A.I. Imaging

Best Selling RPGs - Available Now @ DriveThruRPG.com
Status
Not open for further replies.
I've had the same experience myself, mostly. However, I just managed to get these in two passes. I don't know if you really wanted these, but in case:
0LumI62jl8wUuKK2awTc--1--5m6t0.jpg
0LumI62jl8wUuKK2awTc--2--qjj83.jpg
0LumI62jl8wUuKK2awTc--3--9jgf0.jpg
0LumI62jl8wUuKK2awTc--4--bqknf.jpg
I used "an old brickwork wine cellar with long shadows and twisted swirly walls, with wine barrels along the walls" in Stable Diffusion v1.5 with Prompt Weight 90%, which gave me four views of brick halls with curved ceilings and no barrels at all. Then I fed the most "wine-cellary" one back in as a seed image, with the only prompt "a row of wine barrels along the walls", and it gave me the above images. I was shocked, as I rarely get such obedient results.
Nice. Though it's a bit clean or modern looking. I was also trying to get a hooded figure in the middle. At one point there was a bottle of wine randomly floating in the air. This is what I have so far.

Screenshot 2022-12-18 213037.jpg
 
AI art, like AI text, is a nice parlour trick, but you can almost always tell, because the AI has no idea what it's actually doing, it's just pattern matching, so you get Mobius effects everywhere and things just in the wrong shape or placed in the wrong position.

Still better than Leifield at his worst though.
 
Here's a stream of more typically weird attempts to add wine barrels to your image :hehe: :
FLuwHIL2FYDR6tdPgDRC--1--am5fm.jpg
FLuwHIL2FYDR6tdPgDRC--2--96v0t.jpg
FLuwHIL2FYDR6tdPgDRC--3--5fnxn.jpg
FLuwHIL2FYDR6tdPgDRC--4--0nn7f.jpg
xl6WWExKDSEbzawjBKpN--grid.jpg

EB9S8jJKWWdPtOX0JIfG--grid.jpg
UZSnitifsq35ThkQqAqd--grid.jpg
 
So morally human art students shouldn’t study contemporary art to avoid using another artist’s art to effectively screw them over by copying their technique and themes?

Legally it OK for this, human art students, to happen because the artist only has the legal right to control the copying of a specific image. Also some countries recognize a moral right for the artist work to remain as is in the form the artist intended. But that doesn’t prevent the copying of themes and techniques.

Just trying to see where the line being drawn here. As the AI models do not store the actual images themselves.

Yeah, I'm still trying to articulate where the line might be, and it might be a moot point in the end, but I suppose the difference is that humans have to develop a skill that's labor intensive to use with only a limited output, while the AI is an automated process that can be replicated indefinitely for mass production. And it involves the development of new technology that would presumably benefit the creators in ways that extend beyond the benefits an art student would get from learning art, since this could potentially have other applications down the line.

It's kinda similar to the way that social media provides a "free" service for users, but ends up farming a bunch of user data in the backend that's invaluable to advertisers and even tech companies themselves in predicting user behavior and learning how to manipulate them. It's kinda innocuous and in the surface it's like "Who cares? It's a 'free' service and nobody's forcing you to use it!" Yet social media can now decide elections and has helped pave the way for mass surveillance, since these platforms can even track your location data.

I'm not sure AI art can be that pernicious, but it still involves tech that's still in development and bound to have unforseeable consequences or uses. And also bound to be far more valuable monetarily and in terms of broad application than any single art student's education.
 
Yeah, I'm still trying to articulate where the line might be, and it might be a moot point in the end, but I suppose the difference is that humans have to develop a skill that's labor intensive to use with only a limited output, while the AI is an automated process that can be replicated indefinitely for mass production.

The thin end of that wedge goes back to the introduction of printing.
 
Last edited:
I find it a bit addictive, trying different settings and strings and seeing what gets generated. Occasionally, it's something pretty good, and sometimes, worthwhile variations are created, but other times, you get surreal weirdness, where one finds somehow it doesn't really know the difference between wine barrels and stone columns, etc.

Some more offerings on the attempt to get a hooded figure in an old wine cellar:
Rs1iLoDEO7zmPVWIzxlM--2--0uwcr.jpg
Rs1iLoDEO7zmPVWIzxlM--4--nt6ic.jpg
Rs1iLoDEO7zmPVWIzxlM--3--9d94z.jpg
dFXqsjPSjfJAz1uMErmn--3--csizb.jpg
gJQMhGOc6thRi1gJHqjA--3--cwcid.jpg
 
I find it a bit addictive, trying different settings and strings and seeing what gets generated. Occasionally, it's something pretty good, and sometimes, worthwhile variations are created, but other times, you get surreal weirdness, where one finds somehow it doesn't really know the difference between wine barrels and stone columns, etc.

