Day 2 of 30 Day AI Challenge – Generating Images and Art

Ok, time for Day 2 of my #30DayAIChallenge. Every day, I’m going to try and do something different or better, using currently available Artificial Intelligence tools…today’s challenge is going to focus on art and image generation.

There’s a (necessary) debate raging about the process and outputs of AI generated visual content. On the one hand, can it really be considered ‘Art’ if it is machine generated and reliant on the creations of countless human artists that didn’t choose to have their creations scraped into an AI model? Can images created using AI tools be copyrighted? Clearly the US Copyright Office thinks not.

Without going too deep down these rabbit holes, I’m going to focus my effort on simply creating visual content I’ve been unable to find through traditional means. I’ve always loved this parable, claimed to be of Native Indian origins:

An old Cherokee is teaching his grandson about life. “A fight is going on inside me,” he said to the boy. “It is a terrible fight and it is between two wolves. One is evil – he is anger, envy, sorrow, regret, greed, arrogance, self-pity, guilt, resentment, inferiority, lies, false pride, superiority, and ego.”

He continued, “The other is good – he is joy, peace, love, hope, serenity, humility, kindness, benevolence, empathy, generosity, truth, compassion, and faith. The same fight is going on inside you – and inside every other person, too.”

The grandson thought about it for a minute and then asked his grandfather, “Which wolf will win?”

The old Cherokee simply replied, “The one you feed.”

The parable is a great reminder of the power of positivity and the unnecessary control we sometimes relinquish to the negative. On a few occasions, I’ve scoured the web for a great representative image I could use as computer wallpaper or something I could print out for my office wall. Unfortunately, most options have fallen short of exactly what I was looking for. Here’s an example from Google Images:

A couple of interesting options but nothing I’d want to use. Here’s a stab at ShutterStock, which couldn’t produce any results (my search wasn’t simple and specific enough) and generously recommends signing up for a ‘free trial’ in order to use AI capabilities.

Ok…time to see what the free AI tools can do. First, let’s start with a pretty detailed description of what I need:

A photo realistic image of two wolves fighting in a dark and suspenseful atmosphere. One wolf is good and represents positivity plus all that’s good with the world. The other wolf is evil and represents negativity and all that’s wrong with the world. Show the good wolf winning, with text at the bottom of the image that says ‘Which Wolf are you feeding?’

Here’s what the top few AI models generated.

DALL-E 2 by OpenAI

Pretty awful…and clearly the model can’t handle text captions too well. Here’s another shot with a slightly tweaked prompt and no text:

I then asked DALL-E to create variations of the last image, which ended up with a few more similar options…though many of them look pretty bizarre! Also interesting to note is the the credit system used by OpenAI for its products, including DALL-E…you get 15 credits every month and I used up about 3 creating the examples here.

Cr(ai)yon

These ended up looking a lot better than DALL-E, but still not amazing (and there are still a few warped looking wolves in the mix). The positive side is that currently Craiyon is offered free, with unlimited access and is quite simple to use.

WOMBO Dream (using the iOS App)

This one lets you start with a text prompt or other images, plus pick a style. Free options are somewhat limited and there are plenty of Ads to suffer through, but here are a few pictures I was able to generate using the app:

This tool has an interesting interface, with loads of input options and styles to pick from…but the end results still leave a lot to be desired.

MidJourney

Finally, time to test MidJourney, accessed through Discord. Now, this is more like it…MidJourney produces some pretty high quality output and more along the lines of what I was looking for.

I asked the bot to create a few more variations of the second image:

…and finally, a higher resolution version of the first image, which I really like:

To wrap things up, I take the image into Canva and add in the caption I like, plus produce image and PDF versions suitable for printing:

That feels like a Day 2 mission well accomplished! I’ve probably only scratched the surface of what’s possible here, but it’s interesting to see the various models and their approaches, plus efforts to already monetize the space. Clearly MidJourney blows the competition away in terms of quality and power…I’m keen to see how this evolves and what other competitors emerge to challenge it. One thing to note is that it does take a bit of fine-tuning to get exactly what you need, so you may need to play with different ways to describe the images or art you’re looking to create (case in point is the story cover image of a robot painter, in the style of Van Gogh).

In the meantime, I’m happily putting up my little creation on my wall. Let me know if you’ve given any of these (or other) tools a try? Any favourites? Share your own creations using the hashtag #30DayAIChallenge

See you tomorrow!