Key takeaways

  • On midjourney.com, the image icon in the Imagine bar opens a panel where you drop an image into Image Prompt, Style Reference, Omni Reference or Attach to prompt.
  • Omni Reference (--oref) works only in V7; in V8.1 and V8.2 the Edit model replaces it and takes up to four reference images.
  • Image weight (--iw) defaults to 1 with a range of 0 to 3 in V7 and V8.1, and style weight (--sw) defaults to 100 with a range of 0 to 1000.

To use a Midjourney reference image on the web, click the image icon in the Imagine bar, upload your picture, and drop it into the section for what you want it to do: Image Prompt for the overall look and content, Style Reference for the style alone, or "Attach to prompt" for the V8.2 Edit model, which keeps a character or object. In Discord you paste the image's URL into the prompt, either at the start (image prompt) or after --sref, --oref or --edit.

Which of these you can use depends on the model version. V8.2 has been Midjourney's default since July 24, 2026, according to its Version page, and it changed the tools: Omni Reference (--oref) works only in V7, and in V8.1 and V8.2 the Edit model takes its place. Everything below comes from Midjourney's docs as they read on November 16, 2026.

Which reference type to use

You want to Use Web section Discord syntax Model versions
Borrow the content, colors and composition of a picture Image prompt Image Prompt URL at the start of the prompt V6, V7, V8.1, V8.2, Niji 7
Copy only the style: color, medium, texture, lighting Style reference Style Reference --sref URL V6 and later
Put a specific character, object, vehicle or creature in a new image (V7) Omni reference Omni Reference --oref URL V7 only
Keep characters or objects, combine up to four images, or edit one (V8) Edit model Attach to prompt --edit URL V8.1 and V8.2
Recreate a character (old method) Character reference Character Reference --cref URL V6 and Niji 6, now a legacy feature

All of them accept .png, .gif, .webp, .jpg and .jpeg files. You can switch versions by adding --v 7 (or another number) to the end of a prompt, or set a default in the settings panel of the Imagine bar.

How to add a reference image on the web

These steps are the same for every reference type on midjourney.com:

  1. Click the image icon in the Imagine bar. The images panel opens.
  2. Upload a new image, or pick one you uploaded before. You can select several at once.
  3. Drag each image into a section: Image Prompt, Style Reference, Omni Reference (V7) or "Attach to prompt" (Edit model). Clicking an image in your uploads library adds it as an Image Prompt.
  4. Type your text prompt and submit.
  5. To remove an image, hover over it and click the X. To keep an image for the next prompts, click the lock icon to pin it to the Imagine bar.

How to add a reference image in Discord

Discord needs a link to an image that is already online. If the file is on your computer or phone, host it in Discord first:

  1. Upload the image in a private place, such as a direct message with the Midjourney bot or your own server, so others don't see it.
  2. Wait until it shows in the chat under your name. If it still shows eye, pencil or trash icons, it hasn't finished uploading.
  3. Copy the image's address. On desktop, open the image, right-click and choose "Copy Image Address" or "Copy Link". On iOS, open it, tap the ellipsis, then Share and Copy. On Android, press and hold it and pick "Copy Media Link".
  4. Paste the link into your prompt in the right place:
/imagine prompt: https://example.com/photo.png a watercolor of a lighthouse at dusk
/imagine prompt: a lighthouse at dusk --sref https://example.com/style.png
/imagine prompt: a knight riding a bicycle through Paris --oref https://example.com/knight.png --v 7
/imagine prompt: put this cat on a sailboat --edit https://example.com/cat.png

The example.com links stand in for your own image URLs. A Discord message link won't work, and a "blob URL" error means the upload hadn't finished, so send the image again.

How to use an image prompt and image weight

An image prompt feeds the main elements of a picture into a new one. Midjourney says the result is a new image inspired by yours, not a copy. You can use one image with text, two or more images alone (which blends them), or several images with text.

Write the text as a description of the image you want, not as instructions for changing yours. Crop the reference to the aspect ratio you plan to use. Image-only prompts don't work with --stylize or --weird.

Image weight, --iw, sets how much the image counts against the text. Higher means more influence:

Model Default --iw Range
V8.1 1 0 to 3
V7 1 0 to 3
Niji 7 1 0 to 2

The image prompt page says the feature works in V8.2, but its weight table lists only V8.1, V7 and Niji 7.

How to use a style reference (--sref)

A style reference copies the look of an image, its colors, medium, texture and lighting, without the objects or people in it. It works in V6 and later and needs a text prompt.

  • Keep the prompt short and describe the subject: "a ballpoint pen sketch of a bunny", not "copy this style and make a bunny". Avoid style words that fight the reference.
  • Style weight, --sw, runs from 0 to 1000. The default is 100.
  • In Discord you can give several references different weights: --sref URL1::2 URL2::1.
  • Instead of an image you can use a style code, a number from Midjourney's style library: --sref followed by the code. --sref random picks one and replaces "random" with the real code. Browse codes in the Styles tab of the Explore page. You can't turn an uploaded image into a code.
  • V7 has six style reference versions, chosen with --sv; the default is --sv 6. Style codes and --sref random work only with --sv 4 and --sv 6. V6 has four versions, with --sv 4 as its default.
  • --sw and --sv don't work with Moodboards.