Some more offerings on the attempt to get a hooded figure in an old wine cellar:
Rs1iLoDEO7zmPVWIzxlM--2--0uwcr.jpg
Rs1iLoDEO7zmPVWIzxlM--4--nt6ic.jpg
Rs1iLoDEO7zmPVWIzxlM--3--9d94z.jpg
dFXqsjPSjfJAz1uMErmn--3--csizb.jpg
gJQMhGOc6thRi1gJHqjA--3--cwcid.jpg
#3 would look good enough to be included art in an RPG to me.
 
The thin end of that wedge goes back to the introduction of printing.

Yeah, that kinda crossed my mind after writing that snipped, but as I tried to articulate (perhaps ineptly) in the rest of the post, there are aspects of this tech that go beyond even the printing press, since this involves the development of what could be termed "smart" technology that's capable of learning from data gathered or feed to it. The printing press by comparison is "dumb", purely mechanistic technology that merely prints a predetermined pattern that still needs to be created by a human. This thing can develop its own patterns by analyzing existing patterns and produce data in the process that can be farmed to make new developments in the tech and provide additional long-term value for the companies producing them, partly at the expense of the art feed to the software.
 
Meanwhile, like a chump, I'm here trying to illustrate my next game by hand, getting frustated with myself as the vision in my brain doesn't refelect what my pencils are doing.
I recommend using A.I. Art. Not to use directly but as a random generator of a bunch of poses and settings and then use that as a template (by eye) for your own effort. Half the battle is finding the right composition and proportions for the image you have in mind.

For example I generated a bunch of images for a medieval Russian village in winter by a river and found two that had the right composition and proportion that allowed me to make a drawing by hand much better than I could draw from scratch.

1671456062617.png
 
I find it a bit addictive, trying different settings and strings and seeing what gets generated. Occasionally, it's something pretty good, and sometimes, worthwhile variations are created, but other times, you get surreal weirdness, where one finds somehow it doesn't really know the difference between wine barrels and stone columns, etc.

Some more offerings on the attempt to get a hooded figure in an old wine cellar:
Rs1iLoDEO7zmPVWIzxlM--2--0uwcr.jpg
Rs1iLoDEO7zmPVWIzxlM--4--nt6ic.jpg
Rs1iLoDEO7zmPVWIzxlM--3--9d94z.jpg
dFXqsjPSjfJAz1uMErmn--3--csizb.jpg
gJQMhGOc6thRi1gJHqjA--3--cwcid.jpg
#4 probably comes closest to what I'm looking for. Still, a lot of oddities going on there. The door at the end looks weird and the proportion of the hooded figure is just off. The vaulted brickwork is pretty much spot on. Now the question is. If you could draw, how much time would it take to draw it compared to how much time its taken to get it nearly right? I do agree that its quite addictive :smile:
 
#4 probably comes closest to what I'm looking for. Still, a lot of oddities going on there. The door at the end looks weird and the proportion of the hooded figure is just off. The vaulted brickwork is pretty much spot on. Now the question is. If you could draw, how much time would it take to draw it compared to how much time its taken to get it nearly right? I do agree that its quite addictive :smile:
Yeah, that is a good question. I can draw . . . ok . . . but drawing something about like that would take an hour or two, and have different qualities and limits.

What I can also do, is use an image editor ok, too. The best bits can be combined (the door in particular would be easy). And, I got #4 by applying a mask to #3 and asking for a row of wine barrels again.

Some of the best AI art results I've gotten have been by taking an image that's something I like, or part of it, and using that as the base image.
 
I've used AI art for 'trim' and filler. But commissioned art and stock images for the actual important stuff.

The whole argument is BS though - even with the "omg it was trained on my art" claim. The "harm" is in their heads even if they had it trained on their art.

In a decade it will be an essential tool for an artist. No-one expects them to design their own gradients. They use heal and retouch with impunity. It will just be a labour saving device.

THAT SAID...

It should be easier for artists to train their own models. I think it should be retasked as a tool for artists.

It seems to me the tech would be very useful as an art enhancer rather than creator. Like a super smart photoshop, where you could take a photo and then have it recreated in a specific style, like line drawing, oil painting, cartoon etc. Many editing programs can already do this but require a fair bit of time and practice to do really well. Similarly you could take a basic sketch and then work with the AI to flesh it out, thereby reducing some of the really weird stuff like 9 fingers or 3 eyes.

I think this would go a long way towards softening the impact, as rather than godless AI putting artists out of work it could be spun as wonderful AI helping the artistically challenged realize their visions.

Well, yes, but these artists ask permission and pay copyright dues.

