Young Jedi Adventures: Encounter at Midjourney Station

So, I’ve been watching Young Jedi Adventures as Jedi Survivor has left me craving more details about the High Republic era. Even though I’m outside the target audience I’m really loving the characters and the look and feel of the setting.

Unfortunately, the show still leaves you wanting more in terms of art. So I thought it’d be a great chance to experiment with Midjourney. If you haven’t yet heard of it, Midjourney is a generative artificial intelligence program. It generates images from natural language descriptions, called “prompts”.

So firstly, the results (Original left, AI right):

Lys Solay

Lys Solay AI

Method

Prompt: A female Jedi Knight, with two buns of popping purple hair. She has lucid blue skin and is holding a silver lightsaber with a radient purple blade. She is defending her animal friends from bandits.

Lys Solay WIP AI

This then took me to the image above (left), which I blended with an image of Riyo Chuchi a Pantoran from the Clone Wars TV show (right) to get the final image.

Nubs

Nubs AI

Method

Prompt: A photo of a blue bear, wearing a simple white poncho. He is holding a blue lightsaber while looking at his friend. 

Kai Brightstar

Kai AI

Method

Prompt: A young black enthusiastic man with short hair, he is wearing white robes with gold trim while holding a lightsaber with a green blade. 

Midjourney struggles:

  • Non-typical colours, it was either purple lightsaber or blue skin, the system couldn’t get both. When it came to fur I was unable to force it to correctly colour Nub’s face blue.
  • Lightsabers were always shortened to fit into the frame
  • Blue skin in general was tricky, it couldn’t keep it consistent for the whole picture.
  • Hands, while they do now have the correct amount of fingers, still come in some horrific poses.

Midjourney highlights:

  • Lighting is actually incredible, the fact that it understands that lightsabers glow is superb
  • Ease of use, these pictures did not take long to generate
  • When something is in the dataset and manages to ‘fit’ the prompt the system does a great job of extrapolating from a very limited prompt