How to Use Midjourney Omni Reference for Characters

Use Midjourney Omni Reference with V7 to guide a character from one reference image into new scenes. For V8.1 and V8.2, use Edit Model instead: it replaces Omni Reference and Character Reference in V8.X. Start with the Omni Reference guide↗ (opens in a new tab) so you choose the workflow that matches your model.

A reference can guide consistency, but you still need to review the results. fenbs can keep the character’s approved traits, test prompts, and image links outside one generation session. That gives another AI the written assignment and your findings without pretending it can transfer Midjourney’s reference behavior to a different model.

Which steps match your Midjourney version?

1. Choose the model version for the work and write it down before starting your comparison.
2. For V7 in Discord, write the new scene prompt and add --oref followed by the URL of your chosen reference image. Omni Reference accepts one image.
3. For V8.1 or V8.2 on the website, click the image icon in the Imagine bar, add your images under Attach to prompt, and use Edit Model.
4. Generate a simple first scene. Review the subject before making the setting, pose, or lighting more demanding.
5. Save the complete prompt, source image link, model version, and accepted output together.

The Edit Model documentation↗ (opens in a new tab) supports up to four references for that workflow. Do not read the extra reference capacity as a promise that identity will remain exact. Decide what your assignment requires and inspect those details yourself.

Use one version for the first test set. If you later change versions, treat that as a new comparison and keep the earlier results. Otherwise it becomes difficult to tell whether a change came from the reference, the prompt, or the model you selected.

What makes a useful character reference?

Choose a source you are allowed to use and can keep available for the assignment. Before generating anything, write a short description of the traits a reviewer must recognize. For a fictional courier, that might be a dark bob, round glasses, a yellow raincoat, and a small blue delivery bag.

Separate essential traits from details you can tolerate changing. Hair shape might be essential; the exact folds in a sleeve might not be. This is an example review policy for your work, not a Midjourney setting. It prevents a teammate from approving a pleasing image that no longer matches the character you agreed on.

• Keep a clear link to the original reference and identify which file is approved.
• List the visible traits that make the subject recognizable to your audience.
• Write down details that must receive extra human review, such as an emblem or distinctive face mark.
• Record permitted changes, including clothing changes that the story specifically requires.

Omni’s guide warns that small details such as freckles and logos may not match. If your delivery depends on an exact mark, review it at the size people will actually see. Do not make “same character” stand in for a specific acceptance checklist.

How should you test the character across scenes?

Start with a quiet, straightforward scene that makes the subject easy to judge. Then try the situations the finished work needs: a different angle, a more distant view, a changed pose, or a different setting. Keep the core written character description stable while you change one scene requirement.

For the fictional courier, you might test standing outside a shop, carrying a parcel through a station, and speaking with another person at a doorway. Review the face, hair, glasses, coat, and bag in each output. A successful close view does not settle whether the same subject will read clearly in a busy wide scene.

Write the result as an observation: “Glasses changed in the wide scene; coat and bag stayed recognizable.” That is more useful than “did not work.” The design critique prompts can help organize your review, while the person responsible for the artwork makes the final choice.

What should you change when the subject drifts?

Return to the last accepted prompt and make a smaller scene change. Keep a copy of the failed result and name the visible difference you need to address. Avoid changing the character description, setting, style direction, and model version at the same time; that leaves you with too many possible explanations.

Distinguish a subject problem from a style problem. Midjourney’s Style Reference guide↗ (opens in a new tab) describes references that carry appearance rather than copying the people or objects in an image. A style code is therefore not a substitute for the character workflow. Use the separate guide to keeping Midjourney style codes for that intent.

Stop and reassess when an essential detail keeps failing your review. You can record that a method is useful for concept options but has not passed the final delivery test. That honest limit is a practical finding your next session needs, not a reason to quietly lower the requirement.

What belongs in the saved character sheet?

Keep a compact sheet with the approved reference, written traits, model version, scene prompts, and reviewed outputs. Label each scene as accepted, rejected, or still needing review. Include the reason for a rejection when it teaches the next person something useful.

A prompt library can hold the reusable scene tests. For the wider assignment, the AI graphic design guide covers production review beyond character generation. Save the character’s current state and next check in fenbs so another assistant can continue from those findings.

How to keep your character findings with fenbs

1. Create a free account.
2. Name the work after the character or story so you can find it from any device.
3. Save links to the approved reference and accepted scenes, then add the traits that must stay the same.
4. Write the model version, prompts you tested, problems still visible, and the next scene to try. Another AI can read that record before helping you continue.

fenbs keeps the work around the character; it does not generate images or guarantee identity. If you only need one disposable concept, a separate record may add little. It helps when the same character returns across assignments, people, or AI apps. Find more reusable review methods in the prompts and skills hub.