midjourney opened its beta testing in July 2022 and I had to use Discord to access it.  I actually use the desktop Discord app rather than the website, but that is me. As an artist I’m not sure whether to say “ouch” or “wow” as to its results to my prompt.  Not quite what I wanted, but still impressive. I really love the colour scheme that it used.

Stable Diffusion

Stable Diffusion has German origins and gets its name from being a neural network with a latent diffusion model. Wow. The results were different again. Black and white. While I found it pretty impressive, the colours of Midjourney still had me sold.

Deviant Art Dreamup

I’ve been a member of Deviant Art for quite a while and am a fan. They describe their AI offering this way: “DeviantArt DreamUp™ lets you create AI art knowing that creators and their work are treated fairly.” I was actually quite happy with the output, though I still feel that Midjourney has the edge.

DALL E2

DALL E2 is part of that wonderful suite that OpenAI have given us (think ChatGPT). Again another wonderful set of outputs, but I still prefer Midjourney.

Crayon AI

Crayon was formerly DALL E mini. I feel that its outputs were incredibly atmospheric even if they did not contain many of the elements that I sought (like the people).

Genmo AI

Again, another worthy bit of output but lacking the horror aspect that I wanted.

Conclusion

Generative AI has a lot to offer in the future if this is the kind of outputs that these Alphas and Betas can produce.