Midjourney

Midjourney Distorted Faces — How to Fix Weird or Warped Facial Features

Distorted faces are one of the most common issues Midjourney users encounter, especially when generating portraits in v5 and v6. The AI can produce extra eyes, melting features, unnatural proportions, or uncanny valley results that look nothing like a real human face. This guide walks you through exactly why it happens and the fastest ways to fix it.

Why does this error happen?

Midjourney generates images using a diffusion model that reconstructs facial features from learned statistical patterns rather than a true understanding of human anatomy. When prompts are vague, when multiple subjects are present, or when the model is asked to render faces at certain angles or in complex lighting, it struggles to maintain consistent facial geometry. Versions 5 and 6 improved photorealism significantly, which paradoxically makes subtle facial distortions more noticeable. The model also tends to blend features from multiple training references simultaneously, leading to merged or duplicated facial elements — particularly around the eyes, mouth, and ears.

✓

How to fix it

Add 'realistic face, natural proportions' to your prompt

Explicitly telling Midjourney what you want helps steer the diffusion process toward anatomically correct outputs. Append phrases like 'realistic face, natural proportions, symmetrical features, detailed skin' directly to your prompt. Avoid vague descriptors and instead be specific about the face style, angle, and lighting you expect.

Use --cref for face consistency across generations

The --cref (character reference) parameter lets you provide a reference image URL so Midjourney anchors facial features to a known source. Add --cref [image_url] at the end of your prompt to maintain consistent facial structure across multiple generations. This is especially useful when iterating on a portrait or building a character across a series of images.

Try Niji mode for stylized or illustrated faces

If photorealistic faces keep distorting, switching to Niji mode with --niji 6 can produce cleaner, more consistent results for stylized or anime-influenced portraits. Niji is optimized for illustrated characters and handles facial features with more predictable output. Use it when the goal is a stylized look rather than strict photorealism.

Use Upscale with face enhancement after generation

Once you have a base image, use Midjourney's built-in Upscale options and select 'Upscale (Creative)' or apply a face-focused upscaler to refine distorted details. For more control, export the image to a tool like Topaz Gigapixel AI or Adobe Firefly's Generative Fill to correct specific facial areas. Running a second-pass upscale often resolves minor distortions left behind in the initial render.

Pro tip

Always specify the camera angle and focal length in your prompt — for example, '85mm portrait lens, eye-level shot' — to give Midjourney strong spatial context that dramatically reduces facial distortion in portrait generations.

Frequently asked questions

Why does Midjourney give people extra eyes or mouths?

This happens because the diffusion model blends overlapping feature patterns from its training data, especially when the prompt lacks specificity about face count or orientation. Adding 'single subject, one face, frontal view' to your prompt significantly reduces the chance of duplicated facial features.

Does --style raw help with face distortion in Midjourney?

Yes, using --style raw reduces Midjourney's default aesthetic processing, which can sometimes over-stylize and warp facial features. It gives you a more neutral output that is closer to a straightforward interpretation of your prompt, making faces easier to control.

Which Midjourney version produces the best faces?

Midjourney v6 currently produces the most anatomically accurate and detailed faces when paired with specific prompting. For maximum face quality, combine v6 with --style raw and a clear facial description including lighting, angle, and expression.

Can I fix a distorted face without regenerating the image?

Yes — use Midjourney's Vary (Region) feature to select only the face area and re-prompt just that section. This lets you fix a distorted face while keeping the rest of the composition intact.

Related Guides

← Back to all guides