Learn › Midjourney Mastery: From First Image to Pro Workflow › Omni Reference (--oref): Consistent Characters and Objects

Omni Reference (--oref): Consistent Characters and Objects

Beginner 🕐 13 min Lesson 7 of 13
What you'll learn
  • Apply --oref to maintain a specific character or object across multiple generations
  • Set --ow to the right weight for your use case (300-400 for characters, lower for loose influence)
  • Combine --oref and --sref to maintain both subject consistency and aesthetic consistency

The Consistency Problem

Text-to-image AI has always struggled with consistency. Generate the same character ten times and you'll get ten different interpretations. The face changes, the hair changes, the proportions shift. This makes it difficult to build visual narratives, product mockups, or anything where the same subject needs to appear in multiple images.

Omni Reference is Midjourney's answer to this problem. Introduced with V7, it lets you upload a reference image of any subject — a person, a product, a mascot, an object — and tell Midjourney to preserve that subject's visual identity across new generations.

How to Use Omni Reference

In Discord: add --oref [image URL] --ow [weight] to your prompt.

a woman walking through a Tokyo street market, late afternoon, cinematic --oref https://example.com/character.jpg --ow 300

In the web app: click the Omni Reference section in the prompt bar and upload your reference image. The weight slider appears alongside it.

The weight parameter --ow controls how closely the output matches the reference subject. Scale is 0–1,000, default is 100.

Choosing the Right Weight

Weight Range Effect Best For
25–50
Loose influence — general feel of the reference
Mood or vibe consistency, not exact matching
100 (default)
Moderate matching — recognizable but with variation
Character continuation across scenes
300–400
Strong matching — clothes, features largely preserved
Product shots, consistent character clothing
500–1000
Very close — output heavily resembles the reference
Specific product recreation, high-fidelity consistency

Midjourney's general recommendation is to keep weight under 400 for most use cases. At very high weights, the output can become stiff — the subject is preserved but the composition and scene lose variety. Around 300–400 is often the sweet spot for consistent characters across different scenes.

What Omni Reference Works Well With

Characters and people: Generate the same character in different situations — running through a forest, sitting at a cafe, in a medieval setting. The face and proportions will stay recognizable across generations.

Products: Keep a product's shape, color, and branding consistent across different lifestyle shots. Upload a clean product image and generate it in various settings.

Brand mascots: Maintain a character mascot's visual identity across marketing materials, illustrations, and icons.

Objects: A specific piece of furniture, a logo shape, a vehicle model — any object that needs to appear consistently across images.

Combining with Style Reference

Omni Reference and Style Reference work independently and can be used together. --oref controls the subject; --sref controls the aesthetic.

a woman walking through a market --oref https://character.jpg --ow 300 --sref https://aesthetic.jpg --sw 150

This combination is powerful for maintaining a consistent character in a consistent visual style across a whole image series.

A Note on Real People

Midjourney's terms of service prohibit generating images of real, identifiable people without consent. Using a real person's photo with --oref at high weights approaches the boundary of what's allowed. Stick to fictional characters, avatars, product images, or generic faces when using this feature.

Key takeaways
  • --oref (Omni Reference) preserves the visual identity of a specific subject — person, object, or product — across new scenes
  • --ow weight range is 0-1000 (default 100); Midjourney recommends keeping it under 400 for most cases
  • Combining --oref with --sref controls subject (oref) and aesthetic (sref) independently
Learn more every week Get 5 hand-picked prompts + new AI tools every Monday. Free.