Midjourney --cref and Omni-Reference: The Complete Guide to Consistent Characters

If you've ever tried to keep the same character recognizable across a set of Midjourney generations using prompt text alone, you know how fast it falls apart - a slightly different face, a different build, an outfit that drifts. --cref (and its newer form, --oref) exists to solve exactly that problem: it references an actual image of the character instead of asking the model to reconstruct their appearance from a written description every single time.

This guide goes deep on the mechanics - syntax, weight control, version differences, and the combinations that actually work in practice - along with the honest limitations most tutorials skip over.

What is Midjourney --cref (character reference)

--cref is a Midjourney parameter that takes an image URL as its input and uses that image's subject as an identity reference for the new generation. Instead of describing a character purely in words - "a woman with short auburn hair, green eyes, freckles" - you point Midjourney at an actual picture of that character and it carries the identity forward into the new scene you describe in your prompt.

Basic syntax:

The prompt text still defines the scene, pose, lighting, and setting. --cref's job is narrower and more specific: keep the character in that scene recognizably the same person as the one in the reference image.

How to use --cref in Midjourney

The practical workflow:

1. Get a reference image. This can be a character you generated earlier in Midjourney, a photo, or any clear image where the character's face is visible and unobstructed.
2. Host the image somewhere accessible. Midjourney needs a direct image URL - uploading it to Discord and copying the resulting link works, as does any public image host.
3. Write your scene prompt as normal - describe the setting, pose, lighting, and mood you want, exactly as you would for any other generation.
4. Add the parameter at the end: --cref [image URL].
5. Generate, then iterate. Check how closely the identity holds up. If it's too loose or too rigid, the next lever to pull is --cw, covered next.

One detail that trips people up: --cref only works reliably when the reference image has a clear, forward-facing or near-forward view of the subject. A heavily obscured, side-profile, or low-resolution reference gives the model much less to work with.

Controlling the effect with --cw (character weight)

--cw (character weight) is a 0-100 value that controls how much of the reference image gets applied beyond just the face:

• Low --cw values (around 0-30) apply the reference mainly to facial features - the character's face stays recognizable, but clothing, body type, and styling are free to follow the prompt text instead.
• Mid-range values (around 40-70) start pulling in more of the reference's body type and general styling alongside the face.
• High values (80-100) match the reference closely across face, body, and clothing - useful when you genuinely want the same outfit and build in every scene, but this also makes the result more rigid and less able to naturally fit a very different pose or setting.

A good default starting point is a low-to-mid --cw value when you only need the face locked and want everything else - outfit, setting, pose - to follow your text prompt freely. Push it higher only when outfit and build consistency genuinely matter for the shot.

--oref and --ow: the V7/V8 evolution

--oref (omni reference) is the parameter Midjourney introduced alongside V7 as the successor to --cref, paired with --ow (omni weight) in place of --cw. The core idea is the same - reference an image to maintain identity across generations - but --oref broadens the concept: it isn't limited to human characters. It works for products, objects, and other subjects you want to keep visually consistent across a set of images, not just faces.

Syntax mirrors --cref/--cw directly:

Which one should you actually use? If you're generating on V7 or later, use --oref/--ow - it's the actively maintained parameter and generally gives stronger, more flexible results. --cref/--cw still functions on earlier model versions, so if you're deliberately working in an older version for a specific reason, that's when it's still relevant. For the full breakdown of every parameter by model version, see our complete Midjourney parameters guide and our Midjourney V8 guide for what changed most recently.

Combining character reference with style reference (--sref)

This is where character consistency gets genuinely powerful: --oref/--cref and --sref solve two entirely different problems, which means you can stack them in the same prompt. --oref/--cref locks who the subject is. --sref locks what the whole image looks like - color grading, lighting mood, rendering style, overall aesthetic. Used together, you get the same character rendered consistently in the same visual style, across as many scenes as you want.

A few practical notes on combining them:

• Keep the two reference images conceptually separate - one purely for identity, one purely for style - rather than trying to reuse a single image for both roles. Mixed-purpose references tend to muddy both effects.
• If results feel like a fight between the two references, lower one weight before raising the other. It's rarely both at once that needs adjusting.
• This combination is especially useful for building a recurring character across a branded content series - the same face, consistently rendered in your brand's established visual style, scene after scene.

Using multiple reference images

Midjourney supports referencing more than one image at once for --cref/--oref, which can meaningfully strengthen consistency - especially when a single reference image only captures one angle or expression. Stacking two or three images of the same character (different angles, different expressions) gives the model a broader picture of what stays constant about that character's identity, rather than anchoring to one specific pose.

Use this when a single reference is giving you an inconsistent or narrow read on the character - multiple references smooth that out. It's not necessary for every generation, but it's a useful lever when one reference alone isn't holding up across very different poses or angles.

What --cref can't fix

In the spirit of being honest about tool limitations: --cref and --oref are strong, but they are not a guarantee, and knowing the common failure modes saves a lot of frustrated regenerating.

• Major pose or angle changes. Asking for a dramatically different angle - a back view when your reference is front-facing, for instance - gives the model very little identity information to work with, and consistency drops noticeably.
• Style conflicts between the reference and the text prompt. If your reference image is photorealistic but your prompt asks for a painterly or anime style, the model has to reconcile two competing instructions, and identity fidelity often loses out to style fidelity (or vice versa, unpredictably).
• Low-quality or ambiguous reference images. Blurry, poorly lit, heavily stylized, or partially obscured reference images simply don't give the model enough clear signal to lock onto - garbage in, garbage out applies directly here.
• Over-reliance producing a "pasted-on" look. Pushing --cw or --ow very high while asking for a scene that's fundamentally at odds with the reference (different lighting, different pose, different mood) can produce a character who looks cut out and dropped into the scene rather than naturally lit and posed within it. If you see this, the fix is almost always to lower the weight or bring the prompt's requested pose/lighting closer to what the reference already shows.

None of this makes --cref or --oref unreliable - it just means they work best when you understand what they're actually doing under the hood, rather than treating them as a one-click fix for every consistency problem.

Want help writing the scene prompt that pairs with your reference image - the part that describes pose, lighting, and setting around your locked character? The free Midjourney Prompt Builder walks you through concept, subject, colors and materials, composition, lighting, and camera in order, so the --cref or --oref parameter has a strong prompt to work alongside.