Prompt Engineering Background
Technical Guide

The Ultimate Guide to Stable Diffusion Prompts

Author Magnuto Engineering
• January 22, 2025 • 20 min read

Contents

Prompting successfully in Stable Diffusion (SDXL) or Magnuto requires understanding how the CLIP model interprets text. It's not about writing a story; it's about stacking tokens.

1. Anatomy of a Perfect Prompt

A high-quality prompt typically follows this structure:

[Subject], [Action/Context], [Art Style], [Lighting], [Camera/View], [Quality Boosters]

Example:

A cyberpunk samurai standing in neon rain, cinematic lighting, volumetric fog, 8k resolution, unreal engine 5 render, photorealistic, wide angle lens

2. Lighting & Atmosphere Keywords

Lighting is what separates a flat image from a "wow" image. Use these tokens:

Cinematic
  • volumetric lighting
  • god rays
  • rim lighting
  • subsurface scattering
Natural
  • golden hour
  • overcast
  • soft shadows
  • natural light

3. Camera Angles & Lenses

Control the composition by specifying the camera gear. SDXL understands photography jargon perfectly.

  • Close-ups: macro lens, extreme close-up, 100mm lens
  • Portraits: 85mm lens, f/1.8, bokeh, depth of field
  • Landscapes: wide angle, fisheye, drone view, aerial photography

4. The Universal Negative Prompt

Copy and paste this into the "Negative Prompt" box in Magnuto to instantly improve quality by removing common artifacts.

ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, extra limbs, disfigured, deformed, body out of frame, bad anatomy, watermark, signature, cut off, low contrast, underexposed, overexposed, bad art, beginner, amateur, distorted face

Test These Prompts Now

Magnuto's engine is optimized for these exact keywords.

Open Generator