Aivora AIQuick Answer
In Midjourney V7, add --oref [image URL] to your prompt to put a character from a reference image into a new scene, and use --ow (0-1000) to control how strongly the reference is followed. Start around --ow 100 to --ow 200, raise it if the face drifts, and describe the outfit and scene in plain text. Omni Reference works only in V7 and accepts one image per prompt.
If you read my guide on how to create consistent character AI images, you saw the V6 method: --cref with --cw. In V7, that feature was replaced by Omni Reference, which is broader: it can carry over a character, but also an object, a vehicle, or a creature.
This guide covers the V7 workflow end to end, so you can keep one character looking the same across a comic, a mascot campaign, or a storybook.
What Is Omni Reference?
Omni Reference lets you take a character, object, vehicle, or non-human creature from one reference image and place it into a new Midjourney creation. Think of it as telling the model: "put this in my image."
- V7 only. It is not available in V6.1.
- One reference image per prompt. That image can contain several characters if you describe each one in the prompt.
- Works with other references. It can be combined with style references (
--sref) and image prompts. - Not for inpainting or outpainting. Those tools still use V6.1, so remove
--orefand--owbefore editing in the Editor.
--cref vs --oref: What Changed
| Feature | --cref (V6) | --oref (V7) |
|---|---|---|
| Works on | Characters | Characters, objects, vehicles, creatures |
| Strength control | --cw (0-100) | --ow (0-1000) |
| Model version | V6 / V6.1 | V7 only |
| Images allowed | One or more | One |
How to Use Omni Reference (Step by Step)
On the Midjourney website
Add your reference
Drag your character image into the Omni Reference area of the Imagine bar. Use a clean, well-lit portrait with the face clearly visible.
Set the Omni Strength slider
The slider is the same as --ow. Start in the low-to-mid range, generate, then adjust.
Write the new scene
Describe the outfit, pose, and setting in text. The model needs context to know what to change. You can lock the reference so it stays in place across multiple prompts.
On Discord
Add --oref and your image URL to the end of the prompt, then --ow for strength:
/imagine prompt: a woman in a yellow raincoat walking through a rainy Tokyo street at night, cinematic --oref https://your-image-url.png --ow 150 --v 7Choosing the Right --ow Value
The range is 0 to 1000, but higher is not always better. Midjourney's own guidance is that you rarely need to go above moderate values (around 400) unless you are also using very high --stylize or --exp, because those compete with the reference for influence.
| Goal | Starting --ow | Tip |
|---|---|---|
| Restyle the character (anime, painting) | Low, about 25 | Over-describe features you want kept: hair color, outfit |
| Same face, new scene | 100-200 | Good default for storytelling |
| Stronger likeness | 300-400 | Raise slowly and compare |
| High --stylize or --exp | Raise to match | Both compete with the reference |
⚠️ Cost note: Omni Reference uses more GPU time than a normal prompt. Test ideas at lower settings first, then spend on the final versions.
Changing Outfits While Keeping the Face
In V6 you set --cw 0. In V7, use a lower --ow and describe the new outfit clearly:
/imagine prompt: the same woman wearing a white spacesuit with blue glowing accents, standing on Mars --oref [URL] --ow 100 --v 7If the old outfit keeps leaking in, lower --ow a little and make the new clothing description more specific. If the face drifts, raise it a little.
Two or More Characters
Only one image can be used, but it can show multiple people. Put your characters together in one reference image and describe each in your prompt, for example "a tall man with a red scarf and a short woman with silver hair".
Common Problems and Fixes
- Face drifts between images: use a cleaner reference (front-facing, even light, simple background) and raise
--owgradually. - Result looks stiff or copied: lower
--owand add more scene description. - Style fights the character: if you use high
--stylizeor--exp, increase--owto balance them. - Edit tool removes the character: inpainting and outpainting do not support Omni Reference, so edit first or remove the parameters.
- Need absolute consistency over hundreds of images: consider training a LoRA, covered in my full consistency guide.
✅ Pro tip: Save your best reference image and your "master prompt" (the fixed description of your character) together. Reusing both for every scene is the biggest single habit for consistency.
🎨 Build Your First Consistent Character
Pick one clean portrait, start at --ow 150, and generate five scenes. Adjust from there.
Frequently Asked Questions
Omni Reference (--oref) puts a character, object, vehicle or creature from a reference image into a new Midjourney image. It works only in V7 and accepts one reference image per prompt.
--cref was the V6 character feature with --cw. --oref replaces it in V7, covers more than characters, and uses --ow with a 0-1000 range.
Start around 100-200 for a consistent face in new scenes, go low (about 25) for style changes, and rarely exceed about 400 unless you also use very high stylize or exp values.
Only one reference image is allowed, but it can contain several characters. Describe each of them in your prompt.
No. Inpainting and outpainting still use V6.1, so remove --oref and --ow before editing.
Features and limits change. Check Midjourney's official Omni Reference documentation for the latest details.