I’ve made tens of thousands of AI generations on Midjourney, Stable Diffusion, Dall-E, and Firefly. Almost always, I get asked, “How did you prompt that?” The answer isn’t straightforward. I rarely rely on Gen AI’s slot machine-like output to give me a random image. Using tools like ControlNet, IP-adapter, and custom LoRAs (in Stable Diffusion), I’ve learnt how to somewhat control the beast that is AI to get the output I want. Sometimes this work is very manual. That being said, sometimes I am lazy and just want a cool-looking image. And prompting is hard.
So I wrote this “super prompt” that gets GPT (or Claude/Llama/Gemini) to help me craft a good prompt based on my idea. It considers things like lighting conditions, weather, environment, colour palette, hues, mood, tone, visual theme, artistic style, etc. — all based on a figment of your idea.
Just copy and paste the entire text found in this link (works best with GPT-4, Claude 3 Opus, or Llama 3 70b), and then the LLM will prompt you for your idea.
The Evolution of AI Prompting
When I first started with AI image generation, I approached it like a search engine. Type in what you want, hope for the best. But after thousands of generations, I realized that effective prompting is an art form that combines technical knowledge with creative intuition.
Understanding the Tools
Each AI platform has its own personality:
- Midjourney: Excels at artistic and stylized imagery
- Stable Diffusion: Offers the most control through extensions
- DALL-E: Great for photorealistic and concept art
- Firefly: Adobe’s offering with commercial safety
The Super Prompt Philosophy
My “super prompt” approach breaks down the creative process into systematic components:
- Visual Foundation: Lighting, composition, perspective
- Aesthetic Elements: Color theory, mood, atmosphere
- Technical Specifications: Style, medium, quality markers
- Creative Direction: Artistic references, cultural context
Beyond Random Generation
The key insight is moving beyond random generation to intentional creation. By understanding how these models interpret language and visual concepts, you can craft prompts that consistently produce results aligned with your vision.
This systematic approach transforms AI image generation from a game of chance into a reliable creative tool.