AI ART (Patreon)

As most of you are likely aware, AI art is everywhere these days. WFRP Facebook and Reddit groups are awash with it. Online communities have entire servers dedicated to it. And GMs everywhere are tentatively testing this new image generation tool and deciding if it has a use for them.

And with that in mind, is it of use for us at LAWhammer?

Addressing this, I recently had a brilliant chat about AI art and GMing that you can watch over on Inside the Rookery. I've also discussed the subject many times online, and have been messing around with AI art pretty extensively to see what the fuss is all about.

I'll admit I started with many reservations and prejudices. Some of those concerns are now behind me, others are with me still. Putting aside the impact AI art will have on the games industry, it is definitely a brilliant tool for GMs everywhere. At the very least, it offers us all an easy tool to visualise and inspire our imaginations, and RPGs are all about that.

So, will I use AI art here on LAWhammer?
For a project like this, where I intend to create art for pretty much every NPC in our campaign, having the support of an AI art program as one of the many tools in my toolbox for building images seems to be a no-brainer. Writing for the Rookery takes up almost all of my day, so anything that helps me build images fast during my free time is, as far as I'm concerned, 100% a good thing.

As those of you who follow our Discord know, the AI generator I've been messing around with most is Midjourney. The images it creates are often rubbish. But amongst the detritus I've uncovered gems, and I'm more than happy to use them if they work for my personal games. Coz, really, why wouldn't I?

The most pressing reason not to use AI art for my personal games is because most of the images are not quite what I want. But that's not necessarily a bad thing. An image may not be right, but it can be a great source of inspiration for getting it right. As I'm an arty chap when I want to be, I'm in the fortunate position where I can repaint any image an AI creates to match my requirements. I'd not do that for professional work as the copyright ownership for AI images is as stable as a tide-locked sandcastle, but for a project like LAWhammer, it's perfect.

So, what does that mean for LAWhammer in practice?
Well, let's say I want an image for a PC or NPC. I can screw around in Midjourney to see if it can nail a basic shape I like. If I find something that speaks to me, I can then repaint it later to match my needs.

Want to see how that works?

When I was initally building images for the PCs in the upcoming game, I couldn't figure out how I wanted Gerhardt, the Noble Scholar, to look. So, I threw the problem at Midjourney.

My 'prompt' - the words I used to 'inspire' the AI art generation - could hardly have been simpler: 'young male noble, slight smile, medieval'. Midjourney threw back the expected odd-looking nonsense including images such as this:

Yeah, no. Should have used a better prompt. But I was on a fishing expedition. I wasn't quite sure what I wanted yet.

So, I pressed on, and more of the same was pumped out as I refined what it offered.

Still very much not what I wanted.

So I just kept the AI running in the background as I worked on other stuff, refining the images as I went. I expermented with styles, tightening things each time. Soon, scores of images had scrolled down my screen, but they were still... #notright.

So, I was still far from inspired.

But, I presevered.

Eventually, after more faces than I could count (all running as I was editing a document for the Rookery) it unexpectedly offered a face that clicked for me. It wasn't perfect, but there was something in it that caught my eye.

Can you guess which one?

It was the fourth one.

Yes, really, that one.

It had the neck of a giraffe, and the less said about those eyes, the better. But, through the anatomical disaster on display, I could see something that looked reminiscent of the Gerhardt that was swimming through my mind.

Now, if I wanted a 'perfect' image, I'd have had to keep reworking what the AI offered and hope it nailed what was in my head. But, that's not what I was doing here. I wasn't looking for a finished piece - I was simply looking for something to send me off in the correct direction. And I figured number 4 might have been able to do that.

So, I upscaled the image to look at it more closely.

Yup. That still looks really stupid. But, it still spoke to me. I thought I could do something with it.

So, I pulled the image over to Photoshop, removed everything but the face, faded it off and used it as the undersketch for drafting up an image of Gerhardt. I liked the smile and the face colours, and wanted to keep those elements. But, the rest?

Well, they went out the window.

I beavered away, and in very little time, I had something that was much closer to the mark I was intending to hit. It now looked like this.

That'll do!

