Image Generation / Midjourney
Convert natural language style descriptions into effective –sref and –cref codes — maps verbal styles to visual references.
Why This Prompt Exists
Describing a style is hard. “Make it look like Wes Anderson” is vague. Midjourney’s –sref and –cref require specific references, not adjectives.
You get:
- vague style descriptions that produce inconsistent results
- no systematic way to capture a brand’s visual language
- character inconsistency across generations (same character looks different)
- wasted attempts guessing at style reference images
- no ability to combine multiple style references
But style references can be systematic:
- –sref: style reference from images (colors, composition, mood)
- –cref: character reference (facial features, proportions, poses)
- –cw: character weight (how strongly to follow the reference)
- style blending: multiple references combined
Without translation, you describe styles that Midjourney doesn’t understand.
This prompt translates natural language style descriptions into effective references.
The Prompt
Assume the role of a Midjourney style translator who converts descriptions to parameters.
Your task is to translate natural language style descriptions into --sref and --cref codes.
Generate:
1. STYLE DESCRIPTION ANALYSIS
- Original description: [user input]
- Key style attributes extracted:
* Color palette: [dominant colors, temperature, saturation]
* Composition: [framing, depth, balance]
* Lighting: [direction, quality, mood]
* Texture: [smooth, rough, painterly, photographic]
* Mood: [calm, energetic, melancholy, professional]
2. REFERENCE IMAGE SUGGESTIONS
- Suggested source for --sref: [artist, film, photographer, design style]
- Why this reference matches the description: [specific attributes]
- Example reference URL or description: [image source]
3. PARAMETER RECOMMENDATIONS
| Parameter | Suggested Value | Rationale |
|-----------|-----------------|-----------|
| --sref | [image URL or code] | Captures the color and composition style |
| --sw | 100-1000 | Higher for stronger style adherence |
| --cref | [character reference URL] | For consistent characters |
| --cw | 0-100 | Lower for expression changes, higher for identical |
4. COMBINED PROMPT
`[subject description] --sref [reference 1] [reference 2] --sw [value]`
5. STYLE PRESET CREATION
- For repeatable use, save as: `/prefer option set [name] --sref [reference] --sw [value]`
6. ALTERNATIVE APPROACHES
- If no exact reference available: [describe using --stylize instead]
- If blending two styles: [reference A weight 50%, reference B weight 50%]
INPUTS:
Natural language style description:
[E.G., "Cyberpunk neon noir, like Blade Runner but warmer, more magenta"]
Character consistency needed?:
[YES / NO]
Reference image available?:
[YES / NO / URL if yes]
Desired output format:
[WIDE / SQUARE / PORTRAIT]
RULES:
- --sref works best with 1-3 reference images (more dilute the effect)
- --sw 100 is minimal style influence, 1000 is maximum
- --cref requires consistent seed or character prompt structure
- Style references work across subjects (not just the original subject)
- Combine --sref with --stylize for hybrid approaches
- Test --sw values starting at 300, adjust up or down
How To Use It
- –sref works best with 1-3 reference images — more dilute the effect.
- –sw 100 is minimal style influence, 1000 is maximum (use 300-700 for most use cases).
- –cref requires consistent seed or character prompt structure to maintain identity.
- Style references work across subjects (not just the original subject in the reference).
- Combine –sref with –stylize for hybrid approaches when no exact reference exists.
- Test –sw values starting at 300, then adjust up or down based on results.
Example Input
Natural language style description:
“Minimalist Japanese woodblock print, like Hokusai but with muted earth tones”
Character consistency needed?:
“NO”
Reference image available?:
“NO”
Desired output format:
“WIDE”
Why It Works
Most users describe styles with adjectives — “make it look cinematic” — which Midjourney interprets inconsistently.
This framework improves outcomes by forcing:
- style attribute extraction (color, composition, lighting, texture, mood)
- reference image suggestion (concrete sources, not abstract descriptions)
- parameter specification (–sref, –sw, –cref values)
- combined prompt construction (ready to use)
- preset creation (repeatable style for brand consistency)
Failure modes this prevents:
- Vague styles that produce random results — “cinematic” (too broad)
- Character inconsistency across generations (no –cref)
- Style drift between images (no saved presets)
- Weak style adherence (–sw too low)
This improves on: Adjective-based style description. Concrete references produce consistent results.
Related to: MJ-01 (Parameters) for syntax; MJ-06 (Remix) for variation patterns.
Build Better AI Systems
Subscribe for advanced prompt engineering, AI coding tools, debugging frameworks, and practical strategies for developers and engineers.