D&D General DALL·E 3 does amazing D&D art

[Kichwas]

Do you really think the 2024 core book art is crap or do just not like the style or theme of it? I can not like an image, an artist, or style and still realize it is not crap. It is just something I don't like.

Though for me, whether I like it or not is completely besides the point when it comes to AI art.

I don't think they're using AI Art in that are they?

But one side impact of all these new training models is a lot of actual artists are now getting accused of doing Ai Art by the 'anti AI' folks. The training models are just using existing styles, and they happen upon your style, suddenly everyone who can't think rationally is all over you.

The long term reality is that the AI tools will just get rolled into art the same way cameras and then photoshop, blender, and similar tools did.

Like with cameras - some people will make quickies, and others will master ways to do creative things with it, and people way more skilled than me will do what I do, only much better: mixing hand made work and AI work into one final piece.

Take my process there, and have a skilled artist do that, and the results will be a whole new form of art that some will dismiss as just 'cheap trips', 'snapshots', and so on - like some do with photography. But those kinds of artists that can master the new medium and mix it with their existing skills - they're the future of all of this.

 

[dave2008]

I don't think they're using AI Art in that are they?

No. @ezo said they like AI art better than the "crap" in the 2024 books. That is what I was responding to.

 

[evilbob]

In my limited experience, some AIs tend to especially draw women with scantily clad "armor" - even when you put in prompts like "woman wearing heavy plate mail." It's pretty bad.

Absolutely had to share this though, which is what one of my players spat out for a harengone bard for the Vecna campaign

 

[Thomas Shey]

In my limited experience, some AIs tend to especially draw women with scantily clad "armor" - even when you put in prompts like "woman wearing heavy plate mail." It's pretty bad.

I kind of agree with some other posters this is probably an artifact of the majority of depictions they've been "taught" on doing the same. They've, effectively, learned bad lessons here.

 

[pukunui]

In my limited experience, some AIs tend to especially draw women with scantily clad "armor" - even when you put in prompts like "woman wearing heavy plate mail." It's pretty bad.

I've had minimal issues getting Bing's Image Creator to portray armored women:

 

[evilbob]

I've had minimal issues getting Bing's Image Creator to portray armored women:

ChatGPT works well. They have it very locked down, from my limited playing with it, including from things like copywrited characters.

And yes, it's because of the images they've been trained on. Which means the "It's bad" was more an indictment of most fantasy art than the AI...

 

[cbwjm]

I asked chatgpt if it knew what I meant if I said "no boob armour" and it did. Admittedly this isn't plate armour which is notorious for boob armour but still, I think these look pretty cool. Really like the consistency here.

 

[Kichwas]

And yes, it's because of the images they've been trained on. Which means the "It's bad" was more an indictment of most fantasy art than the AI...

A bit of both I think. Especially when it comes to local installs of AI art tools.

Bing and ChatGPT have had people put in content filters to moderate what they deliver and to ensure results.

When you do this stuff locally and go out there to grab training files - the top sites hosting them just let people moderate and filter themselves. So every now and then you'll grab "SFW skin tones" training, and for some reason every so often a random NSFW gender-blended undessed furry (or something worse) pops into the images and you have to figure out that one completely unrelated training model was why, then toss it in the bin.

It's what makes a new model nice to get a hold of. Folks haven't yet fully messed up the addons for it.

Over in the Bing and ChatGPT world - they've got very strong filters so you at least won't break into the 'weird' unless that was your goal.

 

[Kannik]

In my experience, to significantly alter the style of the image with an image-to-image workflow, you need enough denoise that only the basic composition of the image remains. I have had good success using i2i with first generating a basic outline with a model very apt at understanding my prompt and then using i2i with a less precise, yet more pretty model, but I don't think starting with flux and i2i in a model with the right style would work that well. The fine details would be lost during denoising.

Aye, I was thinking mostly from the perspective of transferring over a gross pose (either through i2i or control net) rather than details or intricate bits. Maybe Flux is a bad example/base model to use, unless there are some variations of it that generate really good at poses/anatomy.

 

[ezo]

Do you really think the 2024 core book art is crap or do just not like the style or theme of it?

There is a lot of art I don't care for personally, but can appreciate how others might. There is little artwork in 2024 that I can understand how people would like it at all. But as they say, one man's gargabe is another man's treasure.