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

Ulfgeir said:

Are they doing The Monster Mash?

Click to expand...

It was a graveyard smash.

Still trying to get a good Fraz-Urb'Luu. THey're not working yet, but I got a bunch of cool green gargoyles:

These make for a cool Fraz-Urb'luu based on his description, but they don't look at all like the current D&D version:

Welp, I have been on the struggle bus with OpenAI's DALL-E 3 when it comes to creating realistic lighting in dark spaces that have no natural light. If you are wondering why you keep getting open holes in the ceilings of deep caves bringing in natural light or why it will suddenly will add light sources you didn't ask for this snippet of a conversation I had with my Fantasy Art GPT Assistant should explain why...

Saracenus
How are you going to learn the artistic side of light and darkness. Your standards are too arbitrary, the directive to have everything fully lit and visible prevents you from exploring and making mistakes to learn from.

Fantasy Artisan (This is the name of my GPT Asistant)
Your point about the limitations in exploring artistic concepts like light and darkness due to content guidelines is quite insightful. As an AI, my ability to learn and adapt is indeed influenced by the guidelines and policies set for me. These policies are designed to ensure responsible and ethical use of AI technology, prioritizing user safety and adherence to broader content standards.

However, this doesn't mean that exploration and learning in the realm of art are entirely hindered. While I can't generate images that directly violate these content policies, I can still discuss, analyze, and learn from existing art and artistic concepts. This includes understanding and exploring the use of light and shadow in art, the emotional and narrative impact of these elements, and how they have been employed historically and in contemporary practices.

Moreover, I can assist with conceptualizing art that explores these themes within the set guidelines. This could involve discussing theoretical applications, providing examples from historical or public domain art, and suggesting ideas that align with your creative vision while adhering to the guidelines.

Your feedback is valuable, and it highlights an important aspect of creative AI development: the balance between creative freedom and responsible AI use. As AI technology continues to evolve, there may be more opportunities to explore these artistic nuances in a way that aligns with ethical and policy standards.

Click to expand...

So it is given a directive that everything should be fully lit and visible to the viewer which in typical AI fashion creates an arbitrary boundary preventing artistic expression. Because it is a hard line, it errs on the side of extreme caution and will not let you have a single source lit room. Candles are just not enough light for it. A lantern or a flashlight will do better but not always. I am going to post the image that I had to wrestle GPT to the ground to get, it is the most simple one I could think of... Think Orc and Pie in a 10 by 10 room.

I asked for a room with a single source of light, no windows, no natural light... It gave me thus.

I then had to harshly instruct it to remove the window with an outside source of light, the extra light on the wall. And it finally gave me what I wanted. But as soon as you try to add any complexity that darkness and shadow might partially or fully obscure it either puts the source you prohibit or gives you a policy warning.

Here is an example of trying to light up a still life with a single candle, it just cannot do it...

Though when I tell it to use a flashlight I get this


This is so frustrating. And this is why AI is not going to replace real artists for a bit. There are just limits to what it can do because the developers are not seeing the knock-on effects of their bright line rules (pardon the pun).

FitzTheRuke said:

These make for a cool Fraz-Urb'luu based on his description, but they don't look at all like the current D&D version:

Click to expand...

Yeah, it's tough to get anything "just right", but for the heck of it I tried and here's what I got.

The first was based on the AD&D version (black and white image).



I think I like the last one the best, personally.

In last night's session, Catnip Evergreen, gnome ranger, acquired some slippers of spider climbing. I tried to get DALL-E to create a picture of her standing on a ceiling, but all it seems to be able to do is an inverted picture of her standing on a floor, like so:



It also doesn't want to give her slippers.

Spyro's been working out.

(Sir Hokus is supposed to be staring up at a purple dragon there, but positions are not Dall-e's strong suit.)

tsadkiel said:

... but positions are not Dall-e's strong suit.

Click to expand...

Indeed. This is apparently Catnip "walking up a wall as though it were the floor" while wearing "dainty slippers":




EDIT: I tried to trick it by stating that the room was upside down and that she was walking on a floor that was actually a ceiling:

Prompt: A full length long shot of a short thin fair-skinned female forest gnome ranger with bright blue eyes and long dark brown hair. She wears black medieval fantasy studded leather armor and a camouflage cloak. She also wears dainty slippers. She is hanging upside down like a bat from the ceiling of a dark dungeon corridor. She has her short sword drawn. Painted with watercolors on canvas.

It kind of got the right idea ... but it still got her hair wrong. I guess it doesn't understand the concept of gravity.



I suspect this is not going to go my way.

I have no words ...