How to Use Image References in Midjourney: Complete Guide (2026)

🖼️ Image generated with: Google Gemini

💡 Prompt optimized for: Midjourney / DALL-E 3 (works on all platforms with minor tweaks)

 

How to Use Image References in Midjourney: A Complete Walkthrough

You've got a killer concept in your head. Maybe it's a character you sketched, a color palette you saved from Pinterest, or a photo with lighting that stopped you mid-scroll. You fire up Midjourney, type a detailed prompt, hit enter—and get back something... fine. Not that thing you were chasing. Just something vaguely adjacent to it.

Sound familiar?

Here's the truth: how to use image references in Midjourney is the single biggest skill gap between people who get consistently great results and those who keep rolling the dice. Text prompts can only take you so far. When you learn to feed Midjourney visual anchors—a face, a style, a whole composition—you stop gambling and start directing.

This walkthrough breaks down every major image reference method in Midjourney, from the classic image prompt to the newer Omni-Reference system. You'll get copy-paste prompts, parameter cheat sheets, and real talk about when to use each one. Let's get into it.

What Are Image References in Midjourney (And Why Should You Care?)

Midjourney's image reference tools let you hand the AI a visual example and say, "Guide me toward this." That could mean matching a character's face across ten different scenes, locking in a specific watercolor texture, or pulling the composition from a photograph into an illustration.

Here's why this matters: words are blunt instruments for visual nuance. You can type "moody lighting with teal and orange tones" all day, but dropping in a reference image communicates that in a single gesture. The AI sees the color relationships, the shadow direction, the texture—all at once.

Midjourney currently offers four distinct ways to use reference images:

•  Image Prompts — the original method. Paste an image URL, and Midjourney uses its content, colors, and composition as inspiration.
• Style References (--sref) — isolates just the aesthetic of an image: the medium, lighting, textures, and color grading.
• Character References (--cref) — locks in a character's facial features and identity across different generations.
• Omni-Reference (--oref) — the V7 powerhouse that pulls any subject—person, object, creature—from a reference image directly into your scene.

Each one solves a different problem. Mixing them is where the magic happens.

Method 1: The Classic Image Prompt (And Its Hidden Power)

This is where most people start. You upload an image, grab its URL, and paste it at the very beginning of your prompt. Simple enough.

But here's what most tutorials skip: the image prompt doesn't just copy the subject. It pulls in composition, color palette, and general "vibe"—often in ways you didn't intend. That's why the --iw parameter exists.

What Is --iw (Image Weight) and When Should You Change It?

The --iw parameter controls how much influence your reference image has compared to your text prompt. The default is 1, and the range goes from 0 to 3 in V6 and V7.

Think of it like a slider between "follow the image" and "follow my words":

•  --iw 0.5 — Text prompt dominates. Great when you want the reference to whisper, not shout.
•  --iw 1 — Balanced (the default). Usually a safe starting point.
•  --iw 2 — The reference image takes the lead. Use this when the visual is more important than your description.
•  --iw 3 — Heavily image-driven. The output will cling tightly to the reference's composition and colors.

Copy-Paste Example: Image Prompt with Weight Control

/imagine prompt: https://your-image-url.jpg a lone astronaut standing on a crystalline planet, bioluminescent flora, deep purple and cyan color scheme --iw 2 --ar 16:9 --v 6

What's happening here: The URL at the front is the image prompt. --iw 2 tells Midjourney to prioritize the reference's composition and palette over the text description. --ar 16:9 sets a cinematic widescreen ratio. Swap the URL for your own reference, adjust --iw between 0.5 and 3 until the balance feels right.

Method 2: Style References (--sref) — Lock In an Aesthetic

If you've ever found an artist on Instagram and thought "I want that look, but with my subject," this is your tool. Style References capture the visual DNA of an image—colors, textures, medium, lighting—without copying the actual objects or people in it.

The syntax is straightforward: add --sref followed by an image URL at the end of your prompt. You can stack multiple images by separating URLs with a space.

The --sw Dial: How Strong Should the Style Be?

