How to Use Midjourney’s Style Reference Feature to Nail Any Aesthetic

Every Midjourney user has been there: you see a piece of art with exactly the mood you want — the color grading, the brushwork, the atmosphere — and you spend twenty minutes trying to describe it in words. The results are fine. Not it. Midjourney’s style reference system, built around the --sref parameter, cuts that loop short. You hand it an image, it reads the visual language, and your next generation inherits the aesthetic. No fine-tuning, no training, no waiting.

This isn’t a single magic button — it’s a set of tools that work best when you understand what they’re actually doing. This guide walks you through the --sref parameter and its companion --sw (style weight) from the ground up, with concrete prompts you can copy and run today.

Midjourney V7, the current release as of early 2026, handles style references significantly better than earlier versions — the model extracts color palette, texture, composition tendencies, and lighting logic from reference images without needing to reproduce the source image’s content. That distinction matters both practically and legally: you’re transferring visual grammar, not copying an artwork.

What You’ll Actually Achieve

By the end of this tutorial, you’ll know how to feed Midjourney a reference image and get outputs that feel like they belong in the same visual universe — same palette, similar texture handling, comparable mood — applied to completely different subject matter. You’ll also know when to push the style weight up, when to dial it back, and how to blend multiple references without everything turning to mush.

What You Need Before Starting

You need an active Midjourney subscription (Basic tier or above works fine) and access to Discord or the Midjourney web interface at midjourney.com. For style references, you’ll need either a URL pointing to an image hosted online, or you can upload directly in Discord. The --sref parameter is available in Midjourney V6 and V7 — make sure you’re not accidentally running an older version by checking your settings with /settings.

Note 💡

You can use multiple --sref URLs in a single prompt — just chain them with spaces. Midjourney will blend the style signals. More on that below.

Step 1: Find or Prepare Your Reference Image

Your reference image does the heavy lifting, so choose carefully. The model reads the whole image for style signals, which means a cluttered reference with ten competing aesthetics will produce ambiguous results. The best references share a few traits: consistent lighting throughout, a clear color palette of three to five dominant tones, and a recognizable texture treatment — whether that’s painterly, flat graphic, photorealistic, or anything else.

You can use any publicly accessible image URL. If you want to reference a specific painting style, grab a high-quality image of a work in that style from a museum archive or a stock site. If you’re working from your own photography for a consistent visual identity, upload it to Discord first, right-click to copy the image link, and use that URL.

Pro tip ✅

Crop your reference to the section that contains the style you actually want. If you love the background treatment in a painting but not the foreground clutter, crop to just the background before using it as a reference. The model reads everything in the frame.

Step 2: Build Your First Style Reference Prompt

The basic syntax is straightforward. Write your subject prompt normally, then append --sref URL at the end. Here’s the simplest working structure:

a portrait of a woman in a garden --sref https://your-image-url.com/image.jpg --v 7

Run that and compare it to the same prompt without --sref. The difference in color and texture handling is usually immediate and obvious. The subject stays what you described — a woman in a garden — but the visual language shifts toward whatever your reference image was speaking.

Now try it with a more specific subject to see how the style transfers across content types:

a futuristic cityscape at dusk, neon reflections on wet streets --sref https://your-image-url.com/image.jpg --v 7

The reference image’s palette and texture logic now apply to an entirely different scene. This is where style reference earns its keep — you’re not cloning the source, you’re borrowing its visual vocabulary.

Step 3: Control Intensity with –sw

The --sw parameter (style weight) controls how aggressively Midjourney applies the style reference. The range runs from 0 to 1000, with 100 as the default. Lower values let your text prompt dominate; higher values push the reference image’s aesthetic harder, sometimes at the cost of prompt fidelity.

Here’s the same prompt at three different style weight settings so you can see the difference in practice:

a still life with fruit and a candle, soft morning light --sref https://your-image-url.com/image.jpg --sw 50 --v 7

a still life with fruit and a candle, soft morning light --sref https://your-image-url.com/image.jpg --sw 200 --v 7

a still life with fruit and a candle, soft morning light --sref https://your-image-url.com/image.jpg --sw 600 --v 7

At --sw 50, the style reference is a whisper — you’ll notice the color temperature shift but your prompt’s descriptors still drive the output. At --sw 200, the balance tips toward the reference. At --sw 600, the reference image’s visual logic is dominant enough that your text description mostly handles subject matter and composition intent, while the aesthetic comes almost entirely from the reference. For most practical use cases, the 100–300 range hits the sweet spot.

Warning ⚠️

Pushing --sw above 500 with complex reference images can produce results that look stylistically accurate but lose coherence in the subject. If your outputs start looking like abstract impressions of what you asked for, pull the style weight back down rather than trying to overpower it with more detailed text prompts.

