Who is Drax Daren?
I’m using Stable Diffusion & an open-source LLM to create a new character in the Star Wars universe. Introducing Drax Daren, a traveling space photographer. He’s currently wandering around Mos Eisley, Tatooine.
Most days of my childhood was spent making Star Wars short films, playing with action figures, and building sets for my stories from cardboard. These generative tools are a natural extension of my imagination.
Stable Diffusion is an amazing, open-source tool for creating all sorts of AI-generated art content. I wanted to bring a storytelling perspective to my AI photography creation and craft a visually striking day-in-the-life photography series. Behind the scenes, candid, never before seen characters and Star Wars species.
Authentically “Star Wars” is the aim. I’ve chosen color palettes for each planet, narratives that fit Drax’s travels, and use the TV shows, images, and movies as reference for the characters and “vibes” of the planets.
There’s a catalog of mini-narratives assigned to each subject of the photograph with the intention of creating a deeper, richer experience to Drax’s life and work.
What is Prompting?
Using NLP (natural language processing) Stable Diffusion & MidJourney generate images based on a user’s description of that image. Some of the challenges of a successful prompt include:
- Crafting the Subject e.g. who, what, where, when
- Choosing the right Medium i.e. painting, photo, digital illustration
- The movement or style i.e. pop art, surrealism, hyper-realism
- Thoughtful emphasis of additional details with punctuation
- Negative Prompting is equally important and a way to refine what’s undesired in the image
The Midjourney/Image Diffusion AI tool "Prompting" allows users to input text prompts that guide the AI in generating images.
- Stable Diffusion w/ Automatic1111
- ControlNet in Automatic1111
- Photoshop for compositing & revisions
- Lightroom for coloring & finishing
- AutoGPT enables Drax’s decision making as an autonomous AI. It’s a Python program powered by GPT-4. He chooses his stories and who to follow
- Custom model training (LoRA) on Star Wars characters and camera
- Effective prompt design
- Maintaining a consistent visual aesthetic given how Diffusion models work
- Using new, open-source tools