⚡ Promptolis Original · Creative & Arts
👤 Midjourney Character Consistency — --cref Workflow for Multi-Image Projects
Keep the same character across 30+ images. The --cref workflow that enables children's books, graphic novels, brand mascots.
Why this is epic
Before --cref (character reference) dropped in v6, character consistency in Midjourney was nearly impossible — same text prompts produced different-looking people each generation. --cref solved this. This prompt walks through the complete workflow: establishing a strong character anchor image, using --cref across follow-up prompts, tuning --cw (character weight) for strict vs. flexible consistency.
Takes your character description + project scope (single scene / series / full book / brand mascot) → outputs: character anchor prompt with multi-angle considerations, --cref + --cw workflow for follow-ups, wardrobe/expression variation strategy, and common-pitfall troubleshooting (face drift, age drift, ethnic ambiguity).
Handles the 2026 reality: --cref works best for stylized characters and for realistic characters where the character is semi-distinctive (specific features). Generic 'man with brown hair' anchors drift more than 'woman with sharp jawline, close-set green eyes, shoulder-length auburn waves, slight asymmetry in smile.'
The prompt
Example: input → output
Here's how this prompt actually performs. Real input below, real output from Claude Opus 4.
<character-description>Elena — a 9-year-old girl, mixed Mexican-Irish heritage, shoulder-length dark curly hair, green eyes, freckles across nose, mischievous smile, usually in overalls and t-shirts, looks like she's about to do something clever</character-description> <style>Watercolor children's book illustration, soft pastel palette, warm + bright</style> <project-scope>Book-length — 25 scenes</project-scope> <consistency-need>Tight (book illustration — same character feels like same character across every page)</consistency-need> <midjourney-version>v7</midjourney-version>
Character Anchor Strategy
For book-length children's illustration, you need a 6-image anchor library: 3 angles (front / side / 3-quarter) × 2 expressions (neutral / mischievous grin). Then use --cref + --sref for all 25 scenes. Watercolor illustration holds --cref tighter than photorealistic would — stylized works in your favor.
The character description is specific enough (mixed heritage, specific features, specific outfit register) to lock well. Freckles and the 'mischievous smile' are distinctive anchors Midjourney will remember.
Anchor Prompt (the main anchor — front view, mischievous)
Elena, 9-year-old girl, mixed Mexican-Irish heritage, shoulder-length dark curly hair, bright green eyes, freckles across nose, mischievous grin, wearing denim overalls and striped t-shirt, watercolor children's book illustration style, soft pastel palette, warm natural light, front-facing portrait composition, clean white background, whimsical children's book aesthetic --ar 1:1 --stylize 500 --v 7Generate this 4-6 times. Pick the single strongest for --cref. Don't use a 'pretty-but-drifted' result; use the one closest to your description.
Multi-Angle Anchor Set (generate each 2-3 times)
Anchor 2 — Side Profile, Neutral
Elena, 9-year-old girl, mixed Mexican-Irish heritage, shoulder-length dark curly hair, green eyes, freckles, side profile view, neutral expression, wearing denim overalls striped t-shirt, watercolor children's book illustration, soft pastel palette, clean white background --ar 1:1 --stylize 500 --v 7Anchor 3 — 3/4 Angle, Curious
Elena, 9-year-old girl, mixed Mexican-Irish heritage, shoulder-length dark curly hair, green eyes, freckles, three-quarter angle view, curious expression head tilted, wearing denim overalls striped t-shirt, watercolor children's book illustration, soft pastel palette, clean white background --ar 1:1 --stylize 500 --v 7Anchor 4 — Full Body, Running
Elena, 9-year-old girl, mixed Mexican-Irish heritage, full body mid-run, dark curly hair in motion, green eyes, freckles, wearing denim overalls striped t-shirt, watercolor children's book illustration, soft pastel palette, dynamic pose, white background --ar 2:3 --stylize 500 --v 7Anchor 5 — Close-Up, Laughing
Elena, 9-year-old girl, mixed Mexican-Irish heritage, close-up portrait laughing mouth open, shoulder-length dark curly hair, green eyes crinkled with laugh, freckles visible, watercolor children's book illustration, soft pastel palette, warm light --ar 1:1 --stylize 500 --v 7Anchor 6 — Thoughtful, Reading
Elena, 9-year-old girl, mixed Mexican-Irish heritage, shoulder-length dark curly hair, green eyes, freckles, thoughtful expression reading a book, wearing denim overalls striped t-shirt, watercolor children's book illustration, soft pastel palette, soft window light --ar 3:2 --stylize 500 --v 7Pick the strongest 3-4 of these anchors. Your 'primary' anchor URL = the one you think best represents Elena. Use it as the main --cref. The others are secondary references for scenes that need those angles/expressions specifically.
