If you have used Midjourney for more than a few weeks, you have probably encountered style reference codes — cryptic numeric strings that somehow make the AI produce images with a specific visual identity. They are one of the most powerful features in Midjourney's toolkit, and also one of the least understood.

This guide breaks down exactly what --sref codes are, how they work, and how they compare to a deeper approach: AI-powered style analysis that decodes the actual visual DNA behind any image.

Style reference codes (--sref) are numeric identifiers that tell Midjourney to apply a specific visual aesthetic to your generation. Instead of describing a style in words — "painterly, warm, golden hour, soft focus" — you pass a code and let the model figure out the rest.

The code 3847291 encodes a particular style that Midjourney has learned. Every time you use it, you get images with the same aesthetic fingerprint — the same color temperature, the same level of detail, the same compositional tendencies.

Parameter	What It Does	Example
--sref	Applies a style reference code. Can be a number or an image URL.	--sref 3847291
--sw	Style weight (0–1000). Controls how strongly the style is applied. Default: 100.	--sw 250
--p	Personalization mode. Applies your personal aesthetic preferences learned from your rating history.	--p
--ar	Aspect ratio. Not a style parameter, but critical for composition.	--ar 16:9
--s	Stylize value (0–1000). Controls Midjourney's own aesthetic interpretation. Default: 100.	--s 750

--sref codes are essentially pointers into Midjourney's learned style space. When you provide a numeric code, the model looks up a specific region of its latent aesthetic space and applies those characteristics to your generation.

What does a style code control? Everything that is not content:

The key insight is that --sref codes are opaque. You know they produce a certain look, but you do not know why. You cannot decompose code 3847291 into "warm color temperature + chiaroscuro lighting + analog grain." The code is a black box.

You can also pass an image URL instead of a numeric code:

This tells Midjourney to analyze the reference image and extract its style characteristics. The --sw parameter controls intensity — at 50 you get a subtle influence, at 500 the style dominates, at 1000 it overpowers the prompt entirely.

Here is the fundamental problem with --sref codes: they work, but you cannot learn from them.

When you use code 3847291 and get beautiful results, you have gained a tool but not knowledge. You do not know what makes that style work. You cannot modify it incrementally. You cannot explain it to someone else. You cannot transfer its principles to another AI tool or to your own photography.

This is where AI style analysis offers a fundamentally different approach.

Instead of encoding a style as an opaque number, AIKIZI's Style Decode engine breaks it into explicit, human-readable layers. Here are four real styles decoded from the AIKIZI gallery:

Each of these styles could be represented by a single --sref code. But the decoded version gives you the recipe: four named layers, key tokens, a color palette, and a master prompt template. You can modify any layer, swap one influence for another, or combine elements from different styles. The style becomes a creative tool, not a black box.

Aspect	Midjourney --sref	AIKIZI Style Decode
Input	Numeric code or image URL	3–10 images
Output	Styled image	4-layer DNA, key tokens, color palette, master prompt
Transparency	Opaque (black box)	Fully decomposed and named
Modifiable	Only via --sw weight	Each layer and token adjustable independently
Transferable	Midjourney only	Works with any AI model via prompt template
Educational	No — you learn nothing about the style	Yes — maps lineage, influences, and techniques
Availability	All Midjourney users	Style Decode (multi-image, 3–10 images) is currently available to select accounts. Single-image Decode is available to all users.

Use --sref when you already know the exact look you want and just need to reproduce it quickly. Style codes are shortcuts — fast, reliable, and require zero understanding of what makes the style work.

Use Style Decode when you want to understand a visual aesthetic, learn from it, modify it, or apply it across different tools. If you are building a brand identity, developing a consistent visual language, or studying art direction, you need the decomposed view.

They are not competitors. --sref is a production tool. Style Decode is a research and creative development tool. The best workflow uses both: decode first to understand, then generate with the knowledge you have gained.

Style references reward a little depth. These companion guides go further, all rebuilt for Midjourney V8.1:

• How to use style references in Midjourney V8.1 — the hands-on workflow, from image references to numeric codes.
• Sref vs cref vs iref — which reference flag actually runs in V8.1, and which is legacy.
• Best Midjourney style references (2026) — real V8.1 images, their exact prompts, and the decoded style DNA behind each.
• Style weight (--sw) — tune how hard a style asserts itself over your prompt.
• Browse the decoded gallery — find a style to match your own reference.

Decode Any Image →