Midjourney V7 Omni Reference: A Practical --oref Guide

Midjourney V7's Omni Reference (--oref) is a powerful parameter designed to maintain the identity of a person, object, vehicle, or non-human creature across multiple image generations, automatically activating V7 and costing twice the GPU time of a regular V7 job [S3]. This guide provides a practical workflow for leveraging --oref and its companion --ow (Omni Weight) to achieve reproducible results, understand its limitations, and decide when to use it over a V8.1 image prompt.

What is Omni Reference in Midjourney V7?

Omni Reference (--oref) is a Midjourney V7-specific feature that allows you to provide a single reference image to guide the generation of a consistent subject [S3]. Unlike general image prompts, --oref focuses on transferring the core identity of a specific entity—be it a character, a product, or an animal—into new scenes and styles. It's particularly useful for maintaining visual continuity when developing narratives, product mockups, or character studies.

Key characteristics of Omni Reference:

• V7 Exclusive: --oref is only available in Midjourney V7 [S1]. It automatically switches your job to V7, even if you're currently using a newer default version like V8.2.
• Subject-Focused: It excels at carrying the identity of a person, object, vehicle, or non-human creature [S3].
• Single Image Input: You provide one URL to an image that serves as the reference [S3].
• Text Prompt Required: An accompanying text prompt is always necessary to describe the desired scene, action, or style [S3].
• GPU Cost: Omni Reference jobs consume twice the GPU time of a standard V7 job [S3].
• Incompatibilities: It cannot be used with Fast, Draft, Conversational Mode, q4, Vary Region, Pan, or Zoom Out [S3].

How to use --oref and --ow

To use Omni Reference, you'll need a reference image URL and a text prompt. The basic syntax is straightforward:

/imagine [your text prompt] --oref [URL to reference image]

The --ow (Omni Weight) parameter controls how strongly Midjourney adheres to the visual characteristics of your reference image [S3]. It ranges from 1 to 1000, with a default of 100 [S3]. The provider recommends staying below 400 for typical stylized generations [S3].

Steps for using --oref:

1. Select a clear reference image: Choose an image where the subject is well-lit, clearly visible, and ideally in a neutral pose or context if you want maximum flexibility.
2. Upload your image: If your image isn't already online, upload it to a platform that provides a direct image URL (e.g., Discord, Imgur, your own website).
3. Construct your prompt: Combine your desired scene description with the --oref parameter and the image URL.
4. Experiment with --ow: Adjust --ow to fine-tune the balance between the reference image's influence and your text prompt's creative direction.

Example Prompt Structure:

/imagine a futuristic detective in a neon-lit alley, rain falling --oref https://example.com/detective-ref.jpg --ow 200

What changes at low, medium and high Omni weight?

The --ow parameter is crucial for controlling the fidelity to your reference image. Understanding its impact at different ranges will help you achieve your desired outcome.

| --ow Range | Characteristics

Character, product and object-reference workflows

Omni Reference shines in scenarios requiring consistent visual identity. Here are workflows for different subject types:

Character Consistency Workflow

Goal: Maintain a character's appearance (face, hair, build, outfit details) across various scenes, poses, and expressions.

Reference Image Selection:

• Ideal: A clear, well-lit headshot or full-body shot of the character in a neutral pose. Avoid strong expressions or complex backgrounds.
• Avoid: Images with heavy stylization you don't want to carry over, or images where the character is obscured.

Workflow Steps:

1. Establish the Base Character: Generate your initial character design using a detailed text prompt. Once satisfied, select one image as your --oref source.

   /imagine a young female elf, long silver hair, emerald eyes, wearing leather armor, fantasy art --v 7
   

2. Apply --oref for Scene Variations: Use the generated character image as --oref and modify your text prompt for new scenarios.

   /imagine the young female elf, sitting by a campfire, looking thoughtful --oref [URL to base elf image] --ow 150
   

3. Adjust --ow for Detail vs. Flexibility:
   • Higher --ow (200-400): For maintaining intricate details like specific facial features or armor patterns. The character will look very similar.
   • Lower --ow (50-100): For allowing more stylistic interpretation or adapting the character to new outfits/expressions while retaining core identity.

