Midjourney Character Sheet Guide: Consistent Brand Characters with Multi-Turn Prompting & Style References
Creating Consistent Brand Character Sheets in Midjourney
Maintaining visual consistency for brand characters across dozens of AI-generated images is one of the hardest challenges in AI art production. Midjourney offers a powerful toolkit—character references (—cref), style references (—sref), seed locking (—seed), and multi-turn prompting—that, when combined strategically, lets you build reproducible character sheets worthy of professional brand guidelines.
This guide walks you through the complete workflow, from initial character design to a locked-down, reusable reference system that produces consistent results across multiple generations.
Prerequisites
- An active Midjourney subscription (Standard or Pro recommended for fast generations)- Access to Midjourney via Discord or the Midjourney web app at
midjourney.com- A folder system for organizing reference images and seeds
Step-by-Step Workflow
Step 1: Establish the Base Character Design
Start by generating a strong initial character concept. Focus on clear, simple descriptions that define the core identity of your brand character.
/settings
Ensure you are on Midjourney Model Version 6.1 or later for best —cref support. Then craft your initial prompt:
/imagine prompt: a friendly fox mascot character, orange fur, wearing a blue scarf, standing upright, full body, character design sheet, white background, multiple poses —ar 3:2 —v 6.1
Review the four outputs carefully. Select the one that best represents your brand character. Click the corresponding U1–U4 button to upscale it. Save this image—it becomes your anchor reference.
Step 2: Lock the Seed for Reproducibility
After upscaling, retrieve the seed value by reacting to your upscaled image with the ✉️ envelope emoji. Midjourney will DM you the job details including the seed number.
Record this seed. For example:
Seed: 2847193650
Using the same seed with the same prompt and parameters produces near-identical results, giving you a stable baseline.
/imagine prompt: a friendly fox mascot character, orange fur, wearing a blue scarf, standing upright, full body, white background —ar 3:2 —v 6.1 —seed 2847193650
Step 3: Apply Character References (—cref)
Upload your anchor reference image to Discord or the Midjourney web app. Copy the image URL. Now use the --cref parameter to enforce character consistency in new generations:
/imagine prompt: a friendly fox mascot waving hello, standing in a park --ar 16:9 --v 6.1 --cref https://your-uploaded-anchor-image-url.png --cw 100
The --cw (character weight) parameter controls how strongly the character reference is enforced. Values range from 0 to 100:
| --cw Value | Effect | Best For |
|---|---|---|
| 100 | Strongest character match (face, body, clothing) | Strict brand consistency |
| 50 | Moderate match (face and general build) | Varied outfits, same character |
| 0 | Face only | Costume changes, seasonal variants |
--sref parameter. This is critical for brand consistency across different scenes.
/imagine prompt: a friendly fox mascot reading a book in a library --ar 16:9 --v 6.1 --cref https://your-anchor-image-url.png --cw 100 --sref https://your-style-reference-url.png --sw 100The --sw (style weight) parameter ranges from 0 to 1000 (default 100). Higher values enforce the reference style more aggressively.
Step 5: Multi-Turn Prompting for Character Sheet Variations
Now build out your full character sheet by systematically generating key poses and expressions. Use the same —cref, —sref, and —seed parameters across all prompts:
# Front view
/imagine prompt: fox mascot standing front view, arms at sides, neutral expression, white background —cref [URL] —cw 100 —sref [URL] —seed 2847193650 —v 6.1
Side profile
/imagine prompt: fox mascot standing side profile view, white background —cref [URL] —cw 100 —sref [URL] —seed 2847193650 —v 6.1
Expression sheet
/imagine prompt: fox mascot face expression sheet, happy sad surprised angry, white background —cref [URL] —cw 100 —sref [URL] —seed 2847193650 —v 6.1
Action pose
/imagine prompt: fox mascot running dynamic pose, white background —cref [URL] —cw 100 —sref [URL] —seed 2847193650 —v 6.1
Step 6: Build a Reference Library Document
Organize your finalized assets into a structured reference document. Track every generation with its parameters:
| Asset | Seed | --cw | --sw | Prompt Summary |
|-----------------|------------|------|------|------------------------|
| Anchor (front) | 2847193650 | 100 | 100 | standing, front view |
| Side profile | 2847193650 | 100 | 100 | standing, side profile |
| Expressions | 2847193650 | 100 | 100 | face expression sheet |
| Action pose | 2847193650 | 100 | 100 | running dynamic pose |
## Pro Tips
- **Stack multiple --cref URLs:** You can provide up to three character reference images for stronger identity locking. Separate URLs with spaces after --cref.- **Use --no for consistency:** Add --no realistic, photographic to maintain a consistent illustrated style and prevent style drift.- **Batch with the /prefer suffix command:** Store your reference parameters as a suffix so you don't have to type them every time: /prefer suffix --cref [URL] --cw 100 --sref [URL] --sw 100 --seed 2847193650 --v 6.1- **Combine --sref random with --cref:** When exploring new art directions while keeping the character, use --sref random to generate fresh styles that you can then lock down with a specific --sref URL.- **Export seeds programmatically:** Use the Midjourney web app's gallery to filter and export job metadata including seeds for large-scale character production.- **Permutation prompts for efficiency:** Generate multiple variations in a single command: /imagine prompt: fox mascot {waving, sitting, jumping}, white background --cref [URL] --cw 100
## Troubleshooting
| Problem | Cause | Solution |
|---|---|---|
Character looks different despite same --cref | Prompt wording conflicts with reference | Simplify the prompt; avoid describing character features already in the reference. Increase --cw to 100. |
| Style drifts between generations | Missing or low --sw value | Add --sref with your style anchor and set --sw to 200–500 for stronger enforcement. |
| Seed produces different results | Different model version or aspect ratio | Ensure --v and --ar are identical across all prompts. Seeds are version-specific. |
--cref URL not working | Image URL is not publicly accessible | Upload the image directly to Discord or use the Midjourney web app uploader. Avoid links that require authentication. |
| Character clothing keeps changing | --cw set too low | Increase --cw to 100 to include clothing and accessories in the reference lock. |
Can I use —cref and —sref together in the same prompt?
Yes. Combining --cref (character reference) and --sref (style reference) in a single prompt is the recommended approach for brand consistency. The character reference controls identity—face, body, clothing—while the style reference controls the artistic rendering. Use both with high weight values (--cw 100 and --sw 100 or higher) for maximum consistency across your character sheet.
How many reference images can I use with —cref?
Midjourney supports multiple reference image URLs with —cref. You can provide up to three images separated by spaces. Using multiple angles of the same character (front, side, three-quarter view) significantly improves consistency. You can also weight individual references by appending ::N to each URL, where N is the relative weight (e.g., —cref url1::2 url2::1 gives twice the weight to the first reference).
Will the same seed always produce identical results?
A seed produces near-identical results only when all other parameters remain constant—including the model version (—v), aspect ratio (—ar), prompt text, and any reference URLs. Changing any of these will alter the output even with the same seed. Additionally, Midjourney model updates can affect seed reproducibility over time, so it is best practice to generate all character sheet assets in a single session and save the final images rather than relying solely on seed numbers for long-term reproducibility.