Step 4: Blend Multiple Style References

Midjourney accepts more than one --sref URL in a single prompt. The model blends the style signals proportionally — useful when you want to combine, say, a specific color palette from one reference with a texture approach from another.

an aerial view of a forest in autumn --sref https://first-reference.com/image1.jpg https://second-reference.com/image2.jpg --sw 150 --v 7

The blend is roughly equal by default. If you want to weight one reference more heavily than the other, you can add :: weighting syntax after each URL — though in practice, most users find that choosing two complementary references and keeping style weight at 100–200 produces cleaner results than trying to fine-tune blend ratios.

Pro tip ✅

When blending two style references, make sure they share at least one dominant characteristic — similar color temperature, similar texture type, or similar compositional tendency. Two completely opposed aesthetics blended together tend to cancel each other out rather than combine interestingly.

Step 5: Combine Style Reference with Other Parameters

Style reference works alongside Midjourney’s other parameters without conflict. A few combinations worth knowing:

Pairing --sref with --cref (character reference) lets you maintain a consistent character while applying a style from a separate reference image — useful for illustration work where you need both visual consistency and a defined aesthetic.

a knight standing on a cliff at sunset --sref https://style-reference.com/art.jpg --cref https://character-reference.com/character.jpg --sw 150 --v 7

Adding --stylize on top of a style reference amplifies Midjourney’s own aesthetic interpretation. Lower stylize values (below 100) keep the output closer to the reference image’s aesthetic; higher values let Midjourney riff on the style rather than adhere to it.

an underwater scene with bioluminescent fish, dark depths --sref https://your-image-url.com/image.jpg --sw 200 --stylize 75 --v 7

Pro tip ✅

If you’re building a consistent visual identity across a series of images — for a project, a brand, or a publication — generate one “anchor” image you like, then use that as your --sref for every subsequent generation. Your outputs will share a visual language without you having to describe it in words every time.

Step 6: Use Your Own Work as a Reference

One of the most practical applications of --sref is referencing your own previous outputs. Generated an image with exactly the right mood? Use it as a style reference for the next one. This creates a feedback loop where your visual style becomes progressively more consistent without any manual fine-tuning.

a busy market street in the rain, evening, warm shop lights --sref https://midjourney-cdn.com/your-previous-output.jpg --sw 120 --v 7

Midjourney stores your generated images on their CDN. Right-click any output in Discord or the web interface to copy the image URL, then drop it directly into your next prompt as a style reference.

Realistic Expectations: What Style Reference Does and Doesn’t Do

Style reference is not a reproduction tool. It extracts aesthetic signals — palette, texture, lighting logic, compositional tendencies — and applies them to new content. It does not copy composition, faces, or specific elements from the reference image. Two images referencing the same source will look related but not identical, and neither will look like the source itself.

That also means some styles transfer better than others. References with strong, consistent color palettes and clear texture treatments (oil painting, watercolor, flat graphic design, high-contrast noir photography) produce reliable results. References with very subtle or highly complex aesthetic characteristics may transfer less distinctly — the model picks up the loudest signals first.

Avoid 🚫

Don’t use low-resolution or heavily compressed images as style references. The model needs enough visual information to extract reliable style signals. A 200×200 pixel JPEG at 60% quality gives it almost nothing useful to work with. Aim for reference images that are at least 512px on the shorter side and reasonably sharp.

Five Ready-to-Run Prompts to Try Now

Swap in your own reference URLs and these will get you results immediately. Each one is designed to test a different dimension of style transfer:

a lone lighthouse on a rocky coast, stormy sky, dramatic waves --sref https://your-reference.com/image.jpg --sw 180 --v 7

a coffee shop interior, warm afternoon light, people working at laptops --sref https://your-reference.com/image.jpg --sw 100 --v 7

a fox in a snowy forest, winter twilight, blue shadows --sref https://your-reference.com/image.jpg --sw 250 --stylize 80 --v 7

a woman reading a book by a window, rain outside, soft natural light --sref https://your-reference.com/image.jpg --sw 150 --v 7

an abstract geometric composition, primary colors, clean shapes --sref https://your-reference.com/image.jpg --sw 300 --stylize 50 --v 7

Pro tip ✅

Run each prompt four times (the default grid) and note which seeds produce the strongest style transfer. You can then use --seed with that number to push variations in the same aesthetic direction.

Why This Approach Matters Beyond the Technique

The --sref parameter shifts how you think about prompting. Instead of spending effort describing an aesthetic in words — always an imperfect translation — you provide a visual example and describe your subject. The creative work moves from “how do I explain this mood in text” to “what do I want to show, and what visual language should it speak.” That’s a cleaner division of labor, and it produces more consistent results faster.

For anyone doing production work — building out a visual brand, illustrating a story, creating consistent assets — style reference is the feature that makes Midjourney genuinely repeatable. Single great outputs are fun. A coherent body of work is useful. The --sref parameter is what gets you from one to the other.

Promptyze

Promptyze covers generative AI in plain English — hands-on reviews, tutorials and daily news, fact-checked and hype-free.

See Full Bio