Midjourney Omni Reference is the V7 feature that carries a person or object from one reference image into new images, using --oref and a weight set with --ow. It does not run on V8.2, the default model since July 24, 2026: the Edit Model replaced it and takes up to four reference images. Add --v 7 to keep using Omni Reference, or attach your reference through the Edit Model.
Midjourney Omni Reference is the V7 feature that puts a person, object or creature from one reference image into a new image, using the --oref parameter and a strength you set with --ow. It does not run on V8.2, Midjourney's default model since July 24, 2026: the Edit Model replaced it and accepts up to four reference images, so today you either add --v 7 to keep using Omni Reference or attach your reference through the Edit Model.
Checked October 2026 against Midjourney's documentation: Omni Reference, Edit Model, Version, Parameter List, Modifying Your Creations and Legacy Features. Everything here comes from those pages; it is a documentation walkthrough, not a test of output quality.
This page covers the one feature. For how Midjourney compares with LoRAs, Flux references and face swap, start at our consistent character AI guide, the hub for every method. For the full persona workflow on both versions, shot by shot, use the Midjourney consistent character workflow.
What is Midjourney Omni Reference?
Omni Reference is a subject reference. In Midjourney's words, it lets you "put characters, objects, vehicles, or non-human creatures from a reference image into your Midjourney creations". For persona work that meant one thing: generate a hero portrait once, then reuse that face in new scenes without training anything.
It arrived with V7 and it stayed there. The version notes list V7 as the model that "introduced Draft Mode and Omni Reference", and the Omni Reference page now opens with a redirect in plain text: "When using V8.X, use the Edit Model instead".
- April 3, 2025V7 released
The version that introduced Draft Mode and Omni Reference.
- June 17, 2025V7 becomes the default
It stayed the default until June 9, 2026, which is when most --oref tutorials were written.
- March 17, 2026V8.0 alpha
Early access on the alpha website; withdrawn on July 24, 2026.
- April 14, 2026V8.1 released
Default from June 10 to July 23, 2026.
- July 24, 2026V8.2 becomes the default
It ships the Edit Model, which replaces Omni Reference, Character Reference and the Retexture tool.
Source: Midjourney Version documentation, checked October 2026
Does --oref still work in 2026?
Yes, on V7 only. The Omni Reference page states it twice: the feature "is compatible with Midjourney version 7" and "can only be used with Midjourney version 7". Which tool you use depends on which version your prompt runs on.
| Version | Subject reference tool | How you add it | Images | Status |
|---|---|---|---|---|
| V8.2 (default) and V8.1 | Edit Model | Attach to prompt, or --edit plus image URLs | Up to 4 | The current route |
| V7 | Omni Reference | Omni Reference section, or --oref plus one image URL, with --ow | 1 | Still works when you add --v 7; twice the GPU time of a regular V7 image |
| Older versions | Character Reference | --cref | Not in the current parameter list | Replaced; do not start a new persona here |
--oref tutorial: how to use Omni Reference on V7
You need a V7 prompt, one reference image and a text description of the new scene. Midjourney's page is explicit that the text matters: a prompt will not run on the reference alone.
- One hero image of a fictional character you generated: face visible, even light, nothing covering it
- The file ends in .png, .gif, .webp, .jpg or .jpeg
- For Discord, the image is already online so it has a URL (upload it to Discord to get one)
- Version set to 7 in the settings panel, or --v 7 at the end of the prompt
- Draft Mode and Conversational Mode switched off; no --q 4
- A written description of the persona to repeat in every prompt
- You hold the rights to the image you upload
- 1Set the version to 7
Open settings from the Imagine bar and choose V7, or plan to end the prompt with --v 7. Expected: the prompt no longer runs on V8.2.
- 2Open the images panel
Click the image icon in the Imagine bar. Upload the hero or pick it from your uploads.
- 3Drop it into Omni Reference
Drag the image into the Omni Reference section. Expected: one thumbnail. Only one image is allowed.
- 4Lock it for a series
Click the lock icon to keep the image pinned to the Imagine bar across prompts.
- 5Write the scene
Persona description first, then action, place, outfit and light. Expected: a prompt that would make sense without the image.
- 6Leave Omni Strength at 100
The slider in the Imagine bar is --ow. Start at the default.
- 7Generate and compare
Check the grid against the hero at full size: eyes, nose, jaw, hairline. Expected: the same person in a new scene.
- 8Adjust one thing
If the face drifts, raise the strength a step. If the style is stuck, lower it and restate the style.
