π Prompt Structure
The best Midjourney prompts follow: [Subject] + [Environment] + [Style] + [Lighting] + [Camera] + [Mood] + [Quality tags] + [Parameters]. More specific subject descriptions produce more consistent results than vague ones.
βοΈ Stylize Parameter
--stylize controls how strongly Midjourney applies its aesthetic. Low (0β100): stays close to your prompt. Medium (250β500): balanced interpretation. High (750β1000): highly stylized, more artistic freedom. Default is 100.
π² Chaos Parameter
--chaos (0β100) controls variation between results. Low chaos: consistent, predictable outputs. High chaos: wild, unexpected interpretations. Use low chaos for product/professional work, high chaos for creative exploration.
π« Negative Prompts
Use --no to exclude unwanted elements: --no blurry, text, watermark, ugly, deformed. In Midjourney v6, negative prompts are more effective than in earlier versions. Always exclude "text" unless you want text in the image.
π Artist References
Adding "in the style of [artist]" or "by [artist]" is the fastest way to get a consistent aesthetic. Greg Rutkowski for epic fantasy, Artgerm for portraits, Simon Stalenhag for retrofuturism. Combine 2β3 artists for unique blends.
πΈ Camera Terms
"Shot on 35mm Kodak Portra 400" gives film grain and color. "85mm f/1.4 bokeh" gives portrait blur. "Shot on Sony A7R V" gives sharp digital. Camera and lens specifications dramatically change the feel of photorealistic images.
Each model has different strengths. Midjourney v6 produces the most aesthetically refined images and excels at artistic styles. DALL-E 3 (via ChatGPT) follows complex instructions most accurately and handles text in images well. Stable Diffusion (SDXL/Flux) is open-source, runs locally, and is best for fine-tuned control with LoRA models. Adobe Firefly is commercially safe β all outputs are licensed for commercial use.
For the best results, use this builder's model selector to switch prompt format β DALL-E 3 needs more descriptive natural language, while Stable Diffusion benefits from keyword-heavy prompts with weights like (subject:1.4).