Follow-Up Prompt Template (for book scenes)
For each of your 25 scenes:
[PRIMARY_ANCHOR_URL] Elena in [specific scene description — where, what she's doing, what's in scene], 9-year-old girl with shoulder-length dark curly hair and freckles wearing denim overalls striped t-shirt, watercolor children's book illustration, soft pastel palette, [scene-appropriate lighting] --ar [page-ratio] --style [stylized watercolor] --cref [PRIMARY_ANCHOR_URL] --cw 90 --sref [PRIMARY_ANCHOR_URL] --sw 80 --v 7Example for specific scene (Elena climbing a tree):
[URL] Elena climbing an oak tree in summer afternoon light, looking down with excitement toward ground, her denim overalls and striped t-shirt bright against leaves, watercolor children's book illustration, soft pastel palette dominant green and warm yellow, dappled sunlight through leaves --ar 2:3 --cref [URL] --cw 90 --sref [URL] --sw 80 --v 7--cw Tuning Guidance
- --cw 100: Strict face lock. Use for close-up portrait scenes where face recognition matters most.
- --cw 80-90: Default for book work. Strong character consistency while allowing natural scene variation.
- --cw 50-70: When character is in distinctive pose/activity (swimming, upside-down, in shadow) — relaxing weight helps pose render better.
- --cw 30-50: Rarely in book work. Use only if you want a sibling-character vibe, not exact same Elena.
Start --cw 90 for all scenes. If first scene result has face problem, raise to 100. If pose feels rigid or face-dominant-at-expense-of-scene, lower to 70.
Drift Prevention (Common with Children's Book Characters)
Age drift: Elena may render as 11-12 or 6-7 in some scenes. Always include '9-year-old' or 'a nine-year-old' in EVERY scene prompt. Don't let Midjourney guess age; state it.
Hair drift: Curly hair can go from tightly-curled to loose waves. Always 'dark curly hair' in scene descriptions. If renders are too straight, add 'tight curls' or 'coiled curls.'
Expression drift: Elena's 'mischievous' personality may render as 'sad' or 'angry' in scenes where she should be happy. State expression explicitly in each scene: 'Elena smiling,' 'Elena with delighted grin,' 'Elena focused but happy.'
Ethnicity drift: Mixed heritage may render as more-one, less-other in some scenes. Stating 'mixed Mexican-Irish' explicitly in each scene prompt helps. If drift persists, include 'warm olive skin tone, slight freckles' as consistent descriptors.
Outfit drift: Even stating 'denim overalls striped t-shirt' in every prompt, outfit may occasionally drift (different shirt, different coveralls). For book consistency: if outfit matters, restate full outfit every time. If outfit doesn't matter for a specific scene: can drop it and let Midjourney dress Elena appropriately for scene.
Workflow Integration
1. Create an 'Elena Character' Notion page or shared doc with:
- Primary anchor URL (the main --cref)
- Secondary anchor URLs (side profile, 3/4, close-up, running, reading) for specific-pose scenes
- Character description block you copy into each prompt
- List of consistent descriptors that lock character (age, hair, eyes, freckles, outfit)
2. Template your scene prompts. Don't write each from scratch; use the template above, fill in [scene] specifics.
3. Budget anchor time. First 30-60 minutes on anchors. After that, each scene is 2-5 minutes generate + iterate.
4. Scene-by-scene iteration. Generate scene → if drift is acceptable → move on. If drift is significant → re-generate with stronger --cw or clearer description. Don't perfect scene 1 forever; iterate through whole book then refine.
5. Flag-and-return workflow. Mark scenes that are 'close but not quite' for later revision. After all 25 scenes are 80% done, revisit the flagged ones.
