Answer First
For Midjourney V7 character consistency, use Omni Reference, not older V6-style Character Reference advice. The current V7 workflow is built around --oref for the reference image and --ow for reference weight. Older --cref guidance still matters for understanding legacy Character Reference workflows, but it should not be presented as the main V7 method.
The practical V7 command pattern is:
/imagine cinematic portrait of the same detective in a rainy alley, noir lighting --oref [image URL] --ow 300Midjourney’s Omni Reference documentation describes the system as a broader reference mechanism for characters, objects, vehicles, and other visual subjects. It can improve consistency, but it is not a guarantee of frame-perfect identity.
Omni Reference vs Character Reference
| Use case | Current V7 guidance |
|---|---|
| Consistent character in V7 | Use Omni Reference with --oref. |
| Stronger or weaker reference adherence | Adjust --ow. |
| Older Character Reference tutorials | Treat --cref as legacy or non-V7 context. |
| Style consistency | Combine Omni Reference with style references where appropriate. |
| Perfect animation continuity | Expect extra manual cleanup or downstream tools. |
The important correction is simple: if you are writing a V7 workflow in 2026, do not tell users to use --cref as the primary character-consistency parameter.
How --oref Works
Omni Reference lets you provide a visual reference that Midjourney should carry into the generation. For a character workflow, the reference image should clearly show the identity you want to preserve. Check the current documentation for supported inputs, parameter compatibility, and cost before a large batch.
/imagine a woman standing in a futuristic subway station, editorial fashion photo --oref [image URL] --ow 250The reference can influence facial structure, hair, clothing cues, body type, signature accessories, and other identity anchors. The prompt still matters. If you ask for a different outfit, age, lighting, or setting, Midjourney will try to satisfy both the text and the reference.
Understanding --ow
--ow controls Omni Reference weight. Higher values tell Midjourney to follow the reference more strongly; lower values allow more freedom. Treat it as an iteration knob, not a magic consistency switch.
Midjourney’s current documentation allows values from 1 to 1,000 and uses 100 by default. It recommends staying below 400 unless very high stylize values justify more reference influence, because higher settings can become unpredictable.
| Weight direction | Use when… | Risk |
|---|---|---|
Lower --ow | You need the same character adapted into a different role, outfit, or art direction. | Identity may drift. |
| Around the default | You want a balance between recognizable identity and prompt flexibility. | Some details may change. |
Higher --ow | Identity, outfit, or object fidelity matters more than scene freedom. | Output can become rigid or unpredictable. |
Start near the documented default, compare results, and adjust upward only when identity drift is the main failure.
Build A Strong Reference First
The best character-consistency workflow starts before --oref.
- Create a clean hero reference. Use a clear face, readable silhouette, and simple lighting.
- Avoid extreme expression lock. A screaming or laughing reference can bias later generations.
- Show the important anchors. Hair, glasses, scars, jewelry, costume shape, and color palette should be visible.
- Use enough resolution. A blurry or cropped reference gives the model less identity signal.
- Save the original reference. Do not keep chaining from later outputs unless you intentionally want drift.
For storyboards, comics, and brand characters, generate a reference sheet first:
/imagine character reference sheet, young explorer with short red hair, green eyes, freckles, front view, side view, three-quarter view, neutral expression, white background --ar 16:9Then use the strongest sheet or portrait as your --oref anchor.
Scene Progression Workflow
Use this sequence for a multi-image character project:
- Generate the definitive reference.
- Write prompts for every scene before producing final images.
- Use the same
--orefreference across the scene set. - Keep
--owconsistent for the first pass. - Review all images side by side.
- Regenerate the outliers with slightly higher
--owor clearer identity wording. - Use editing or inpainting tools for small fixes instead of regenerating the whole set.
This prevents the common drift problem where every new image becomes the next reference and the character slowly changes.
If you are still learning the web workflow around references, queues, and saved images, pair this guide with the Midjourney Alpha web guide. For budget planning, the Midjourney V7 subscription guide explains how GPU time affects repeated character iterations.
Multi-Character Scenes
Multi-character consistency is harder because the model must assign the right reference to the right person. Make the prompt explicit:
/imagine two detectives in a cafe, the older man on the left studies a notebook, the young woman on the right smiles at the window --oref [one two-character reference image URL] --ow 250To improve reliability:
- Put both characters into one reference image; current Omni Reference accepts only one image.