No. Wait. They don’t.

It’s exactly what they do and what they have been doing for centuries. But it sucks when a machine does it quicker.

meanwhile … farm workers, supermarket workers, print workers and looking on and thinking “ya”

Certainly no wanna be fantasy artist has stacks of art books the containing work of Vallejo, Frazetta, Hildebrant, Elmore, Trampier, Danforth etc providing them with inspiration and style cues... :wink: AI is just more efficient than the human brains at digesting imagery, but on the other side it lacks the ability to look at sexually explicit material and decide if it is pornography or not.

Meanwhile, like a chump, I'm here trying to illustrate my next game by hand, getting frustated with myself as the vision in my brain doesn't refelect what my pencils are doing.

Yeah, I hear you. I always seem to buy the wrong brand of pencils because they never seem to draw what is in my head. I think the paper is faulty as well. :grin:

I recommend using A.I. Art. Not to use directly but as a random generator of a bunch of poses and settings and then use that as a template (by eye) for your own effort. Half the battle is finding the right composition and proportions for the image you have in mind.

For example I generated a bunch of images for a medieval Russian village in winter by a river and found two that had the right composition and proportion that allowed me to make a drawing by hand much better than I could draw from scratch.

View attachment 53267

Looking at some of the AI art shown here it does seem like it could be a helpful tool for quickly providing a baseline to work from. Exploring different ideas, or even throwing random stuff at the wall until something gives you some good inspiration. Not much different than doing an image search on Google.

I put the same phrase into a Google Images search and this was one of the photos that came up. Sure it is a modern photo (not many photos from the Middle Ages...), but very useful as a foundation to work from. Adjust the architecture as needed, back date or remove the bridge etc to better fit your needs. Add dragons or sea serpents to taste.

1671496045192.png
 
DriveThru's latest policy is that you have to designate your product as containing A.I. art during upload and then another disclaimer somewhere else... I think in the product's description. Apparently, customers demanded it.

This is very good to hear and something that I'd like to see enshrined in law overall, in the same way as ingredient labels.
 
I've unsubscribed from Midjourney and given up AI art for the moment. It was making me existentially queasy, and it was pricey to boot.

Keeping my AI-generated ship avatar though.
I run it locally on my iPad (using Guernika). I'm ne'er going to use it for a book*.

*I've been chatting to some folks about training the AI on public domain art. Then the only objection is "Well, you could have hired a human artist" - I buy plenty of art, including heaps that will likely never be used.

BEC6CFF4-42BC-487C-8901-433450C9CDD9.jpeg
 
I've been using the free dream.ai to produce a few visuals for my players and for characters I'm playing. It isn't as good as other generators, but on the other hand, it is free. As someone who has negative skill in art, it's nice to be able to make pictures instead of writing 1000 word descriptions for things.
 

look, they have a pile of stock art they own the rights too, they have an AI engine. some intern put this together in a couple hours, and bam.
 
I've found AI art great for inspiration for random monsters. Just ask it for something - it gives you something totally different from what you asked for and all you need to do is describe it and give it stats.
 
I've found AI art great for inspiration for random monsters. Just ask it for something - it gives you something totally different from what you asked for and all you need to do is describe it and give it stats.
Yeah absolutely. I also make Funko Pop-style versions of all our characters cos it's funny. It's also fun to take like random "snapshots" of the game.
 
I hope there will be appropriate copyright and payments made to these masters. I mean, this is literally what "we" have been criticising AI art for...

You seem confused about what the word copyright means and what are the actual objections to AI art, of which there are a myriad, but I've not seen a single ne revolve around "copyright payments", which would be stupid.

Moreover, since they certainly don't represent my personal objections, or anything that I've said, I certainly don't appreciate you throwing shade my way and using my post to construct a strawman argument for your soapboxing.

7bk156.jpg
 
You seem confused about what the word copyright means and what are the actual objections to AI art, of which there are a myriad, but I've not seen a single ne revolve around "copyright payments", which would be stupid.

Actually it's all about copyright. And copyright is all about money - literally why we have copyright is money.
But thank you for "throwing shade" that I'm confused about what copyright means or what the actual objections are.

Moreover, since they certainly don't represent my personal objections, or anything that I've said, I certainly don't appreciate you throwing shade my way and using my post to construct a strawman argument for your soapboxing.

Well, you certainly exercise your power when you don't personally like something. Wait. That's the OTHER thread.
 
I hope there will be appropriate copyright and payments made to these masters. I mean, this is literally what "we" have been criticising AI art for...
I don't think that's the criticism- copying art styles has been going on for ages. It's in using the art in the models to train the AIs. It's a subtle difference, and definitely one for more discussion (is using the eyes to train one different than using the digital medium as the input to an AI?), but it's not a discussion that I feel qualified to take on.
 
