Midjourney V7 Consistent Characters: Omni Reference
On this page
Answer First
For a recurring character in Midjourney V7, use Omni Reference. On Discord, the current pattern is:
/imagine a detective waiting on a rainy train platform, full-body cinematic still --oref [image URL] --ow 100On the web, upload or select one image and place it in the Omni Reference slot. A text prompt is still required. Midjourney says Omni Reference can carry a character, object, vehicle, or non-human creature into new generations, but it does not promise exact copying.
V7 vs. Older Character Reference Advice
| Question | Current answer |
|---|---|
| Main V7 reference method | Omni Reference (--oref) |
| Reference strength | Omni Reference weight (--ow) |
--cref tutorials | Older Character Reference workflow for V6 or Niji 6 |
| Style matching | Add a Style Reference or reinforce style in the text prompt |
| Exact identity continuity | Not guaranteed; review details across every approved frame |
Midjourney’s current Character Reference documentation explicitly tells V7 users to use Omni Reference instead.
Build a Strong Reference Image
Choose one source image with:
- a single clearly visible character;
- an unobstructed face and distinctive silhouette;
- readable hair, clothing, and signature accessories;
- neutral lighting that does not hide identity cues;
- only imagery you have the necessary rights to upload and use.
Omni Reference accepts only one image, though Midjourney says users can experiment with a single image containing multiple people and describe them in the prompt. For a production character, one-person references are easier to evaluate.
Write Prompts That Separate Identity From Scene
Keep stable identity anchors short and consistent, then vary the scene:
same detective, short silver hair, navy field coat, brass compass;
walking through a bright desert market, documentary photography --oref [URL] --ow 100The text matters as much as the reference for scene, action, composition, and extra details. If the desired style differs from the reference, Midjourney recommends reinforcing that style in the text; a Style Reference can also help.
Tune --ow With a Small Grid
The current official range for --ow is 1–1,000, with a default of 100. Midjourney advises generally keeping it below 400 unless a high stylize or experimental value is competing with the reference.
Do not jump directly to the maximum. Run the same prompt at a low, default, and moderately higher weight, then score:
- face and hair continuity;
- costume and accessory continuity;
- pose and scene compliance;
- unwanted copying of reference composition;
- visual quality and moderation outcome.
Choose the lowest weight that preserves the identity cues you need. A stronger reference can also fight the requested scene or style.
Current Compatibility and Cost Boundaries
On the August 15, 2026 source check, Midjourney documented that Omni Reference:
- works with V7;
- costs 2× more GPU time than regular V7 images;
- is not compatible with Fast Mode, Draft Mode, or
--q 4; - produces results that are not directly compatible with Vary Region, Pan, or Zoom Out;
- can be combined with Style References and Image Prompts;
- can be edited in the web Editor after removing the image reference and related parameters.
These are volatile product boundaries. Recheck the live Omni Reference documentation before planning a large batch.
A Six-Scene Consistency Test
Generate the same character in:
- close-up neutral portrait;
- full-body standing pose;
- profile view;
- strong warm light;
- strong cool light;
- action pose in a new outfit.
Score only the identity details that matter to the project. Midjourney warns that intricate details—such as freckles or clothing logos—may not reproduce perfectly, so keep a character sheet and a manual cleanup path for high-continuity work.
Rights and Safety
Midjourney’s external-image rules make the user responsible for inputs, outputs, and sharing. Use only images you have the necessary rights to use, and do not manipulate real people in abusive, sexualized, deceptive, or otherwise prohibited ways. A successful generation is not proof of copyright, trademark, publicity-rights, or commercial clearance.
For a wider interface choice, compare Midjourney V7 on web vs. Discord. If recurring-character reliability matters more than staying in Midjourney, use the Leonardo Phoenix vs. Midjourney consistency guide to run the same scene test in both products.
Frequently Asked Questions
How do I keep a character consistent in Midjourney V7?
Use one clear Omni Reference image with a specific text prompt. On Discord, add --oref followed by the image URL; on the web, place the image in the Omni Reference slot. Test a small scene set before a large batch.
Should I use --cref in Midjourney V7?
Midjourney’s current documentation directs V7 users to Omni Reference. Character Reference and --cref belong to the older V6 and Niji 6 workflow, so they should not be the primary V7 instruction.
What does --ow do?
--ow controls Omni Reference weight. The current documented range is 1 to 1,000 with a default of 100; Midjourney advises generally staying below 400 unless high stylize or experimental settings require stronger reference influence.
Does Omni Reference guarantee an identical character?
No. Midjourney warns that intricate details such as freckles or clothing logos may not match perfectly. Treat Omni Reference as generation guidance and use a review-and-rerun workflow for continuity.
What are the current Omni Reference limitations?
Midjourney currently documents one reference image, V7-only compatibility, double GPU time versus regular V7 images, and incompatibility with several features including Fast Mode, Draft Mode, Vary Region, Pan, Zoom Out, and some quality settings. Recheck the live documentation because compatibility can change.