A consistent character is the difference between a fun AI image and a usable visual story. If you have ever generated a brilliant Midjourney portrait, tried to place that same person in a new outfit or location, and watched their face slowly turn into a cousin of the original, you already know the problem. Character Reference is Midjourney’s answer to that problem: a practical way to keep a person, mascot, creature, or stylized hero recognizable across multiple prompts.

For creators, marketers, game designers, educators, and solopreneurs, this feature changes the workflow. Instead of spending hours rerolling images and hoping the same face appears again, you can guide Midjourney with a reference image and use parameters like --cref and --cw to control how strongly the character carries over. The result is not perfect identity locking, but it is dramatically better than prompt text alone, especially when you pair it with smart prompting and a repeatable production process.

In this tutorial, you will learn exactly how to use Character Reference in Midjourney, from choosing a good source image to building a multi-scene character set. We will use concrete examples, practical prompt structures, troubleshooting patterns, and pro tips that reflect real creator workflows in Discord and the Midjourney web interface. By the end, you will know how to generate a character once and reuse that character across portraits, action shots, product visuals, editorial images, and storyboards with far more control.

What Character Reference Does in Midjourney

Character Reference tells Midjourney to borrow the visual identity of a character from an image and apply it to a new generation. In practice, that means Midjourney pays attention to traits such as face shape, hairstyle, age impression, clothing cues, body type, color palette, and the overall vibe of the subject. It is especially useful when you want a recurring character in different poses, camera angles, settings, moods, or illustration styles.

The core parameter is --cref, which stands for character reference. You place it at the end of your prompt followed by an image reference. In Discord, that reference is usually an image address copied from an uploaded image. In the Midjourney web interface, the workflow can feel more visual because you can select or drag images into the prompt area, but the underlying idea is the same: the model uses your chosen image as a character anchor.

It is important to understand what Character Reference is not. It is not a guaranteed face-swap tool, a biometric identity system, or a perfect continuity machine. Midjourney interprets images creatively, so exact facial geometry may shift, especially under extreme changes in pose, lighting, age, or style. However, for fictional characters, brand mascots, social media personas, editorial illustrations, concept art, and marketing visuals, --cref can produce a consistent enough identity to support professional-looking visual series.

Character Reference vs Image Prompting

Traditional image prompting influences the whole image: composition, colors, style, subject matter, background, and mood. Character Reference is more targeted. When you use --cref, you are telling Midjourney, in effect, to preserve the character identity while allowing the rest of the prompt to change. This distinction matters because a standard image prompt might copy the reference composition too heavily, while a character reference is more useful for placing the same subject into a completely new scene.

You can also combine Character Reference with other Midjourney features. For example, you might use a character reference for identity, a style reference for the visual treatment, and a text prompt for the setting and action. A children’s book creator could keep the same fox protagonist with --cref, apply a watercolor look with a style reference, and then prompt scenes like forest picnic, moonlit library, or rainy city street. This layered approach is where the feature becomes genuinely powerful.

Prerequisites Before You Start

Before building a consistent character workflow, make sure you have access to Midjourney through either Discord or the Midjourney web interface. The exact buttons and layout may evolve, but the practical concepts remain stable: you need a prompt box, a way to upload or select images, and access to parameters such as --cref and --cw. If you are using Discord, you will also need to know how to upload an image, open it, and copy its image address for use in a prompt.

You should also have a clear idea of the character you want to produce. Midjourney can improvise, but consistency improves when your source image is clean and intentional. A strong reference image typically shows one primary character, has a visible face or defining body shape, avoids heavy occlusion, and does not contain multiple competing subjects. If you are working with a mascot, creature, robot, or fantasy hero, make sure the defining features are visible: horns, armor, color markings, silhouette, hairstyle, jacket, emblem, or accessory.

For best results, use a current Midjourney model that supports Character Reference well. Many creators see stronger prompt adherence and character handling in newer model versions than in older ones. If you work with anime or manga-style visuals, Niji models may be a better fit; if you work with cinematic, editorial, or realistic images, the standard Midjourney model is often more suitable. The key is to test one model at a time and avoid changing too many variables between generations.

