Image generation as self-reflection

I joined Midjourney in the fall of 2022. Since then I’ve generated 228,259 images. That’s something like 175 images per day on average.

When it was still novel, I was banging out images with Midjourney. These days I go weeks without generating images but use up my credits by the end of the month.

Image generation gets a lot of flak for the potential to replace artists. People say the images are meaningless. It also uses up a lot of power.

All these are valid concerns. Yet my personal experiences with it have been rewarding.

Starting out, I was firing away at random. I’d look at what other people were doing and riff on that. I’d paste in snippets of text from books or songs. I’d try putting in descriptive text from tabletop roleplaying games.

At some point, I became more systematic and experimental. This is partly why I’ve generated so many images. Midjourney introduced new ways to control style and these would interact with prompting. They also added ways to run large sets of prompts and I’d use these to create galleries of output.

These experiments had me looking at thousands of images. I’d see various styles and formats. By doing this I started to observe what I liked and didn’t like. What made me feel something and why. I spent a lot of time

I started to learn a lot about my own taste and how my taste connects to my lived experiences and values.

By seeing many images I could understand what made them click for me. Then I could iterate and go down various rabbit holes to try new directions.

For example, I discovered that I like images with religious and ceremonial elements. I like images that convey a certain cinematic nostalgia. I learned a lot of terminology by reading prompts and researching. I really like vending machines.

People often use artist’s names in prompting. The sentiment around this is mixed and probably for good reason. But, by seeing names that others use I have discovered a lot of artists. I’ve gone out and bought their books and seen their art in person when possible.

You could get a similar experience by going through many different sources. Go to museums. Look at artist books and websites. Browse Instagram or Pixiv.

AI tools let you put this all in one place and control the experience for yourself.

I’ve branched out and used other services than Midjourney lately. They all have various strengths and weaknesses.

ChatGPT’s image generation is very good at style transfers. You can put in a photo and get back something with the same composition in any style you can describe. Others are good at more direct control or SVG.

Midjourney still has the best aesthetics to my eye. It still is the best at rapid experimentation. And, for me, that’s where the value is.