DEV Community

七品
七品

Posted on

The Prompt Engineering Guide: FLUX vs Midjourney vs Stable Diffusion vs DALL-E 3

The Prompt Engineering Guide: FLUX vs Midjourney vs Stable Diffusion vs DALL-E 3

If you've used AI image generators, you know the frustration: the same prompt gives wildly different results across models. What works on Midjourney looks terrible on Stable Diffusion. What works on DALL-E 3 doesn't work on FLUX.

I've spent months testing across all four major models. Here's what I've learned about crafting prompts that work — and the patterns that transfer across models.


The Universal Prompt Architecture

Every great prompt follows this structure, regardless of model:

[SUBJECT] + [ACTION] + [ENVIRONMENT] + [LIGHTING] + [STYLE] + [MOOD] + [COMPOSITION] + [COLORS] + [QUALITY]
Enter fullscreen mode Exit fullscreen mode

Component Breakdown

Component Purpose Example
Subject Main focus "A Siberian husky"
Action What they're doing "running through snow"
Environment Background "in a dense pine forest"
Lighting Light source "golden hour, soft volumetric"
Style Aesthetic "cinematic, photorealistic"
Mood Emotional tone "peaceful, majestic"
Composition Framing "wide angle, rule of thirds"
Colors Palette "cool blues and warm oranges"
Quality Technical "8K, highly detailed"

The secret most people miss: lighting alone can transform a mediocre prompt into a stunning image.


Model-Specific Strategies

FLUX — Best for Photorealism

FLUX excels at hands, faces, and text rendering. It's the best choice for photorealistic results.

Best prompt structure for FLUX:
cinematic photograph of [subject], [action], [environment], [lighting], 
ultra-detailed, 8K, shot on Sony A7R IV, 85mm lens, f/1.8, shallow 
depth of field, natural textures
Enter fullscreen mode Exit fullscreen mode

Strengths: Anatomy, text in images, complex compositions
Weaknesses: Less style flexibility than Midjourney

Midjourney — Best for Artistic Work

Midjourney shines at stylized, creative, and artistic outputs. Use parameters for control.

Best prompt structure for Midjourney:
[subject] [action], [environment], [lighting], [style], [mood] 
--ar 16:9 --v 6.1 --s 250 --style raw
Enter fullscreen mode Exit fullscreen mode

Key parameters:

  • --ar — aspect ratio (16:9, 4:3, 1:1, 9:16)
  • --s — stylization (0-1000, higher = more artistic)
  • --style raw — more photorealistic, less "beautification"
  • --no — what to exclude

Stable Diffusion — Best for Control

SD gives you the most control through CFG scale, samplers, and LoRA models.

Best prompt structure for SD:
masterpiece, best quality, [subject], [action], [environment], 
[lighting], [style], [colors]

Negative prompt: ugly, tiling, poorly drawn, deformed, blurry, bad anatomy
Enter fullscreen mode Exit fullscreen mode

Key settings: CFG scale 7-12, DPM++ 2M Karras sampler

DALL-E 3 — Best for Complex Scenes

DALL-E 3 understands natural language best. Write in complete sentences.

Best prompt structure for DALL-E 3:
[Detailed natural language description of the scene, including 
subject, action, environment, lighting, style, and mood. 
Be specific about composition and details.]
Enter fullscreen mode Exit fullscreen mode

The Lighting Cheatsheet

The single highest-ROI change you can make to your prompts:

Lighting Type Effect Prompt Keywords
Golden Hour Warm, soft, directional "golden hour, warm sunlight, long shadows"
Blue Hour Cool, moody, twilight "blue hour, twilight, cool tones, dusk"
Rembrandt Classic portrait "Rembrandt lighting, triangle of light"
Backlight Silhouette, rim light "backlit, rim light, dramatic silhouette"
Soft Diffuse Even, flattering "soft diffused lighting, overcast, softbox"
Hard Light Harsh shadows, dramatic "hard light, harsh shadows, high contrast"
Volumetric Light rays "volumetric lighting, god rays, crepuscular"

Pro tip: Always specify lighting. It's the most underused prompt component.


Common Mistakes and Fixes

Mistake Symptom Fix
Overloaded prompt Weird combinations 1 subject + 3-4 max descriptors
Conflicting styles Mushy results Pick ONE style and commit
Wrong aspect ratio Cropped subjects Specify --ar or "wide shot"
Too many quality words No effect "8K, highly detailed" once is enough
Missing lighting Flat, boring Always specify lighting source

100+ Templates at Your Fingertips

I've compiled all of this knowledge — and much more — into a AI Prompt Engineer skill for Claude Code, available for $10.

It includes:

  • 100+ template prompts organized by category (portraits, landscapes, sci-fi, product shots, etc.)
  • Model-specific prompt structures for FLUX, Midjourney, SD, and DALL-E 3
  • Complete lighting reference with keywords
  • Style reference guide (20+ styles with model recommendations)
  • Prompt testing workflow for iterative refinement
# Installation is one command:
mv ai-prompt-engineer ~/.claude/skills/
# Then ask Claude: "Create a cinematic portrait prompt for FLUX"
Enter fullscreen mode Exit fullscreen mode

Better prompts, better images. 🎨

Top comments (0)