Midjourney --cref Guide: Character Consistency Across Images (2026)

Midjourney's --cref (character reference) parameter, introduced in v6 and significantly improved in v7, lets you maintain a consistent character across multiple images — different poses, environments, lighting conditions, and styles. It's the most-requested feature Midjourney has ever added.

What --cref does

--cref [image URL] tells Midjourney to use the character in your reference image when generating new scenes. The result maintains recognizable features — face, hair, general build — while placing the character in entirely new contexts.

This solves one of AI image generation's biggest limitations: generating multiple images featuring the same person consistently. Before --cref, every Midjourney generation produced a different face.

Basic syntax

[scene description with character] --cref [image URL]

Example: A woman standing in a rain-soaked Tokyo street at night, neon lights reflecting, cinematic --cref https://your-character-image.jpg --ar 2:3

The --cw parameter (character weight)

--cw [0–100] controls how strongly the character reference is applied:

• --cw 100 — maximum fidelity. Face, hair, skin tone, and body proportions are all closely matched. Less creative freedom.
• --cw 75–85 — strong character recognition with natural variation across scenes. Recommended starting point.
• --cw 50 — loose character resemblance. More stylistic freedom; character feels 'inspired by' the reference rather than 'copied from' it.
• --cw 0 — only general style elements (hair color, overall aesthetic) are carried over. Essentially style transfer rather than character consistency.

Getting the best reference image

The quality of your reference image directly determines --cref consistency. Best practices:

• Face forward, neutral expression: The model needs to clearly see facial features. Angled or obscured faces produce weaker consistency.
• Good lighting, no harsh shadows: Shadows on the face obscure features the model needs to extract.
• Clean background: A simple background reduces visual noise. White or solid dark backgrounds work well.
• No accessories obscuring features: Sunglasses, masks, or large hats reduce character consistency.
• Single subject: One clear subject in the frame. Group photos confuse the character extraction.

Combining --cref with --sref

The most powerful workflow combines both: --cref for character consistency and --sref for visual style consistency.

[scene description] --cref [character URL] --cw 80 --sref [style URL] --sw 80

This maintains both the character's appearance and the overall aesthetic across an entire project — useful for editorial series, comics, book covers, and brand photography.

Common --cref workflows

Editorial / fashion series

Woman in a red coat on a Paris street in autumn, golden hour --cref [url] --cw 80 --ar 2:3

Same woman at a café window, morning light, newspaper on table --cref [url] --cw 80 --ar 2:3

Woman running through Montmartre in rain --cref [url] --cw 80 --ar 2:3

Character design / illustration

Knight in black armor standing in a moonlit forest, fantasy setting, oil painting style --cref [url] --cw 70 --ar 2:3

Same knight in a medieval tavern, warm firelight, celebrating with villagers --cref [url] --cw 70 --ar 16:9

Brand / commercial photography

Product ambassador in a minimalist white studio, holding [product], friendly expression, commercial photography --cref [url] --cw 85 --ar 4:5 --style raw

Limitations to know

• --cref works best with photorealistic references. Illustrated characters, cartoons, or highly stylized references produce inconsistent results.
• Extreme pose changes (e.g., reference is front-facing, prompt asks for back-view) reduce consistency.
• --cref does not perfectly replicate clothing from the reference — it applies character features, not wardrobe.
• Very different art styles (e.g., photorealistic reference → anime output) will reduce character fidelity.

Using a prompt generator for character projects

Long character consistency projects involve writing many similar prompts with subtle variations. A Midjourney prompt generator can scaffold these variants for you — maintaining the --cref, --cw, and --sref parameters consistently across your entire series while varying only the scene description.