In Discord, the same request is one line. The reference goes at the end as a URL:
27-year-old woman, shoulder-length dark brown wavy hair, reading at a bookshop window seat, olive jacket, candid photo, overcast afternoon light --ar 4:5 --v 7 --oref https://your-image-url.png --ow 100Three limits from the documentation shape how you plan a batch. Omni Reference costs twice the GPU time of a regular V7 image. It is not compatible with Fast Mode, Draft Mode, Conversational Mode or --q 4. And its results cannot go straight into Vary Region, Pan or Zoom Out: to edit one, load it into the Editor and remove the image reference and any --oref or --ow parameters first.
Midjourney Omni Reference weight: what --ow does
--ow controls how much detail from the reference appears in the new image. It accepts any value from 1 to 1,000 and defaults to 100. Midjourney gives one firm piece of guidance: unless you are using a very high stylize value, keep the weight below 400, "otherwise your results may be unpredictable".
| --ow range | What Midjourney says | Persona use |
|---|---|---|
| Below 100 | Less detail is taken from the reference. Midjourney says to reinforce the physical characteristics you want to keep in the prompt text. | A stylised or illustrated version of the persona, paired with a style reference |
| 100 (default) | The default weight. | Every new scene. Change it only when you can name the problem |
| Above 100, below 400 | More detail from the reference. Midjourney suggests a higher weight when high stylize or --exp values compete for influence. | The face drifts, or you run a high --s for the look of the feed |
| 400 to 1,000 | Allowed, but "your results may be unpredictable" unless stylize is very high. | Avoid for persona work unless stylize is very high |
Two interactions are worth knowing before you touch the slider. Stylize and --exp compete with the reference for influence, so a feed that runs high on either may need a higher weight to hold the face. And style changes pull the other way: for a different style from your reference, Midjourney recommends naming the style at both the start and the end of the prompt, adding a style reference and lowering the weight, then restating the physical features in text because a low weight carries fewer of them.
The documentation does not publish a table of what each weight looks like, and neither does this page. Move in steps from 100, change nothing else between runs, and keep the value that holds the face without freezing the pose and outfit of the reference.
Midjourney Character Reference vs Omni Reference
Character Reference (--cref) is the older tool and Omni Reference (--oref) is its V7 successor; both are now replaced on the default model. Midjourney's version notes say the V8.2 Edit Model "replaces Omni Reference, Character Reference, and the Retexture tool", the current parameter list no longer shows --cref and marks --oref as "replaced by the Edit Model in V8.X", and when checked, the old Character Reference help address returned the Edit Model article.
- Scope. Omni Reference is not limited to people. The documentation lists characters, objects, vehicles and non-human creatures, which is why it also suits a recurring prop or pet.
- Inputs. Omni Reference takes exactly one image. For several characters, Midjourney suggests one image that contains all of them, described in the prompt.
- What it combines with. Personalization, moodboards, stylize, style references and image prompts all work alongside it.
- What it does not carry. Midjourney warns that "intricate details like specific freckles or logos on clothing may not perfectly match your reference". Pick signature details that are large: hair, glasses, build.
A style reference is a different tool again: it carries the look of an image, not the person in it. Our Midjourney prompts guide for people covers how to pair one subject reference with one style code.
What replaced Omni Reference: the Edit Model
On V8.1 and V8.2, the Edit Model does the same job with more inputs. Midjourney describes three uses: changing an existing image from written instructions, generating new images from up to four reference images, and inpainting and outpainting inside the Editor.
- One reference image
- Strength set with --ow, 1 to 1,000
- Prompts describe the scene
- Twice the GPU time of a regular V7 image
- Results need the reference removed before Vary Region, Pan or Zoom Out
- Up to four reference images
- No weight parameter documented
- Prompts can describe a scene or give instructions
- Works with style references, moodboards, Personalization and --hd
- Inpainting and outpainting built into the Editor; no --tile, no Remix
- 01Attach the hero
Image icon, then Attach to prompt. Lock it to reuse it.
- 02Add views
Up to four images in total: front, three-quarter, profile, full body.
- 03Write the scene
Describe it, or instruct it: both styles work.
- 04Set --ar
The Edit Model matches the first image and ignores your default ratio.
- 05Compare
Same check as before, against the hero.
In Discord, add --edit and paste the image URLs with a space between each:
27-year-old woman, shoulder-length dark brown wavy hair, reading at a bookshop window seat, olive jacket, candid photo, overcast afternoon light --ar 4:5 --edit https://hero.png https://three-quarter.pngIf you built habits around Omni Reference, this is how each one translates:
| Job | Omni Reference on V7 | Edit Model on V8.1 and V8.2 |
|---|---|---|
| One hero image dropped into new scenes | --oref with one URL | Attach the hero in Attach to prompt and describe the new scene |
| Controlling how strongly the reference applies | --ow 1 to 1,000 | No weight parameter is documented for the Edit Model. Use more references, --raw and clearer instructions |
| Two characters in one image | One reference image containing both | Up to four references, or one combined image if their features get mixed up |
| A new angle of the same character | Describe the angle in the prompt text | Instruction prompts such as "let's see this image from the front" |
| A different art style | Lower --ow, state the style at the start and end, add --sref | Retexture by instruction, with a style reference or moodboard |
| Fixing one area of a result | Remove the reference and --oref in the Editor first | Inpainting and outpainting run on the Edit Model in the Editor |
The missing weight is the real change. There is no documented --ow equivalent, so control moves into the inputs: more views of the character, an instruction that says what must stay, and --raw, which Midjourney recommends for "more precise control" of Edit Model results.