Recommended Source Image Checklist

A usable character reference does not need to be technically perfect, but it should communicate identity clearly. A front-facing or three-quarter portrait is often the safest first choice because the model can read facial proportions, hair, clothing, and expression. Full-body references are useful when costume and silhouette matter, but they may provide less facial detail if the character is small in the frame.

  • Use one main subject so Midjourney does not blend traits from multiple people or characters.
  • Choose clear lighting because harsh shadows can distort facial structure and colors.
  • Avoid busy backgrounds when possible, especially for your first reference image.
  • Keep defining features visible, such as glasses, scars, braids, armor, jacket color, or mascot markings.
  • Prefer high-resolution images because a crisp reference gives the model more useful visual information.

You should also decide whether your goal is strict identity, flexible character branding, or broad inspiration. Strict identity means you want the same person or fictional hero to stay very recognizable. Flexible branding means you mainly care that the character feels like the same brand asset, even if minor facial details shift. Broad inspiration means you want the new image to borrow a general archetype, such as a silver-haired cyberpunk detective or a cheerful claymation robot, without needing exact continuity.

Step 1: Create or Select Your Character Reference Image

The first real step is to create a reliable anchor image. You can use an image generated in Midjourney, an illustration created in a tool like Procreate, a 3D render from Blender, or a licensed brand asset from your own design library. If you are using a real person, be careful with consent, likeness rights, and platform rules. For most creative and commercial workflows, the safest option is a fictional character you generate or own.

If you do not already have a character image, generate one in Midjourney before using Character Reference. Start with a direct prompt that defines the person or creature clearly. For example: portrait of a 32-year-old female space engineer, short black curly hair, amber utility jacket, small silver nose ring, warm confident expression, cinematic studio lighting, clean gray background. The clean background is not mandatory, but it helps produce a reference image that is easy to reuse.

Once Midjourney generates a grid, pick the image that has the strongest identity. Do not simply choose the prettiest image. Choose the one with the most memorable and repeatable features. If the face is partially hidden, the hair is ambiguous, or the outfit has chaotic details, it may look impressive once but become unstable across future prompts. A slightly simpler design often performs better when you need 10, 20, or 50 consistent scenes.

Build a Reference Sheet for Better Continuity

One pro-level workflow is to create a simple character reference sheet before producing scenes. Ask Midjourney for a front view, side view, and three-quarter view of the same character, or generate several portraits and choose the most consistent one as your main reference. You can later test whether one image or multiple references give better continuity. In many cases, a single clean portrait provides a stronger signal than a collage with too many competing details.

If costume matters, create a second reference focused on the outfit. For example, a game designer might keep one close-up face reference and one full-body costume reference. The close-up preserves facial identity, while the full-body image helps retain the red cloak, brass shoulder plates, leather boots, or asymmetrical gloves. This is useful when building character packs for pitch decks, visual novels, tabletop role-playing campaigns, or branded social media stories.

Name and organize your character assets from the beginning. Use practical filenames in your local folder, such as maya_space_engineer_face_v01, maya_full_body_suit_v02, and maya_expression_smile_v01. Midjourney itself may not care about your filenames, but your production workflow will. After a week of experimentation, clean file names can save you from guessing which image produced the best results.

  1. Generate or import a clean image of one character.
  2. Choose the strongest identity, not just the most decorative image.
  3. Save the image in a dedicated project folder with a descriptive name.
  4. Upload it to Discord or select it in the Midjourney web interface.
  5. Test it with one simple scene before building a full campaign.

Step 2: Add the --cref Parameter to Your Prompt

After your reference image is ready, the next step is using it in a prompt with --cref. The basic structure is: describe the new image you want, then add the character reference parameter at the end. A practical example would be: cinematic photo of the character repairing a damaged rover on Mars, orange dust storm in the distance, practical sci-fi suit, dramatic rim light --cref [character image]. In Discord, the bracketed part would be replaced by the copied image address or inserted image reference.

The text prompt still matters. Character Reference does not replace good prompting; it enhances it. If your prompt says the character is a medieval knight, Midjourney may alter the outfit heavily even if the reference wears a modern jacket. If your goal is to keep the jacket, say so. A better prompt might be: the same character wearing her amber utility jacket over a light spacesuit, standing inside a Mars research garage, cinematic realism, 35mm lens, shallow depth of field --cref [character image].