- Give each character distinct visual anchors.
- Use positional language such as “left”, “right”, “foreground”, and “background”.
- Avoid prompts where characters overlap or swap clothing.
- Generate several candidates and reject blended faces quickly.
If identity assignment keeps failing, create separate images and composite later rather than forcing a crowded one-shot prompt.
Combining Character And Style Consistency
For a graphic novel, brand campaign, or game concept set, you usually need both character identity and visual style. Use --oref for identity and a style reference or style language for the look.
/imagine the same detective examining clues in a dim office, moody 1970s film still, muted greens and amber practical light --oref [character URL] --ow 250Keep one variable stable at a time. If both identity and style are drifting, first lock identity with --oref and --ow, then refine the style reference or wording.
Common Problems And Fixes
The face changes across scenes
Increase --ow, use a clearer reference, and repeat identity anchors in the prompt. If the reference is a full-body image with a small face, crop or regenerate a stronger portrait reference.
The outfit never changes
Lower --ow and describe the new outfit clearly. A high reference weight can preserve clothing along with identity.
The character keeps the same expression
Use a neutral reference image and prompt the expression directly: “quiet smile”, “worried expression”, “focused gaze”, or “laughing with eyes open”.
The model blends two characters
Separate the subjects more clearly in the prompt or generate them separately. Multi-character consistency is less reliable than single-character consistency.
The scene ignores the reference
Raise --ow, simplify the prompt, and remove competing visual descriptions that contradict the reference.
Limits To Acknowledge
Omni Reference improves V7 consistency, but it does not produce pixel-identical continuity. Character details can still shift across lighting, pose, camera angle, clothing, and art style. For animation, comics, and production brand work, expect a finishing pass.
Midjourney also notes compatibility and usage limits around Omni Reference. The official Omni Reference page mentions extra GPU-time cost and parameter constraints, so check the current docs before planning a high-volume production workflow.
When Older --cref Advice Still Matters
If you are reading old tutorials, --cref explains the earlier Character Reference mental model: use a character image to guide future generations. That concept is still useful, but the V7 execution path has moved to Omni Reference.
Use old --cref content only as background. For V7 instructions, update examples, screenshots, and checklists to --oref and --ow.
Bottom Line
The V7 character-consistency workflow is now Omni Reference-first. Build a clean reference, use --oref, tune --ow, keep the same reference across scenes, and review outputs as a set before polishing.
That approach is more source-accurate than old --cref guidance and gives creators a better shot at consistent characters across storyboards, brand scenes, and visual campaigns.
Continue the Character Workflow in Flowith
- Use the AI furry art generator for a character-brief workflow where species, markings, palette, clothing, pose, and consistency are reviewed explicitly.
- Use the AI avatar generator when the deliverable is a profile-ready character crop rather than a multi-scene sequence.
- Use the AI image workflow when the model choice is still open.
- Compare Recraft alternatives when the job is a repeatable brand-asset system rather than scene continuity.
For any public or commercial character, also review reference-image permission, likeness risk, brand ownership, and the current Midjourney terms. Visual consistency does not establish the right to use the source or output.
Frequently Asked Questions
How do I make a consistent character in Midjourney V7?
Use one clear Omni Reference image with --oref, keep the text prompt explicit, start near the default --ow 100, and compare a scene set side by side. Raise or lower --ow only after identifying whether identity drift or prompt rigidity is the problem.
Should I use --cref or --oref in Midjourney V7?
Use --oref for V7. Midjourney’s current Character Reference documentation says Omni Reference replaces Character Reference in V7; --cref remains relevant to older V6 and Niji 6 workflows.
What does --ow do?
--ow controls how strongly Omni Reference influences the result. Current documentation allows 1 to 1,000, defaults to 100, and recommends staying below 400 unless high stylize settings require more influence.
Can I use two character reference images with --oref?
No. Current Omni Reference accepts one image. For two characters, prepare one reference image containing both people, describe their positions and visual anchors clearly, and expect more manual review than with a single character.
Does Omni Reference guarantee perfect character consistency?
No. Midjourney notes that intricate details may not match perfectly. Lighting, pose, angle, clothing, style, and competing prompt instructions can still introduce drift, so production work needs review and often editing.