Should a persona stay on V7 for Omni Reference?
Stay if the feed already has a V7 look and one hero image is doing the job. Move if you are starting now, because V8.2 is the default and the Edit Model is where Midjourney now points reference work.
Either way, Midjourney is one route among several. When you need the same face across hundreds of images and every angle, references stop being enough and a trained model takes over: the AI Influencers program covers creating the persona and LoRA training for consistent faces, and our Flux consistent character workflow shows the reference-based route on an open model family.
Troubleshooting Omni Reference
- The reference is ignored. The prompt ran on V8.x. Add
--v 7, or attach the image through the Edit Model. - The job will not start. Check for Draft Mode, Conversational Mode, Fast Mode or
--q 4, and make sure the prompt has text as well as the image. - Vary Region, Pan or Zoom Out are unavailable. Open the image in the Editor and remove the reference and the
--orefand--owparameters. - The face is close but small details differ. Expected: freckles and logos may not match. Judge identity on the large features.
- Every image copies the reference's style. Name the style at the start and end of the prompt, add a style reference and lower the weight.
- A harmless prompt is blocked. Midjourney says to expect moderation friction with external images and that blocked jobs do not cost GPU time. Reword and retry.
One rule sits above all of these. Midjourney's external image guidelines require that you hold the rights to what you upload and forbid manipulating images of public or private individuals in abusive ways. Build a persona from a face you generated, not from a real person's photo. For checking a result against the hero, see our AI face consistency guide.
Midjourney Omni Reference: FAQ
What is Omni Reference in Midjourney?
Omni Reference is a Midjourney V7 feature that puts a character, object, vehicle or non-human creature from one reference image into a new image. On the website you drag an image into the Omni Reference section of the Imagine bar; in Discord you add --oref and an image URL. Its strength is set with --ow. On V8.1 and V8.2 the Edit Model does this job instead.
Does --oref work on Midjourney V8?
No. Midjourney's documentation says Omni Reference can only be used with version 7 and tells V8.X users to use the Edit Model instead. V8.2 has been the default since July 24, 2026, so an --oref prompt copied from a 2025 tutorial needs --v 7 added to run, or it needs rewriting with the reference attached through the Edit Model.
What is the best Omni Reference weight?
Start at the default, --ow 100. Midjourney allows any value from 1 to 1,000 and says it is best to stay below 400 unless you use a very high stylize value, because results may become unpredictable. Raise the weight when high stylize or --exp values compete with the reference, and lower it when you want a different art style from the reference image.
What is the difference between Character Reference and Omni Reference?
Character Reference, --cref, is the older feature and no longer appears in the current parameter list. Omni Reference arrived with V7 and is broader: it carries characters, objects, vehicles and creatures, takes one image, and is weighted with --ow. Midjourney's version notes say the V8.2 Edit Model replaces both, with up to four reference images instead of one.
How many images can I use with Omni Reference?
One. Midjourney's documentation says you can only use one image with Omni Reference, on the website and in Discord. If you need two characters, it suggests a single reference image that contains both, described in the prompt. The Edit Model on V8.1 and V8.2 accepts up to four reference images, which is the main practical reason to move.
Does Omni Reference cost more GPU time?
Yes. Midjourney says using Omni Reference costs twice the GPU time of a regular V7 image. It is also not compatible with Fast Mode, Draft Mode, Conversational Mode or --q 4, according to the same page, and blocked jobs are not charged. Budget for the doubled cost when you plan a batch of persona images on V7.
What replaced Omni Reference in Midjourney?
The Edit Model, introduced with V8.2 and compatible with V8.1 and V8.2. It changes an existing image from written instructions, generates new images from up to four reference images, and powers inpainting and outpainting in the Editor. In Discord the parameter is --edit followed by image URLs separated by spaces; on the website you attach images to the prompt.
One reference holds a face. A system holds a persona.
AI Influencers, included in All Access, covers creating the persona, prompt engineering, LoRA training for consistent faces, ControlNet for body consistency and video, with the other three programs, weekly coaching and the community in one subscription.
Write the persona line you will repeat
The free prompt generator drafts a persona description to paste in front of every Midjourney prompt. Share your hero image and first series in the free Telegram for feedback.