Start with simple prompt changes. First test the character in a similar portrait, then a half-body scene, then a full-body action shot. This staged approach helps you see when the identity begins to drift. If you immediately ask for underwater lighting, a helmet visor, a different age, a new hairstyle, a wide-angle action pose, and a painterly style, you will not know which variable caused the character to lose consistency.

Prompt Template You Can Reuse

A reliable template is: [medium] of the same character, [action or pose], [setting], wearing [important clothing], [mood], [camera or art direction] --cref [reference]. This template keeps identity language near the front, preserves costume details, and gives Midjourney a clear job. You can adapt it for photography, illustration, 3D render, comic art, product mockups, or poster design.

For example, a brand team creating an AI mascot might use: 3D render of the same friendly teal robot mascot, waving beside a laptop on a clean desk, rounded body, glowing eyes, soft studio lighting, minimal startup brand aesthetic --cref [robot reference]. A fiction writer might use: editorial fantasy illustration of the same elderly mapmaker, reading a glowing compass in a candlelit archive, blue velvet coat, silver beard, intricate shelves, warm magical atmosphere --cref [mapmaker reference].

Keep your prompts consistent when testing. If you are comparing four outputs, avoid changing the model version, aspect ratio, stylization level, reference image, and prompt wording all at once. A disciplined test might generate four variations using the same prompt and only adjust --cw, which controls character reference weight. This gives you useful information instead of random impressions.

Step 3: Control Consistency with --cw

The --cw parameter stands for character weight. It controls how strongly Midjourney should apply the character reference. The scale commonly runs from low influence to high influence, with higher values preserving more of the reference character. At high character weight, Midjourney may hold onto clothing, hair, face, and general appearance more aggressively. At lower weight, it may keep only the general identity or vibe while allowing more transformation.

A typical default behavior aims to carry over the character strongly, but not every project needs maximum consistency. If you are creating a comic sequence where the same heroine appears in multiple panels, you may want a higher character weight. If you are creating a campaign where the same mascot appears as a chef, astronaut, skateboarder, and teacher, a slightly lower weight may give Midjourney room to change costumes and props while retaining the mascot’s face and silhouette.

Think of --cw as a creative dial rather than a magic fix. If the identity is drifting, raising character weight can help, but it can also make the image cling too closely to the source pose or outfit. If the outfit refuses to change, lowering character weight can help, but it may weaken facial continuity. The best value depends on the reference image, prompt complexity, model version, and the amount of transformation you are asking for.

Practical --cw Testing Workflow

Run a simple three-part test before committing to a big batch. Use the same prompt and same reference image, but generate versions with low, medium, and high character weight. Compare the face, hair, clothing, silhouette, and overall personality. This test takes only a few minutes but can prevent hours of inconsistent generations later.

For example, use one prompt such as: the same character sitting in a neon-lit diner at midnight, holding a cup of coffee, cinematic cyberpunk mood, rain on the window --cref [character reference] --cw 25. Then run the same prompt with --cw 50 and --cw 100. If the low value captures the mood but loses the face, and the high value preserves the face but keeps the original clothing too much, the middle value may be your production setting.

Keep notes in a tool like Notion, Google Docs, Airtable, or a simple spreadsheet. Record the prompt, model version, aspect ratio, reference image, character weight, and your rating for identity consistency. A practical scoring system from 1 to 5 works well: 1 means the character is unrecognizable, 3 means usable with minor drift, and 5 means highly consistent. After 20 tests, you will have real data instead of relying on memory.

Step 4: Build Scenes, Expressions, and Outfits

Once you have a working reference and character weight, you can begin producing useful scene variations. This is where Character Reference becomes a productivity hack rather than a novelty. A social media manager can create a month of posts featuring the same AI spokesperson. A game designer can visualize character classes and environments. A course creator can build a friendly guide character for slides, thumbnails, and worksheets.

Start by changing one category at a time: setting, pose, expression, outfit, camera angle, or art style. If you change all of them at once, continuity will suffer and troubleshooting becomes difficult. For instance, first place the same character in a laboratory, office, desert, and train station while keeping the outfit consistent. Then test expressions like laughing, worried, determined, and surprised. Finally test outfit changes after you know the face is stable.

