AI and the Art of Propmaking

I'm now on board with the generative AI hype.  It's an amazing technology that's going to revolutionize the world.  More specifically, it's going to open up new possibilities for propmaking.

I spent a few hours fiddling around with Midjourney over the weekend.  If you're not familiar with the service, it uses artificial intelligence technology to generate images based on text prompts.  To be honest, I initially didn't have much interest in it.  For every truly impressive image I've seen posted there were a multitude that looked terrible. Those crappy images have flooded social media over the last few months, triggering outright bans of any AI art by quite a few discussion groups.

All that helps explain my mindset going in.  I was looking for a way to generate images for spell scrolls to hand out to TTRPG players.  The illustrations would basically be visual filler.  If there's one thing the tidal wave of AI glurge demonstrated, the technology excels at producing quantity over quality. 

And then something totally unexpected happened. I was getting some really cool artwork.  

It took some time, but once I learned how to parse the text prompts my results steadily improved.  Part of that process was using the "/describe" command to upload an image and have the AI tell me what keywords and phrases would generate it.  As my significant other pointed out, the engine was effectively teaching me how to use it.

I've posted some of the results below.  For the spell scroll images I found prompts like "/imagine wizard's spell scroll, on parchment, featuring a wheel of occult symbols and runes, in the style of a precise draftsman" were productive.  The second image is supposed to be a druidic spell.  I customized the original prompt by swapping out "occult symbols" for "druidic runes" and adding "green plants".

The bottom two pictures are attempts at creating a "grimoire illustration" of the "King in Yellow".  The first popped up after six or seven attempts.  The second finally showed up close to twenty cycles later.  For some reason, the Midjourney AI has problems assigning colors.  I had to add multiple prompts to get the King in Yellow to, you know, actually be yellow.  Adding "dressed in tattered yellow robes", "wearing yellow clothing", and "bedecked in yellow" eventually put it on the right track.

From the standpoint of prop use these are fantastic results.  Much of the imagery is sheer visual gibberish, but that's fine when it comes to generating graphics for scrolls and tomes.  Give it a try.  I'd be interested in hearing about your experience.