Old style codes may look different now, because the feature changed between V6 and V7. To get the old results back, use --sv 4 in V7 or switch to V6.

How to keep a character or object: Omni Reference in V7

Omni Reference puts a character, object, vehicle or creature from your image into a new scene. It works only in V7, so add --v 7 or set V7 as your default. It is not supported in V8.2.

  • Use one image per prompt. That image can show several people; describe each one in the text.
  • Always add a text prompt describing the whole scene.
  • Omni weight, --ow, runs from 1 to 1000 with a default of 100. On the web it is the Omni Strength slider. Midjourney advises staying under 400 unless you use a very high --stylize, because higher values get unpredictable.
  • To change the subject's style, name the style at the start and end of the prompt, add a style reference and lower --ow.
  • Small details such as freckles or logos on clothes may not match.

Omni Reference costs about twice the GPU time of a normal V7 image, which matters on the plans described in Midjourney's pricing and GPU time. It doesn't work with Fast, Draft or Conversational mode, with --q 4, or with inpainting and outpainting. Its results can't use Vary Region, Pan or Zoom Out. To edit one, load it into the Editor and remove the image reference and the --oref and --ow parameters.

How to use up to four reference images: the Edit model in V8

In V8.1 and V8.2, the Edit model replaces both Omni Reference and Character Reference. It does three things: it makes new images from up to four references, it changes an existing image from a written instruction, and it powers the Editor.

  1. Open the images panel from the image icon in the Imagine bar.
  2. Drag or click up to four images into "Attach to prompt".
  3. Write either a normal description or a direct instruction, such as asking to show the object from another angle. Midjourney suggests trying both.

The output takes the aspect ratio of the first image unless you set --ar; your default ratio setting doesn't apply. --raw cuts Midjourney's own styling so your text counts for more. The Edit model works with image prompts, style references, Moodboards, Personalization and --hd, but not with --tile, and its results can't be used with Remix. If two characters blend into each other, Midjourney suggests merging them into one reference image.

How to edit your own photo in the Editor

To change a picture you already have, rather than use it as inspiration, use the Editor:

  1. Click Edit in the midjourney.com menu.
  2. Upload a file from your device or paste an image URL.
  3. Paint over what you want changed with the Erase brush, or use Smart Select to mask an area. The Restore brush brings back erased parts.
  4. To add space around the image, change the aspect ratio or shrink the image on the canvas.
  5. Write a prompt and submit. In V8, a prompt with nothing erased or selected restyles the whole image.

In V8, Editor results, including ones from uploaded images, show on your Create and Organize pages and are public on midjourney.com unless Stealth mode is on.

Character reference (--cref) is now a legacy feature

Older guides tell you to use --cref. According to Midjourney's Legacy Features page, it works only in V6 and Niji 6, and was replaced by Omni Reference in V7 and the Edit model in V8. To use it, set V6 or Niji 6 as your default and add the image to the Character Reference section, or add --cref URL in Discord. Character weight, --cw, defaults to 100, which is also the maximum; lower it to keep mainly the face.

Rules for uploaded images

Midjourney requires you to have the rights to any image you upload. Its docs forbid manipulating photos of real people in abusive or offensive ways, sexualized deepfakes included, and say breaking the rules can get an account suspended or banned without a refund. Moderation can block prompts that look harmless; blocked jobs don't use GPU time. To check what each plan includes before you start, see whether Midjourney is free.

Frequently asked questions

How do I use --sref in Midjourney?

Add --sref followed by an image URL or a style code to the end of your prompt, or drop an image into the Style Reference section on the web. A text prompt is required, and --sw from 0 to 1000 (default 100) sets how strong the style is.

Does Midjourney still have character reference?

Character reference (--cref) is a legacy feature that works only in V6 and Niji 6. Midjourney replaced it with Omni Reference in V7 and with the Edit model in V8.1 and V8.2.

How many reference images can I use in Midjourney?

Omni Reference in V7 takes one image. The Edit model in V8.1 and V8.2 takes up to four in its Attach to prompt section, and image prompts can blend two or more images.

What does --iw do in Midjourney?

Image weight (--iw) sets how much an image prompt counts compared with the text. It defaults to 1, with a range of 0 to 3 in V7 and V8.1 and 0 to 2 in Niji 7.

Sources

  1. Image PromptsMidjourney
  2. Style ReferenceMidjourney
  3. Omni ReferenceMidjourney
  4. Edit ModelMidjourney
  5. VersionMidjourney
  6. EditorMidjourney
  7. Legacy FeaturesMidjourney
  8. Hosting Images in DiscordMidjourney

Add ClaudeGPT Si as a preferred source on Google