Common use cases
- Children's book illustration — one character across 20-40 scenes
- Graphic novel / comic — main character + supporting cast across panels
- Brand mascot — consistent character across marketing assets
- Storybook app — character in various interactions/scenes
- Educational content — consistent 'guide' character across course materials
- Personal avatar / author photo variations for social media
- Concept artist character development for games / film (consistent hero design through iterations)
- Marketing campaigns featuring same model/spokesperson across assets
Best AI model for this
Claude Sonnet 4.5 or GPT-5 for prompt construction. This is text-prompt work, not image generation.
Pro tips
- Character anchor needs 4-6 iterations minimum. Generate multiple close-ups + full-body shots. Pick the strongest for --cref reference.
- --cw (character weight) 100 = strict face match. 50 = balanced. 0-30 = loose 'inspired by.' Start 80-100; relax if outputs feel too rigid.
- Specific, non-generic features help --cref lock in. 'Close-set blue eyes, strong jaw, slight scar above right eyebrow' > 'handsome man with blue eyes.'
- For children's book: generate character in 5-6 emotional expressions + 3 angles BEFORE starting scenes. Reference library of characters = consistency across book.
- Age drift is common. If character looks younger or older than intended, explicitly say 'mid-30s' / '8-year-old' in every follow-up prompt.
- Ethnicity / features drift if ambiguous. Specific non-stereotypical descriptors work best. Avoid both generic ('Asian man') and stereotypical descriptors.
- --cref + --sref stack: character consistency + style consistency in same prompt. Powerful for illustrated books.
Customization tips
- For multi-character books (protagonist + 2-3 supporting): build anchor library for EACH. Use primary character's --cref; supporting characters may need their own --cref when they're focus.
- For children with specific disabilities / adaptive equipment: include in character description and anchor. Wheelchair, hearing aid, glasses, walker — these need explicit inclusion or they drift. Inclusive representation requires intentional anchor work.
- For characters aging across a book (Elena at 9, then 12, then 16): generate separate anchor sets for each age. Each age has own --cref. Don't try single --cref across age changes.
- For adult characters (storybook for adults, graphic novel, brand): same workflow applies. Specific features still matter. 'Handsome man in 40s' drifts; 'man in 40s, sharp cheekbones, stubble, small scar on left temple, slightly graying temples' holds.
- For animal characters (mascots, talking animals in kids' books): --cref works for animals too. Specific markings matter — 'orange tabby cat with white chest and notched left ear' > 'orange cat.'
- For photorealistic model consistency (marketing spokesperson, brand face): --cref works but drift is real. For critical-consistency shoots (all same person), consider real photography with AI backgrounds rather than full-AI generation. Hybrid workflow often more reliable.
- For comic panels (graphic novel): --cref + dynamic pose description. Character in action needs explicit 'running,' 'jumping,' 'crouching' — otherwise Midjourney defaults to standing poses. Dynamic work needs more prompt effort.
Variants
Default Character Workflow
Standard anchor + --cref for series/book
Children's Book Multi-Angle
Full character design including 3 angles + 5 expressions before scenes
Graphic Novel Panel
Character in action poses, dramatic lighting, comic-style consistency
Brand Mascot
Commercial character for cross-platform use, tight consistency required
Realistic Model / Spokesperson
Photorealistic person consistency (harder than stylized)
Frequently asked questions
How do I use the Midjourney Character Consistency — --cref Workflow for Multi-Image Projects prompt?
Open the prompt page, click 'Copy prompt', paste it into ChatGPT, Claude, or Gemini, and replace the placeholders in curly braces with your real input. The prompt is also launchable directly in each model with one click.
Which AI model works best with Midjourney Character Consistency — --cref Workflow for Multi-Image Projects?
Claude Sonnet 4.5 or GPT-5 for prompt construction. This is text-prompt work, not image generation.
Can I customize the Midjourney Character Consistency — --cref Workflow for Multi-Image Projects prompt for my use case?
Yes — every Promptolis Original is designed to be customized. Key levers: Character anchor needs 4-6 iterations minimum. Generate multiple close-ups + full-body shots. Pick the strongest for --cref reference.; --cw (character weight) 100 = strict face match. 50 = balanced. 0-30 = loose 'inspired by.' Start 80-100; relax if outputs feel too rigid.
Explore more Originals
Hand-crafted 2026-grade prompts that actually change how you work.
← All Promptolis Originals