The --sw (Style Weight) parameter runs from 0 to 1000, with a default of 100. In V7, style weight tends to have a more noticeable impact when you're using style codes rather than uploaded images, but it still matters for references.

• --sw 50 — A subtle nod to the style. Your prompt's own descriptors carry more weight.
• --sw 100 — The default. Balanced influence.
• --sw 400 — The style is clearly dominant. Good for brand consistency work.
• --sw 1000 — Maximum style saturation. Can override your text prompt's aesthetic entirely.

Copy-Paste Example: Style Reference with Weight

/imagine prompt: a serene mountain lake at sunrise, mist rising from the water --sref https://your-style-reference.jpg --sw 250 --ar 3:2 --v 6

Tweaking tips: Keep your text prompt simple when using --sref. Don't fight the reference with conflicting style words like "photorealistic" when your reference is a watercolor. Describe what you want to see, not how to style it—let the reference handle the "how."

Pro move: You can use --sref random to let Midjourney assign a random style code from its internal library. Once you get a result you like, note the code—it becomes reusable across any future prompt.

Method 3: Character References (--cref) — Keep Your Character Consistent

This is the one that changed everything for storytellers, comic artists, and anyone building a visual brand. The --cref parameter lets you lock a character's facial features, hair, and general identity across completely different scenes, outfits, and art styles.

The catch? Your reference image matters more than anything. Garbage in, garbage out.

The Ideal Source Image for --cref

Based on community testing and Midjourney's own guidance, your best source image is:

•  Forward-facing with a clear, unobstructed view of the face
• Well-lit with neutral or soft lighting
• Neutral expression — no extreme angles or exaggerated emotions
•  Simple background — a plain studio backdrop or solid color works best

The --cw Parameter: Face Only vs. Full Look

The --cw (Character Weight) parameter ranges from 0 to 100, and it determines what gets carried over from your reference:

•  --cw 0 — Face only. Perfect for changing outfits, hairstyles, or putting your character in wildly different contexts while keeping their identity intact.
• --cw 50 — A middle ground. Face plus some clothing and hair influence.
• --cw 100 — The full package: face, hair, and clothing. Use this when you want the character to appear in exactly the same look, just in a new environment.

Copy-Paste Example: Character Reference with Weight

/imagine prompt: a female cyberpunk hacker sitting in a neon-lit alley, rain-slicked streets, cinematic lighting --cref https://your-character-reference.jpg --cw 0 --ar 2:3 --v 6

Why --cw 0 here: The scene is completely different from the source image. We want the character's face to stay consistent, but the clothing and context need to shift. --cw 0 tells Midjourney to focus purely on facial identity.

One important note: In Midjourney V7, character-specific references have shifted. The --cref parameter is primarily a V6 feature. If you're working in V7, Omni-Reference (covered next) is the recommended replacement for character work.

Method 4: Omni-Reference (--oref) — The V7 All-Purpose Power Tool

If --sref and --cref had a baby, it would be --oref. Omni-Reference lets you pull any subject from a reference image—a person, a vehicle, a creature, an object—and place it into your new creation with adjustable fidelity.

This is V7's headline feature for visual consistency. It's more flexible than --cref (which only handles characters) and more "subject-focused" than --sref (which ignores subjects entirely).

How to Use --oref: Discord and Web

In Discord: Add --oref followed by your image URL at the end of your prompt.

On the Midjourney website: Click the image icon in the Imagine bar, drag your reference into the Omni-Reference section, and set your weight with the slider.

Important limitation: Omni-Reference supports only one image at a time. For multi-reference compositions in V8.2, the newer Edit Model supports up to four reference images.

The --ow Dial: From Subtle Influence to Near-Total Copy

The --ow (Omni-Reference Weight) parameter runs from 0 to 1000, with a default of 100. Here's how the community is using it:

• --ow 25 — Light touch. Ideal for style transformations, like turning a photo into a cartoon character.
• --ow 100 — Default balance. The reference guides but doesn't dominate.
• --ow 400 — Strong fidelity. Keeps key details like a character's face or specific clothing consistent.
• --ow 1000 — Near-total replication. The output will look almost identical to the reference, which can feel rigid if you wanted creative variation.

