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

I’ve tried all sorts of descriptive prompts but it usually wants to give them a nose, teeth, a beard, five fingers, five toes, etc.

Whenever I specify that it has “root-like cords beneath smooth metal body plating” or the like, it can give me something decent for the upper body but will then give me roots for feet or some such silliness.

D&D spell illustration challenge, day 12: Clone


Since clones can be created younger, they are a decent, if underused, path to eternal life. Lichdom is all cool and gothic, though.

Prompt: In a shadowy chamber lit by flickering candles, a large glass sarcophagus stands upright, filled with a thick, amber fluid. Suspended inside is a pale, hairless body, eyes closed, arms folded as if sleeping. Runes glow faintly on the metal frame, and arcane tubing snakes from the base into the stone floor. Standing beside it is a mage: a red-haired wizard with a flamboyant moustache and long hairs, dressed like a magical musketeer: plumed hat, embroidered doublet, and a confident stance. He watches his clone with calm pride, one hand resting on a staff ending with a glowing blue gem, the other holding a leather-bound grimoire.

Commentary: True Resurrection one spell level earlier for 4% of the monetary cost, at the cost of a 120 days setup. Getting the cubic inch of the original body might be painful, but it can double as a weight-reducing surgery. Take that, clerics. There is no reason high level adventuring party wouldn't have a clone bank in one of their extraplanar emergency hideouts. That must be the reason why GMs don't run high-level adventures.

pukunui said:

Has anyone found an AI image generator that can reliably produce decent warforged results for free? I'm struggling to find one that can do it. Even leonardo and mage space - which allow you to upload a reference image and allow for negative prompts - can't produce one.

Click to expand...

No luck here. It's a pretty specialized concept that warrants more explanations... and even then it's complicated. I gave up because my approach of telling the model about a steampunk robot didn't work out.



I like it, even if it's not a warforged.

D&D spell illustration challenge, day 13: Cloudkill.


When your neighbours violate the HOA rules on lawnmowing, it's proper to retaliate by a combination of Arcane Lock and Cloudkill.

Prompt: A large suburban detached house with clean white siding and wide porch. Sickly green-yellow fog curls from the chimney and under the window frames. The door is barred by a magic wall of oange glowing energy. A smiling African-inspired female witch-doctor calmly gestures downward with one hand, casting a spell, surrounded by magical energies. Her cloak flaps in the breeze. On the lawn, a small red tricycle lies tipped on its side.

Commentary: the tricycle isn't on the side. Also, my magic users tend to be slightly morally-impaired.

D&D spell illustration challenge, day 13 again: Cloud of Dagger (I had missed it)


A single, young, enterprising elf wizard can easily outcompete the butchering guild.

Prompt: An anime style drawing: in a sunlit medieval village, a cheerful butcher chats with townsfolk beside his market stall, cheerfully weighing sausages. Just beyond, in a modest garden pen, a lone cow stands quietly among flowers, near a young elven mage raises a hand, facing the cow, concentration etched on his face. Around him spins a forming cloud of daggers — glinting blades orbiting in a tightening storm.

Commentary: I wanted a change of style. Anime elves have silly, long ears. Also, it takes as little as 112 cows to go from level 3 (the earliest you can cast Cloud of Dagger) to level 5 (where you get access to fireball and can also replace the roaster's guild).

Nightfly said:

Niji is Midjourney's anime engine. A smidgen of anime works well for D&D characters -- it emphasizes dramatic poses and strong styling over realism, which is good. Of course, too much anime input and you end up with saucer-eyed cartoons.

Click to expand...

That bit about Anime poses and overall scene composition is a good point! A bit of weighting/influence to have it engage in that aspect w/o going into the aesthetics/style could be really fruitful.

pukunui said:

I’ve tried all sorts of descriptive prompts but it usually wants to give them a nose, teeth, a beard, five fingers, five toes, etc.

Whenever I specify that it has “root-like cords beneath smooth metal body plating” or the like, it can give me something decent for the upper body but will then give me roots for feet or some such silliness.

Click to expand...

The AI seems to think the "Warforged" are "Orcs".

Here the prompt is: "full-body, heroic, clean medievalarmor, inside wearing it is a treeroots robot".






Sometimes the images want to moss over it, kinda like it.






Here the prompt is: "full-body, heroic, treeroot robot wearing shiny armor". The images are only occasionally helpful. (Heh, "rootlike cords" ended up a great horror monster.)

Yaarel said:

The AI seems to think the "Warforged" are "Orcs".

Click to expand...

Yeah, they're not familiar with the term "warforged" at all. If I don't provide any additional descriptions, it tends to just give me an armored knight when I used that term.

I tried uploading an official warforged image to ChatGPT and asked it to create a prompt for me, but that turned out not to be particularly helpful. (Although it is familiar with Eberron and warforged, it wasn't any better at creating an image of a warforged than any other AI image generator.)


For the record, I've been trying to generate an image of Smith, the warforged cleric of Onatar who tends the temple at the Pool of Onatar's Tears in Blackbones (beneath Sharn). There's no official image, nor even an official description, so I went with a "typical" warforged wearing a dark blue cloak with gold trim. I said he would be holding a smith's hammer and would have the symbol of the Sovereign Host emblazoned on his chest plate. This is what ChatGPT gave me:



Not bad - but it's not a "typical" warforged.

This was the closest I got using leonardo.ai:



I can't seem to find the images magespace generated for me, and it won't let me create anymore without paying, so never mind that one! Suffice it to say they weren't any better than anything else I've shared here.

A while back, when I was converting a 3.5e Dungeon adventure to run for my Eberron campaign, I opted to set it in the Eldeen Reaches. I toyed with the idea of making one of the local NPCs in the adventure be a warforged. In some cases, I had the idea that they might have customized themselves with more wood rather than metal.

Here are the various results Bing's Image Creator gave me, based on variations of this prompt: A full length portrait of a tall medieval fantasy robot with root-like cords beneath shiny smooth armor plating. It has a hinged toothless mouth and eyes made of blue crystal. A rune is etched into its forehead plating. It wears a forest green cloak with a circular wooden brooch. It holds a sword and a shield as it patrols. Backdrop is a blurred bucolic forested medieval riverside village. Drawn in detailed modern comic book style.



As you can see, some of them aren't too bad! These two are probably the best of the bunch:



EDIT: Sadly, I seem to be unable to get it to give me noseless robots now. I guess the training data has changed too much or something.

@pukunui

Your Warforges are great.

I wander if you can train the AI by saying in the prompt, "a warforge is ...". Maybe later its patterns will start to associate the term.