When prompting expressions, be explicit but realistic. Phrases like subtle smile, focused expression, looking exhausted but hopeful, or laughing naturally tend to produce better results than vague emotional labels. For facial continuity, moderate expressions are often safer than extreme ones. A wide scream, heavy squint, or exaggerated cartoon laugh can change the face enough that the character appears less consistent.

Example Production Plan for a Character Series

Imagine you are creating a fictional AI productivity coach named Nova for a startup blog. Your reference image is a half-body portrait of a young professional with silver bobbed hair, round glasses, a cobalt blazer, and a calm, confident expression. Your content calendar needs header images for topics like email automation, meeting notes, coding assistants, and prompt engineering. Character Reference lets you reuse Nova across the entire series without redesigning her from scratch.

Your first batch might keep the blazer and glasses in every image. Prompts could include Nova standing beside a holographic kanban board, Nova reviewing automated email flows on a laptop, Nova presenting a dashboard in a modern conference room, and Nova holding sticky notes in a bright coworking space. Because the clothing and face remain visible, the audience begins to recognize the character as a recurring editorial asset.

Your second batch can push variety. Put Nova in a winter office, a late-night coding scene, a podcast studio, or a clean isometric workspace. If the identity holds, you can safely expand. If it drifts, return to the reference image, simplify the scene, increase character weight, or mention the defining traits directly in the prompt: silver bobbed hair, round glasses, cobalt blazer. Text reinforcement often helps the visual reference stay on track.

Pro Tips for Better Character Reference Results

The best Character Reference workflows combine visual anchors with written anchors. Do not rely entirely on --cref. Repeat the character’s defining features in the prompt, especially the features you cannot afford to lose. If your character is defined by red glasses, a green scarf, a white streak in black hair, or a crescent-shaped facial mark, write those details into the prompt. Midjourney reads both the reference image and your text instructions, so make them work together.

Use aspect ratios deliberately. A close portrait in a square format often preserves identity better than a wide cinematic frame where the character occupies only 10 percent of the image. If you need a banner image, consider generating a medium shot where the character is still large and recognizable, then expand or crop afterward if your workflow allows it. The more visual space the character occupies, the more opportunity Midjourney has to render consistent details.

Another advanced tactic is to separate creative exploration from production. During exploration, use higher stylization, unusual lighting, and dramatic scenes to discover what is possible. During production, reduce the number of variables and use a proven prompt structure. Many professional creators maintain two prompt libraries: one for experimentation and one for repeatable outputs. This habit is especially useful when working with client-facing assets or brand campaigns.

Use Multiple References Carefully

Multiple character references can help when one image does not communicate enough information, but they can also confuse the model. If you provide a face close-up, a full-body image, and a different expression, make sure they are clearly the same character. If the hair length, clothing, age, or facial proportions differ too much, Midjourney may average them into an unstable hybrid.

A good multiple-reference setup might include one clean portrait and one full-body costume sheet. A risky setup would include three beautiful images that all look slightly different. For consistent storytelling, boring clarity beats artistic variety. You want the model to receive a strong signal, not a mood board with contradictions.

When using multiple references, test them against a neutral prompt before using them in complex scenes. For example: studio portrait of the same character, neutral gray background, soft lighting --cref [reference one] [reference two]. If Midjourney cannot maintain identity in a simple portrait, it will struggle even more in a crowded action scene with complex props, weather, and camera movement.

Troubleshooting Common Problems

Even with a strong reference, Character Reference can produce frustrating results. The most common issue is identity drift, where the character looks similar but not the same. When that happens, simplify the prompt. Remove unnecessary style words, reduce extreme camera angles, keep the face visible, and mention defining traits directly. If your character has short curly black hair and amber jacket, say that in the prompt instead of assuming the reference will carry it perfectly.

If Midjourney keeps copying the original outfit when you want a new costume, lower --cw or describe the new outfit more strongly. You can also keep a few identity traits while allowing wardrobe changes: same face, same short curly black hair, same warm expression, now wearing a white lab coat over a navy shirt. This tells the model what to preserve and what to change. If the old clothing still appears, the source reference may be too outfit-dominant, so create a cleaner face reference.