Copy-Paste Example: Omni-Reference with Weight

/imagine prompt: a vintage sports car parked in front of a brutalist concrete building, overcast sky, muted tones --oref https://your-car-reference.jpg --ow 400 --ar 16:9 --v 7

Tweaking guide: Start with --ow 100, generate, and evaluate. If the subject isn't showing up clearly enough, bump it to 200 or 300. If the output feels like a carbon copy and you wanted more creative interpretation, drop to 50. The sweet spot is usually between 150 and 500 for most subject-placement work.

Cost warning: Omni-Reference costs 2x the GPU time of a standard V7 generation. It's also incompatible with Fast Mode and Draft Mode.

Combining References: Where Things Get Really Interesting

Here's where you separate yourself from 90% of Midjourney users. You can stack --sref and --cref (or --oref in V7) to lock both the character and the aesthetic simultaneously.

Copy-Paste Example: Character + Style Lock

/imagine prompt: a battle-worn knight standing at the edge of a cliff, storm clouds gathering --cref https://your-character-url.jpg --cw 0 --sref https://your-style-url.jpg --sw 200 --ar 2:3 --v 6

What this does: --cref with --cw 0 keeps the knight's face consistent. --sref with --sw 200 applies your chosen visual style (maybe it's a graphic novel look, or a painterly fantasy style). The text prompt describes the scene. Three layers of control, one prompt.

Heads up on parameter interactions: In V7, --stylize (--s) and --exp compete with --oref for influence over the final image. If you're using Omni-Reference and your subject isn't showing up strongly enough, try lowering --stylize before cranking up --ow.

Quick Reference: Parameter Cheat Sheet

Parameter	What It Controls	Range	Default	Best For
--iw	Image prompt influence	0–3	1	Balancing image vs. text in basic image prompts
--sw	Style reference strength	0–1000	100	Dialing aesthetic influence up or down
--cw	Character reference scope	0–100	100	Face-only (0) vs. full look (100)
--ow	Omni-Reference strength	1–1000	100	Subject placement fidelity in V7
--sref	Style reference source	URL or code	—	Matching a visual aesthetic
--cref	Character reference source	URL	—	Locking a character's identity (V6)
--oref	Omni-Reference source	URL	—	Placing any subject into a new scene (V7)

Three Mistakes That Kill Your Results (And How to Dodge Them)

Mistake 1: Using a messy source image. For --cref and --oref, your reference quality is everything. A blurry, angled, or cluttered source image will produce inconsistent results. Spend two extra minutes generating or finding a clean, well-lit reference before you start.

Mistake 2: Fighting the style reference with your text. If your --sref is a soft watercolor and your prompt says "hyper-detailed photorealistic 8K render," you're sending mixed signals. Keep text prompts content-focused and let the reference handle aesthetics.

Mistake 3: Maxing out weights immediately. Slamming --iw 3 or --ow 1000 feels like the "strongest" choice, but it usually produces rigid, uncreative outputs. Start at the default, evaluate, and nudge upward only if needed.

Your Turn: Put These Prompts to Work

Here's a challenge that will teach you more than any tutorial: pick one reference image you love—a character sketch, a photograph with a palette you vibe with, anything. Run it through three different methods:

1.  As a basic image prompt with --iw 1.5
2.  As a style reference with --sw 300
3.  As an Omni-Reference with --ow 300 (if you're on V7)

Compare the four-image grids side by side. You'll immediately feel the difference between "content influence," "style influence," and "subject placement." That gut-level understanding is what turns you from someone who uses Midjourney into someone who directs it.

Got a question about a specific parameter combo? Drop it in the comments—I read every one. And if you found this useful, share it with that friend who's still typing 200-word prompts and wondering why the results feel random.

Now go make something worth staring at.

🚀 Get Fresh AI Prompts Now ➔

📜 Disclaimer: Prompts are shared for educational purposes. Feel free to use generated images for personal projects. For commercial use, we recommend creating your own unique version using the provided prompt.