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

[Parmandur]

mexican soap opera TV show of an Orc Lady, green-skinned blue-haired pig-faced beautiful tall strong powerful feminine graceful wearing a modesr well-fit 18th century military uniform with a large sword on her side, inspecting a platoon of uniformed green pig-faced Orc troops, troops are in a blasted landscape with explosions in the distance on a sunny day with smokey haze in the air. recorded on digital with colors corrected digitally

 

[tsadkiel]

In which I attempt to join in the "two characters on a dinner plate" trend after everybody else has finished with it.

 

[Saracenus]

So are you suggesting this wording of a prompt will more consistently get us a desired result?

Not what I was intending, no*. The originator of that prompt structure is trying to give GPT4 and DALL E structure so that it doesn’t “freelance” a prompt. It is meant to go with the rest of the instructions. However, this structure does lend itself to cleaner results because it reduces the complexity of a prompt without making it meaningless.

Think of the training data that the AI has as the entire Universe (and in a sense as far as the AI is concerned it is) and if you just feed any old prompt it will try and use things that may not have any relevance to what you ask for. The prompt structure is a means of bounding the possibilities to say a group of (or a single) galaxies. You could dial it down to just a solar system but you have to be careful that you don’t get so specific that it becomes more of a hindrance that a help.

What I am working with is a version of DALL E that will create my prompts from what I tell it do, but the custom instructions will make it ignore anything that is not bound by its parameters. So for instance I could have a parameter that specifies the size of my image ”canvas” and any size I specify in the Natural Language prompt would be ignored because of the prohibition in the custom instructions.

Those of you working in Bing Image Creator do not have that guard rail in place, you have to be more careful what you ask it to do because all it has is what you tell it. The structure of the prompt just makes it easier to get consistent results.

Bing chat on the other hand you could cut and paste in the structure of the prompt and some other doos and don’ts and then the prompt it would help you create would be similar to the ones I am.

TL;DR It should help, but it is just one part of what I am doing.

* Really I was just showing you all what I am working with and hope it helps everyone. It also shows the power of the OpenAI version of DALL E vs the one that is being used with Bing. I cannot claim complete understanding or mastery of this, but I am finding that Custom Instructions are a powerful shortcut to get to what I want.

 

[Parmandur]

1980's AD&D Rulebook cover of an Orc Lady, green-skinned blue-haired pig-faced beautiful tall strong powerful feminine graceful wearing a modesr well-fit 18th century military uniform with a large sword on her side, inspecting a platoon of uniformed green pig-faced Orc troops, troops are in a blasted landscape with explosions in the distance on a sunny day with smokey haze in the air. Painted and printed onto durable hardcover, aged and sitting on a coffee table.

 

[Parmandur]

a oil painting of an Orc Lady Druid, green-skinned blue-haired pig-faced beautiful tall strong powerful feminine graceful wearing a well fit buckskin jumper over a loose red button-up blouse with a lot of jewlery and magic charms, drawinf frsh runes on an ancient standing stone, part of a circle of standing stones in a muggy marshland on an overcast morning. painted on a canvas.



1950's technicolor hollywood epic

 

[Saracenus]

So here is some images I have made playing with the size of the image... I wanted something that would work as a smartphone background (specifically for iPhone) but it told me the typical size of images for such devices were not a parameter I could use (mostly because I was using physical printing terms and DALL E uses pixels, it suggested that I try 1024 by 1792 pixels.

These were all generated from a suggested prompt that my Custom Instructions (CI) generated, the prompt was "An enchanted forest with trees that have glowing fruits." It took that and generated the above using the CI parameters.

GENERATED PROMPT FROM IMAGE 1: An enchanted forest scene showcasing towering trees with bark that glimmers softly and branches laden with luminous, multicolored fruits that cast a gentle glow on the surroundings. A full-body portrait of a woodland creature, perhaps a deer with crystalline antlers, stands amidst the magical flora, its eyes reflecting the spectral light of the fruits. The ground is carpeted with a mosaic of flowers that release a faint, bioluminescent radiance. The atmosphere is tranquil and saturated with the scent of jasmine and lavender, evoking a sense of wonder. The setting is at twilight, with the last rays of the sun filtering through the canopy, intertwining with the enchanted glow of the forest.

The last two images could be from the Feywild or a Dreamscape.

 

[pukunui]

My unnamed elf lady in white, acrylic painting on pottery:

And as an acrylic mural on an exterior wall:

And as a Roman mosaic:

And on Minoan pottery:

 

[FitzTheRuke]

 

[Scribe]

 

[FitzTheRuke]

Rust Monsters