AI image generation over 2.5 years

People, it needs-must be remarked, tend to think of new technologies as a point rather than a vector. What I mean by this is that people grasp new technologies, their immediate reaction is to think about what the next five years are going to be like with the technology as it exists now. But that neglects new technology isn’t stationary.

I wanted to illustrate this by showing how fast things change. About two and a half years ago I happened to use an image generator to create images to go with quotes from Sufjan Stevens songs. Now I want to show you how much things have changed in 30 months or so.

(Orginal image generator here).

A jump of similar magnitude again, whether in large language model performance, or image generator performance is almost unimaginable. I’m not saying a jump like this is going to happen again, but certainly, there will be a jump.

You stare at the sun to see the sublime

All things grow

The lion and the lamb are restored

I was dressed in white

His father was a drinker

In the tower above the earth

Justice delivers its death

I’ll find sleep, I’ll find peace

Blackbird on my shoulder

For this one I wanted to include all four generated Midjourney images because I was shocked at how similar they all were!

I built your walls around me

All of me wants all of you

Signs and wonders

My friend is gone, he ran away

We were ashamed of her

We’re all going to die

It’s your own damn head on that plate

For this one, Midjourney refused to generate an image, because its language AI flagged it as too violent, a change that seemed as interesting as any image.