Perfectly Imperfect

When generative AI creates an image (aka 'Synthography') - is that a threat to creativity or is it just Clipart and Stock Photos for the modern era?

Clipart, Stock Photos, Synthography - there's no shame in any of them. They're all super helpful when you don't have the budget or time to engage with a proper designer/artist. Sometimes the "good enough" is all you need because the focus isn't on the artwork - it's on the speed of information consumption.

20+ years ago, if you needed to represent a concept you'd turn to a vast clipart library.  

Because there was no infinite library of clipart available, it often meant looking for generic ways to express your idea. And like any trend, it was cool when the kids started using it, started to get overused, then jumped the shark when your parents or teachers started dropping them into everything.

As the internet grew, and data storage became cheaper, we were treated to a much larger library of stock photos. It seemed like photographers were constantly churning out new scenarios they could represent. (As a fun fact for those too young to remember, before stock photo websites you used to buy CDs filled with stock photos...)

This allowed content creatores more freedom to represent their ideas graphically - but the stock photos sites still had their limitations. But again - the stock photos were approximations. Generalized visual metaphors of the the idea being expressed.

Today, with generative AI - the possibilities feel endless. But are they? Generate enough images through any model, and you start to see the same patterns and limitations.

It's super helpful in getting us closer to the idea we're looking to represent graphically, but like Clipart and Stock Photos, it only when works when the ideas you are looking to generate are in the model's 'vocabulary.'

It allows us to express ideas quickly, but when the focus is on the art/design/look it still falls short. That's where designers, artists, are still very needed. 

Case in point: Children love drawing. It's not a question of skill, talent, motor control, or ability to be picture perfect. In fact, they love to experiment, get colorful and represent things - either the way they seem them, or the way they wish they could see them.

Children draw in a perfectly imperfect way. It's fun to look at their drawings because if a picture tells a thousand words, a children's drawing tells a million.

As a simple experiment, I asked a few different AI models to draw like a child. I was very explicit with my instructions:

Draw a drawing of a family, the way a 4 year old would draw one. Have a dad, a mom, an older girl and younger boy. Have them stand in front of a house, and include a car. I want it to be authentic to how a 4 year old would draw. Make sure proportions are off, there's no symmetry, and the shapes are simple.

I didn't get creativity in response. I didn't get imagination. I didn't get scribbles, or multi-directional coloring. What I got, regardless of the model I used, was a modern version of clipart:

All the drawings were far too good. It couldn't capture the originality, the innocence. It all felt flat. It was all an approximation.

Don't get me wrong: I love using generative AI - but for speed, to help get ideas across quickly.

When it comes to prose, when it comes to art, when it comes to quality being defined by the aesthetics of the work itself and not solely as a measure of the idea and speed of its consumption - AI isn't there. 

...Yet.

(That being said: the rate of improvement has so accelerated that I'm sure once I hit 'save' I'll be proven wrong with the release of a new model)