D&D General AI Art for D&D: Experiments

2: Swords and daggers are often ok. they will have problems with specific types. Things like simple quarterstaffs are more hit-and miss. Spears are often very problematic (at least on Lexica.art). Bows are problematic if you know anything about bows (but then humans drawing them often tend to mess up as well when drawing someone actually using the bow, so not too surprising.)

I'd suspect you have the answer generally there; probably the material they've been trained on has enough decent samples of simple blades that it can spit out something decent, but it gets much more hit or miss with other weapons because the samples are less frequent and/or less good.

Things I haven't been able to get any engine to do is having a vampire actually biting someone else. Blood is questionable. and if you use an online engine, it will often refuse to draw something it thinks is "spicy".. Even if it is nothing in the description for that. And they sometimes react that way on a prompt that you have used before without problems. The free version of Chatgtp was really bad on that, and well I managed to get my account there banned. I think I figured out what it reacted to that got me banned. Never managed to get it unbanned, as the supportmails just seemed to go into the void, and no human looked at it.

That's of course an intrinsic problem; with the publically accessible ones there are going to be things that even people halfway trying to follow the law and/or ethics will have tried to block, and its going to be hit or miss how it reads those.
 

log in or register to remove this ad

No matter which engine I have used, I have found that you often need to generate a number of images to get something you like.
If you're looking for something extremely specific, yes. What's worked for me is letting go of the need for precision. It doesn't need to be exactly what I want. Close enough is good enough. I stopped fighting the generators a long time ago and pushed them towards an art style that can cover a lot of mistakes. A little abstract instead of going for realism and you're set. I'm happy enough with the results that I generally don't need more than 4-8 images to have 4-8 images that I really like. If you're wondering what I mean, just search this thread for my other posts.
And sometimes it will just plain ignore what you write. even if you are very specific.
Yeah. There's some kind of keyword nonsense going on that just stops certain things from working. I've tried to make WH40k-style orks, only for it to show me space marines with tusks. The only reasonable response, I think, is to shrug and move on. To me that's far better than trying to force the thing to do exactly what I want. It's just not worth the time.
 

No matter which engine I have used, I have found that you often need to generate a number of images to get something you like. And sometimes it will just plain ignore what you write. even if you are very specific.

1: It depends. The engines seems to prefer the person just standing there. And interaction with other beings are often so-so.

2: Swords and daggers are often ok. they will have problems with specific types. Things like simple quarterstaffs are more hit-and miss. Spears are often very problematic (at least on Lexica.art). Bows are problematic if you know anything about bows (but then humans drawing them often tend to mess up as well when drawing someone actually using the bow, so not too surprising.)

3: not sure I understand the question. I have found that different engines will react on certain keywords. So for example my Scion-character, she has intense emerald green eyes. If I don't specify anything else, then Lexica.art for example will then draw her in green clothing. Z-Image Turbo made her look Asian (here the trigger was the use of intense/emerald. if I didnd't include any of those words it worked. Funny thing was that when I asked it to also include another character that was a goth girl, it came out ok, even if those words were included.)

Things I haven't been able to get any engine to do is having a vampire actually biting someone else. Blood is questionable. and if you use an online engine, it will often refuse to draw something it thinks is "spicy".. Even if it is nothing in the description for that. And they sometimes react that way on a prompt that you have used before without problems. The free version of Chatgtp was really bad on that, and well I managed to get my account there banned. I think I figured out what it reacted to that got me banned. Never managed to get it unbanned, as the supportmails just seemed to go into the void, and no human looked at it.
Boy that censorship would bother me, I want my barbarian holding that severed head trophy, but its not a deal breaker.
 

I am thinking CharGen, pretty spendy. But it looks like I can generate map tiles with it, and consistent items and such. Looks like it combines the best of many worlds with several different engines, is that fairly accurate for anyone who knows?
 

Does the paid version have the sad dog issues or is it free to do whatever?

Sad dog issues? I don't know that reference.

If you mean what won't it let you do...

1. ChatGPT won't (and this has been the case before recent Musk AI issues) let you create an image from a photo of a real person. Not will it let you create a photo realistic image of a real person - even to the point of not creating photo realistic people in the background in case they look like a real living person.

2. Won't let you create something of an "adult" nature, which is fair, but it even stops at a plunging neckline or certain costumes, like a maid's outfit. Such things are deemed by the program as "suggestive".

3. Won't let you show someone running in panic/fear, even if it isn't photo realistic or facing the camera. Tried this with some Jurassic World art I was toying with and it's a big no no for some reason.

However, with the paid version, it learns better and can create/save a specific art style. I use this for my home brew campaign setting. Likewise, you can save specific characters to bring into art. The only issue I find is that it struggles with weapons, mainly axes, polearms and the like, where it cannot get the head of the weapon on at the right angle or similar.
 

2. Won't let you create something of an "adult" nature, which is fair, but it even stops at a plunging neckline or certain costumes, like a maid's outfit. Such things are deemed by the program as "suggestive".

This is what I mean. The program is told 'ball gown' and IT decides what to render. It then also decides if what it rendered, is appropriate, and the flags it and hides it or outright suspends the account.

Seems a bit suspect but hey.
 

This is what I mean. The program is told 'ball gown' and IT decides what to render. It then also decides if what it rendered, is appropriate, and the flags it and hides it or outright suspends the account.

Seems a bit suspect but hey.

A friend of mine wanted me to create a picture of her for a 1920's Call of Cthulhu one shot where she was playing the maid. It wouldn't let me use the photo so I had to ask it to create someone similar with a couple minor changes. Unfortunately it views French Maid's outfit as too suggestive.

I've never had it flag anything just say it can't do something, and if you ask for a reason it will explain the decision. As for suspending an account... I don't know but then I've never tried to make anything "adult" on there.
 

A friend of mine wanted me to create a picture of her for a 1920's Call of Cthulhu one shot where she was playing the maid. It wouldn't let me use the photo so I had to ask it to create someone similar with a couple minor changes. Unfortunately it views French Maid's outfit as too suggestive.

I've never had it flag anything just say it can't do something, and if you ask for a reason it will explain the decision. As for suspending an account... I don't know but then I've never tried to make anything "adult" on there.

I'm referring to the free version. Essentially it blocks the images, sometimes it even provides a warning on the prompt, I'd image French Maid would do it, and if you persist in trying to get around it, or it thinks you are, it blocks your access.
 

I'm referring to the free version. Essentially it blocks the images, sometimes it even provides a warning on the prompt, I'd image French Maid would do it, and if you persist in trying to get around it, or it thinks you are, it blocks your access.
I used the free version for a few weeks at the start of last summer but the lack of images generated in 24 hours was the turning point to subscribe.

That makes sense for blocking access.
 

I'm referring to the free version. Essentially it blocks the images, sometimes it even provides a warning on the prompt, I'd image French Maid would do it, and if you persist in trying to get around it, or it thinks you are, it blocks your access.
It also blocks things like "pulp" for some reason. I've tried to get cool pulp action stuff but it's always blocking that. I can get close if I describe it rather than try to use the shorthand of pulp. So weird.
 

Remove ads

Top