Yes, it's super sketchy in many places, but I was going for speed here. Normally, because I'm such a worrywort, I would have spent absolutely bloody ages messing around with different sketches, colours, and shapes trying out different faces types until I got one that I liked. And that can take hours for me. Sometimes days if my obsessive need to hit the details overcomes me (which it frequently does - it's one of the reasons I normally stick with maps: those are detail-orientated by nature, so match my mental map well).

As the above was created in a single sitting, I scored a significant time saving. So, 100% worth it. Not because the results were necessarily better or different to what I would have created without the Midjourney reference. But because I created the image in a fraction of the time I normally would have. And, given my current circumstances, that's exactly what I need as I don't have much free time.

Sure, I still had to sketch in a new body and green velvet cloak, splat in a background, add the Jungfreud heraldry, and draw a giant feather in that hat. But, I still had a pretty big time saving.

So, job done. AI reference success. Big time-saving win!

This wasn't the first image I'd tried to pin with an AI. I'd dropped a few others in place when discussing the merits and flaws of it in Discord - some more heavily AI, some barely AI supported at all, some posted online for others to comment on - but this was the first image that convinced me that using AI art to speed up the process for creating LAWhammer content was absolutely worth it. Very much a no-brainer.

But is that everything I've to say on the AI art subject?

Not quite.

Is the copyright surrounding AI art an issue?
There is a big question mark hovering over the copyright for images created with Midjourney and other AI art programs. Outside the rights I grant to Midjourney as a part of their terms of service, this is where things currently stand (taken from here):

You basically own all Assets you create using Midjourney’s image generation and chat services.

Here's a screenshot of that section.

That means, long story short, that the copyright for all assets created in Midjourney currently belongs to the prompter of the AI. Yes, there are caveats laced into the terms of service, but they do not change that core. So, image creators who subscribe to Midjourney currently own the rights to the images they create using the AI.

However, that was not always the case.

When I first started using Midjourney, the copyright was explicitly stated to belong to no one. So, all images were copyright free.

But, in time, that changed. 

Indeed, it is likely to change again.

If you're interested as to why it is all so changeable, we can discuss it further in another blog post. Long story short, there is human interaction and refinement involved with the image generating process, so the created images meet the current legal requirements to be defined as art in most countries, and are thus copyrightable by the prompter, not by the algorithm creator, or the teacher of the AI. Or, indeed, anyone else. This is likely to be tested in court in many countries over the next few years, so where this all eventually lands is still unsure. But, as it stands, the image creator owns the copyright.

Of course, it's not quite that simple. The above only covers a small percentage of the complicated issues at play here. For example, how AIs are 'trained' and the copyright implications of that is enormously relevant, but that's not really what this blog post is about. If you're interested in that subject, tell me and I'll possibly drop more words on the topic. It's all quite fascinating, and it has many historical parallels to other emerging technologies into the art space, including photography, computer art, and Photoshop. AI art, to me, is just another step on that journey, and, so far, it's been a lot of fun to both engage with it at these early stages, and to see where it's all going to go. The program is improving weekly, and the images it can create today are already much better than those I posted above.

And, with all that understood, we return to the underlying point of this entire post.

Will I use AI art to generate or inspire any art for LAWhammer?
Sure, why not? As long as it matches what we need and doesn't look rubbish. It can potentially save a hell of a lot of time. And, further, it's all art that I apparently not only have the rights to use as a creator, but is also, currently, copyrighted to me. Nice!

See the Gerhardt image above as an example of how AI art can inspire images.

Will the images I create for this campaign be offered on the blog, whether they are AI-created, AI-assisted, or entirely original?
Hell, yes. Share the wealth. This kind of stuff will all be free for all readers of the blog to use in their private games in any way they wish.

Rock on.

Win-win for everyone, I reckon.

Like it? Use it. Don't like it? Don't use it.

To finish off, if this topic interests you as much as it interests me, do drop into the LAWhammer Discord and join the conversation. Also, take time to watch the Inside the Rookery video I mentioned at the start of the post. Lots of cool stuff in there that's directly relevant to this post.

Andy Law

PS: The copyright details in this blog post were, to the best of my knowledge, correct at the time of writing. It's very likely that those details will change in time.