It's in using the art in the models to train the AIs.
Is using the art of others to train human artists bad? If you feed an ai only Rob Leifield bad foot art and tell it is a human with no other context, then ask it to give you a realistic picture of a human, it’s going to give you bad feet. Every single time. It actually doesn’t know the difference between style and reality until you tell it that difference. It doesn’t have any other experience, including day to day life, to determine these things. It has a trainer that says this is a human. It might get “this is what feet look like”.

this is why midjourney fails on weapons a lot. It doesn’t know the object being held is a separate entity with its own scope. Tell it “ draw me a person with five fingers on their hands” they probably don’t know


is using the eyes to train one different than using the digital medium as the input to an AI?
ultimately, this is irrelevant, as computer vision is a thing and works. You can have it use it’s cameras just like eyes and learn it the same way. It’s just how you are getting the data into the data store. The useful part comes in labeling. Take computer vision use for driving. All of those Captchas that have you check for stop signs? That’s you labeling for an AI. that’s how it works. You tell it what it looks like from every single possible angle and obstruction because it doesn’t infer especially well.

it is very neat, but also incredibly naive. We are also completely unaware of much of the information we assume and correlate and infer on a day to day basis. It is completely staggering. At tesla, They had to write code for object permanence, which is a cognitive milestone for children at around 8 months. Think about that a minute. We have software that can control a car, it can stay between the lines and avoid other cars it can sense with its sensors, but it is less than 8 months old cognitively until pretty recently. Some parts it’s very advanced. It has lots of data, but it’s missing that a car moving behind another car is still there even if it can’t detect it with cameras or radar or LiDAR.

It is the same with Ai art. It doesn’t know reality. It didn’t grow up with it. It didn’t do head, shoulders, knees, and toes. It doesn’t know what a tool is. Even with millions of pictures, if they aren’t very explicitly labeled it’s going to give you the six fingered man with a spear as a tail.

it’s incredibly powerful, but just so naive at this point. It will become better. It will overtake humans at some point on many things. We may even teach it abstract notions and it will understand and strive towards those goals. The ramifications of that are staggering and are barely even being though about.
 
My "power" meaning I performed a standard clerical function when you broke a prior established ruling? sure.
Dude, Cure Light Wounds, Guidance and Blessing are still pretty big deals, yanno? That *and* chain mail? Nuthin standard aboot it!
 
ultimately, this is irrelevant, as computer vision is a thing and works.
Not totally. You're training your model on several different artists at the same time using yhe input from all. Visual or not its still different. Humans just don't learn in that way nor are they as efficient.
 
We spoke about this last night at my game. This was the sorta-consensus, as much as we ever agree on anything:

It's 100% A-OK-fine to use AI art to enhance a game we're playing. To give an idea of tone, or mood or whatever. Even to make on-the-fly pictures of stuff that's happening, or just happened, silly or serious. Like postcards. We've been doing that with my BW game and my mate's PF2 game and it has definitely been an enhancement.

It's not-fine to use AI to enhance the RPG games a couple of us have been working up, as it cuts real RPG artists off at the knees, is unethical, and is very much against the current realpolitikinternetinthotingeist.
 
People learning and creating art and computers doing the same thing are different things. No value judgement inherent there, but some truth IMO.
 
People learning and creating art and computers doing the same thing are different things. No value judgement inherent there, but some truth IMO.
Like I've been trying to art since I knew what it was. I am fucking un-artable. You know the only arty-thing I ever arted? Could produce facsimiles of, and riff off of, the art done by that bad-porn-arty-D&D guy you can't mention anymore. Zak Badguy. That guy. That's my fucking art legacy. I can art like Zak Badguy. Might as well babysit like Woody Allen.

But Skynet? Feed Skynet enough art and it will art better than all the artists. For free, or yanno, five bucks a month. That's the difference.
 
Like I've been trying to art since I knew what it was. I am fucking un-artable. You know the only arty-thing I ever arted? Could produce facsimiles of, and riff off of, the art done by that bad-porn-arty-D&D guy you can't mention anymore. Zak Badguy. That guy. That's my fucking art legacy. I can art like Zak Badguy. Might as well babysit like Woody Allen.

But Skynet? Feed Skynet enough art and it will art better than all the artists. For free, or yanno, five bucks a month. That's the difference.
See, I don't have any issue with people using AI to art stuff. That said, I also found an artist and paid them for the work they did for my current RPG project, so there are some lines and whatnot.
 
Status
Not open for further replies.
Banner: The best cosmic horror & Cthulhu Mythos @ DriveThruRPG.com
Back
Top