Using ChatGPT to create prompts for Stable Diffusion: Part II
In the first part, that I didn’t name “Part I”, we explored how to use ChatGPT to create simple prompts and feeding them into Stable Diffusion, or any SD model.
This time we’ll be going a step further and ask ChatGPT to make a description, from that description extract the main concepts and keywords into a list, and then conveniently return that list as words separated by commas.
The Prompt
We start by asking ChatGPT to “Describe an imaginary God in its physical appearance. It has to be original and very detailed. Write in the present tense.”
In this case ChatGPT returned:
“The god is a towering figure, standing at over 50 feet tall. It has a humanoid shape, with four arms and a pair of wings on its back. Its skin is a deep, shimmering blue that seems to shift and change color depending on the angle and the light. Its eyes are bright golden orbs, and they seem to glow with an inner light.
The god’s head is adorned with a crown of flickering flames, and its long, flowing hair is made of writhing tendrils of energy. Its four arms are each ending in a hand with five fingers, each one tipped with a sharp claw. The wings on its back are massive, with feathers that shimmer and shift like the god’s skin.
The god exudes a powerful aura of majesty and strength, and it seems to radiate an energy that is both calming and invigorating. It is a being of great power and wisdom, and it is revered and respected by all who encounter it.”
From here we could already take some ideas and build or own prompt to start creating this imaginary God, but ChatGPT can go a step further.
We just have to instruct it to, based on the description, to extract the main concepts and keywords in finds in the text it just generated.
Telling ChatGPT to “Extract the key concept and keywords to a list” returns:
Here are the key concepts and keywords that I identified in the description of the imaginary god:
towering figure
humanoid shape
four arms
wings
deep, shimmering blue skin
bright golden eyes
crown of flames
energy-filled hair
sharp claws
massive wings
shimmering feathers
powerful aura
majesty and strength
calming and invigorating energy
great power and wisdom
revered and respected
Now we just need to go full lazy mode, and ask ChatGPT to turn this list into a SD prompt style by asking it to “Pass the list to words separated by commas”, that returns:
towering figure, humanoid shape, four arms, wings, deep, shimmering blue skin, bright golden eyes, crown of flames, energy-filled hair, sharp claws, massive wings, shimmering feathers, powerful aura, majesty and strength, calming and invigorating energy, great power and wisdom, revered and respected.
The results now will depend on what software and model you using to generate your images.
The cover image for this photo was the result of just pasting the returned list of words in Diffusion Bee with the openjourney-v2-unpruned_fp16 model, and adding at the beginning “an award winning photo of a”
Using other model, in this case 3D Model (Redshift v1) I was able to produce the above image, that captures the essence of the prompt given by ChatGPT. The software used was Draw Things
Conclusions
If you are somehow stuck in your creative process, you can use ChatGPT to give you some directions. With that information you can experiment with different methods, models, and even changing some of the keywords. The possibilities are endless, and these ChatGPT prompts can help understand what works, what doesn’t and, most of all, have fun experimenting.
If you enjoyed this post please follow me, and leave what you were able to create in the comments.