Product Mockup Consistency Workflow

Goal: Showcase a specific product (e.g., a unique bottle, a custom gadget) in different environments or with varying lighting.

Reference Image Selection:

• Ideal: A clean, isolated shot of the product against a neutral background. Multiple angles can be useful for different shots.
• Avoid: Images with distracting elements or heavy shadows that might be misinterpreted as part of the product's design.

Workflow Steps:

1. Isolate the Product: Ensure your reference image clearly features only the product you want to reference.

   /imagine a sleek, minimalist smart speaker, matte black finish, cylindrical design, studio lighting --v 7
   

2. Place Product in Context: Use the product image as --oref and describe the desired setting.

   /imagine the sleek smart speaker on a modern wooden desk, sunlit window in background --oref [URL to smart speaker image] --ow 300
   

3. Test --ow for Detail Retention:
   • Higher --ow (300-500): Crucial for maintaining precise product dimensions, logos, and material textures. Essential for product visualization.
   • Lower --ow (100-200): Might allow for more artistic interpretation of the product, potentially altering minor details, suitable for conceptual art.

Object/Prop Consistency Workflow

Goal: Replicate a specific non-human object (e.g., a unique ancient artifact, a custom vehicle) in various scenes or styles.

Reference Image Selection:

• Ideal: A well-defined image of the object, showing its key features and proportions. Consider different angles if the object has complex geometry.
• Avoid: Blurry images or images where the object is heavily stylized in a way you don't want to replicate.

Workflow Steps:

1. Define the Object: Generate or find a clear image of the object.

   /imagine an ornate golden chalice, encrusted with rubies, medieval style --v 7
   

2. Integrate into Scenes: Use the object image as --oref to place it in new contexts.

   /imagine the ornate golden chalice on an altar in a dark, ancient temple --oref [URL to chalice image] --ow 250
   

3. Fine-tune with --ow:
   • Higher --ow (250-450): Best for preserving the exact design, material, and intricate details of the object.
   • Lower --ow (75-150): Allows for more stylistic variations of the object, potentially adapting its form or material slightly to fit the new scene's aesthetic.

What Omni Reference cannot do

While powerful, Omni Reference has specific limitations you should be aware of to avoid frustration:

• Not a Style Transfer: --oref is for identity transfer, not style transfer. For style, use --sref (Style Reference) [S2]. Combining --oref with --sref can be effective for transferring both identity and style.
• No Guaranteed Exact Duplication: While it aims for consistency, it won't produce pixel-perfect copies. There will always be some variation, especially in details or when the text prompt strongly conflicts with the reference.
• Incompatible with Advanced V8.1/V8.2 Features: Omni Reference is a V7 feature and cannot be used with V8.1/V8.2's HD mode, Pan, Zoom Out, or Vary Region [S1, S3]. If your workflow relies on these, --oref is not the right choice.
• Not for Complex Scenes: --oref works best with a single, clearly defined subject. Using it with a reference image containing multiple subjects or a complex scene might lead to unpredictable results.
• Limited to One Reference Image: You can only provide one image URL for --oref [S3]. If you need to reference multiple distinct elements, you might need to generate them separately or use traditional image prompts.

Should you use V7 Omni Reference or a V8.1 image prompt?

The choice between V7 Omni Reference and a V8.1 image prompt depends on your specific goals and workflow requirements. Here's a decision framework:

| Feature/Goal | Use V7 Omni Reference (--oref)

FAQs

Q1: Can I use --oref with multiple reference images?
A1: No, --oref currently supports only one image URL [S3]. For multiple image influences, consider using traditional image prompts in V8.1/V8.2 or generating components separately.

Q2: Why is my --oref job taking longer or costing more GPU time?
A2: Omni Reference jobs automatically activate V7 and cost twice the GPU time of a standard V7 job [S3]. This is normal due to the increased processing required for identity transfer.

Q3: My character's outfit keeps changing. How can I fix this?
A3: Ensure your reference image clearly shows the desired outfit. Increase your --ow value (e.g., to 200-400) to give more weight to the reference's details. Also, explicitly describe the outfit in your text prompt to reinforce it.