Midjourney cref Not Working? Try These Fixes

Midjourney’s Character Reference (--cref) feature is designed to maintain consistent characters across different generated images. However, users frequently encounter issues where Midjourney completely ignores the reference face, distorts key physical features, mixes up clothing, or throws generation errors.

If your character references are not rendering correctly, it is usually due to syntax errors, file format issues, model incompatibility, or prompt contradictions. Follow these practical troubleshooting steps to fix your character consistency prompts.

Why Midjourney fails to match your character reference

When Midjourney fails to apply a character reference, it is because the algorithm is struggling to parse the target image. The AI does not copy-paste the face from your source image; instead, it analyzes key facial vectors, hair color, and body structure, then attempts to reconstruct them in the new scene.

If your source image is too cluttered, has bad lighting, contains multiple people, or is formatted incorrectly, the model fails to isolate the character's details. Additionally, using wrong parameter syntax will cause Midjourney to ignore the parameter entirely.

How to fix Midjourney character reference issues

1. Verify your Midjourney model version The `--cref` parameter is only supported on **Midjourney v6** and **Niji v6**. It will not work on v5.2, v5.1, or older versions. If your settings default to an older model, the parameter will be ignored. - Type `/settings` in Discord and ensure **Midjourney Model V6** or **Niji Model V6** is selected. - If you are prompting manually, check that you have not appended an older version tag (like `--v 5.2`) to the end of your prompt, as this overrides your default settings.

2. Correct your URL syntax and image format Midjourney requires direct links to image files. If you link to a webpage or a Discord message instead of the image file itself, the system cannot read the data. - Ensure your reference URL ends with a standard image extension such as `.png`, `.jpg`, `.jpeg`, or `.webp`. - In Discord, right-click the uploaded reference image, select **Copy Link** (do not use "Copy Message Link"), and paste it after the `--cref` tag. - Ensure the syntax is formatted correctly: `[your text prompt] --cref [image URL] --cw [weight]`. Keep a space between `--cref` and the URL.

3. Adjust the Character Weight parameter (`--cw`) By default, Midjourney applies a character weight of 100 (`--cw 100`), which attempts to copy the character's face, hair, and clothing. If you are trying to change your character's clothes or context, this default setting causes rendering conflicts, resulting in distorted characters. - Set `--cw 0` if you only want to copy the face. This is ideal for putting the same character in new outfits, different historical settings, or changing their action. - Use values between `1` and `99` to scale how much of the original clothing and hair style Midjourney should retain.

4. Crop and clean your reference image Midjourney struggles to isolate a character if the reference image is visually busy. - Crop your reference image so it focuses tightly on the character's face and shoulders. - Avoid using images with heavy shadows, dramatic side-lighting, face paint, or extreme filters, as these elements distort facial vectors. - Ensure there is only one person in the reference image. If there are background figures, Midjourney may blend features from both people.

5. Remove contradictory details from your text prompt If your text prompt contradicts your character reference, the generator will experience a rendering conflict. For example, if your reference image shows a person with brown hair, but your text prompt says "a blonde woman sitting in a cafe," the model will produce a distorted hybrid or ignore the reference entirely. - Keep text prompts focused on the setting, action, and camera style rather than re-describing physical traits already present in the reference image. - If you must change a trait (like hair color), lower the character weight to `--cw 0` so the text prompt can successfully override the non-facial features.

6. Chain multiple reference images If a single image is not providing consistent results, you can provide Midjourney with multiple angles of the same character to train the model more accurately. - Paste multiple URLs separated by a space after the `--cref` tag: `--cref URL1 URL2`. - Use images of the same character from different perspectives (e.g., front-facing, profile, three-quarter view) to help the model build a better 3D profile of the face.

When to escalate

If you have verified that your syntax is correct, you are using the v6 model, and your reference images are clear, but Midjourney still fails to process --cref, check the #status channel in the official Midjourney Discord to see if image processing nodes are experiencing degraded performance. If the status is normal, post your prompt and reference link in the #member-support channel, where community guides can check for hidden formatting errors or system bugs.