If the character appears in the wrong age, gender presentation, or species, check whether your prompt is fighting the reference. A prompt asking for a child version, elderly version, robot version, or fantasy creature version may transform the identity more than expected. That can be useful for creative variants, but it reduces strict continuity. For production assets, keep the demographic and physical fundamentals stable unless you intentionally want an alternate version.

Problem-Solution Quick Guide

When the face changes too much, increase character weight, use a closer crop, and keep the character larger in the frame. When the pose is too similar to the source image, lower character weight or describe a stronger action pose. When the image style overwhelms the character, reduce stylization language and test a more neutral rendering. When the background from the reference keeps appearing, use a cleaner reference image or state a new environment clearly.

Hands, accessories, and logos can also cause issues. Midjourney may preserve the idea of an accessory but alter its shape, placement, or text. If your character always wears a badge, a pendant, or a branded jacket, generate several options and choose the one that is closest rather than expecting perfect reproduction. For logos and readable text, use a dedicated design tool after generation if accuracy matters.

Finally, watch for overfitting to one image. If every output feels like the original portrait with a slightly different background, you may be using too much character weight or too restrictive a prompt. The goal is not to freeze the character like a sticker; the goal is to make the character feel alive across situations. Balance continuity with variation, and you will get more believable results.

Best Workflow for Teams and Repeat Projects

If you are using Character Reference for a team project, treat it like a mini style system. Create a character bible that includes the approved reference image, key visual traits, preferred prompt phrases, model version, aspect ratios, character weight ranges, and examples of approved outputs. This prevents every designer, writer, or marketer from reinventing the prompt and accidentally changing the character’s identity.

A practical team workflow starts with one owner generating and approving the master reference. Then the team tests a small set of use cases: portrait, full-body, action scene, product scene, and banner composition. Once the results are acceptable, document the winning prompts. Store outputs in a shared folder with labels such as approved, needs edit, and rejected. This may sound formal, but it is exactly how you turn AI image generation from random experimentation into a repeatable production pipeline.

For high-volume content, build batches around consistent requirements. For example, an editorial team might need 12 blog headers per month featuring the same character. Instead of prompting from scratch each time, keep a base prompt and swap only the topic, setting, and props. This improves consistency and makes review faster. If a new batch suddenly looks different, you can compare it against the documented settings and identify what changed.

Quality Control Checklist

Before publishing or delivering Character Reference images, review them for continuity and usability. Ask whether the face is recognizable, the core traits are present, the character’s age and personality feel consistent, and the image fits the intended format. Also check for common AI artifacts such as distorted hands, mismatched earrings, warped glasses, extra buttons, strange teeth, or inconsistent shadows.

For commercial use, add a human editing pass. Tools like Photoshop, Affinity Photo, Canva, Figma, and Lightroom can help clean small artifacts, match colors, crop compositions, and add accurate typography. Character Reference gets you much closer to the desired image, but finishing tools are still important when the output represents a brand, product, or paid client deliverable.

Keep version history. If your character evolves, label the new reference as version two instead of silently replacing the old one. This matters when you need to revisit older campaigns or maintain continuity across seasons. A simple versioning habit can prevent a character from slowly drifting over months of content production.

Key Takeaways

Character Reference in Midjourney is one of the most useful features for creators who need repeatable AI visuals. By using --cref, you can guide Midjourney to preserve a character’s identity across new scenes, outfits, moods, and compositions. It is not flawless identity locking, but it is a major upgrade over relying on text prompts alone.

The best results come from clean references, controlled testing, and clear prompts. Start with one strong character image, use a simple scene, test character weight values, and document what works. Once your workflow is stable, expand into expressions, action shots, editorial scenes, product visuals, and campaign assets. Consistency is built through process, not luck.

If you remember only one principle, make it this: combine visual reference with written specificity. Tell Midjourney what the image shows, what must stay the same, and what should change. That combination gives you the best chance of creating a character that audiences recognize from one image to the next.

  • Use --cref to apply a character’s visual identity to new Midjourney generations.
  • Use --cw to control how strongly the reference character influences the output.
  • Start with a clean reference featuring one clear subject and visible defining traits.
  • Change one variable at a time when testing scenes, outfits, poses, and styles.
  • Document your settings so you can repeat successful results across teams and projects.
  • Troubleshoot drift by simplifying prompts, reinforcing key traits, adjusting character weight, and using better reference images.