Testing Stable Diffusion

Generate images of whatever you can dream up with stable diffusion

I decided to investigate Stable Diffusion. Stable Diffusion is an AI model that takes text input and turns it into an image. You’ve probably seen it in the news; John Oliver even had an episode about it.

It turns out it’s incredibly easy to install on a Windows PC with this installer: https://stable-diffusion-ui.github.io/ Just run the installer and it’ll download all the dependencies, the model, and launch a web UI for you to work with.

Even though everything I’ve read said you need a high end PC to render images, I found on an almost 7 year old GTX 970 with 4 GB of RAM, I was able to render images with the default settings in about 30 seconds.

Initial Experiements

At first I tried coming up with very specific ideas and seeing if I could get the AI to spit out what was in my head. Most of my experiements just turned out weird results. For example, here’s the prompt “sonic the hedgehog skydiving with a parachute, sky, daytime, ((parachute))” (excuse the extra parathensises; I misunderstood how to use them). I actually tried multiple attempts and this was the best that I got… and he still didn’t have a parachute.

sonic the hedgehog skydiving with a parachute, sky, daytime, ((parachute))

I tried being a little more generic and asking for some characters from the old TV show Saved by the Bell… and instead of two characters, I got the same character looking at himself, except with dark hair. Here’s “A.C. Slater and Zach Morris in class together, bayside high, saved by the bell”:

A.C. Slater and Zach Morris in class together, bayside high, saved by the bell

Things got pretty funny when I asked for two characters from It’s Always in Philadelphia, “Frank Reynolds and Charlie Kelly, It’s Always Sunny in Philadelphia, Paddy’s Pub”. I will say, it got the background right; it looks like these fake characters are actually in Paddy’s Pub:

Frank Reynolds and Charlie Kelly, It's Always Sunny in Philadelphia, Paddy's Pub


At this point I’m thinking, okay, this is funny but what’s the point? All this I hear about AIs taking people’s jobs are overblown. But then I asked for one character from Always Sunny done like a drawing. Here’s “Charlie Kelly, It’s Always Sunny in Philadelphia, Paddy’s Pub, Cel Shading”:

Charlie Kelly, It's Always Sunny in Philadelphia, Paddy's Pub, Cel Shading

… wow. This looks professionally done. And my 7 year old gaming PC spit it out in 30 seconds.

Things got much more impressive when I gave up some control and asked the AI more generic things. For example, here’s “fugitive, dystopian future, hong kong, rain, concept art”:

fugitive, dystopian future, hong kong, rain, concept art

Trying to get rid of the umbrella and tweak it a bit, I tried “fugitive, woman, tattoos, dystopian future, hong kong, rain, concept art, night, police, ((((umbrella))))”. (It’s supposed to exclude the umbrella with that syntax but it left it in… but whatever, I think it looks cool!) This is what I got:

fugitive, woman, tattoos, dystopian future, hong kong, rain, concept art, night, police, ((((umbrella))))

Trying a bit further, I just asked it the name of one of my favorite video games of all time, “myst”. It gave me some box art to a game starring what looks like a woman with pink hair in a pink dress.


I then tried expanding it to asking the AI for some specific things from the game, and it spit out what looks like an area that was inspired by the level. Here’s “myst, channelwood age, linking book”:

myst, channelwood age, linking book

Please keep in mind, I’m showing you the best stuff. At least 50% of what I generated was garbage or nightmare fuel. For example, here’s Eric Cartman from South Park with the prompt “eric cartman, south park, eating”:

eric cartman, south park, eating

Take Aways

So what’s my take away from all this? We’re at the beginning of an AI revolution. It reminds me of the Internet explosion in the 90s. We’re at the beginning of something that will forever transform the way we operate as a society.

I understand now why so many artists are worried for the future. Let’s say I’m a game development company. I might forgo hiring a concept artist and instead just use AI tools to generate concept art. Or maybe I’m making an animated TV show for children (which tends to have cheaper production values and lower expectations for quality). How long until these prompts start spitting out 30 minute TV shows?

My bigger concern is around spam and scams.

These tools are still early, but it’s crazy how far they are. Sure, they still require some manual work and picking and choosing the best images. But I don’t have to get a four year art degree to make a painting. Now I can make it in 30 seconds on a 7 year old gaming PC.


I was playing around with this some more, and I have to share two other awesome images. First, here’s “Eric Bischoff, Anime”:

Eric Bischoff, Anime

Second, here’s “labradoole president of the united states”:

labradoole president of the united states

Back to Blog Posts

You may also like:

I Tried Google Stadia

Is this the future of gaming?

Software Engineer vs. Computer Programmer

What is the difference between a software engineer and a computer programmer? Read more...

New Windows Game: Green Tea Reversi

Command-line reversi that supports all rules and any combination of human and AI players Read more...