Using Reference Images for Character Consistency in Midjourney (Deep Dive) (Patreon)

If you’ve ever felt like Midjourney almost understands your character, but then randomly changes their face, age, or vibe, reference images are where things really start to click.

This post is a deeper look at how I use reference images for character consistency, what’s currently working in Midjourney, and the different ways you can supply those references depending on your workflow.

First: the mindset shift

Midjourney doesn’t remember characters, it interprets visual instructions.
Your job isn’t to make it “recall” someone, but to re-teach it the character every time, as clearly and consistently as possible.

That’s where reference images come in.

Ways to Use Reference Images in Discord

1. Uploading an image directly

This is the simplest and most reliable method.

How it works:

• Upload your character image directly into

• Copy the image link

• Paste it at the very beginning of your prompt

This works best when:

• The image is clean and well-lit

• The face is clearly visible

• There’s minimal background clutter

💡 Tip: Avoid dramatic lighting, extreme angles, or heavy effects in your base reference. Save those for later scenes.

2. Using a hosted image URL (PostImages, etc.)

If you don’t want to keep re-uploading files, hosting your image can make life easier.

Popular option:

• PostImages.org (free, simple, reliable)

Workflow:

1. Upload your reference image

2. Copy the direct image URL

3. Paste it at the start of your Midjourney prompt

This is especially useful if:

• You’re reusing the same character across many prompts

• You want to swap scenes quickly

• You’re working on longer projects (videos, comics, storyboards)

Same rule applies: URL first, prompt second.

How to Upload a Reference Image Using the Midjourney Website

If you prefer working directly in the Midjourney website (instead of Discord), uploading a reference image is actually very straightforward.

Step 1: Open the Midjourney Web App

Go to the Midjourney website and log into your account. From there, open the Create page where you normally enter prompts.

Step 2: Start a New Prompt

Click into the prompt box just like you normally would when generating an image.

Step 3: Upload Your Reference Image

Next to the prompt field, you’ll see an image upload icon (or a “+” icon, depending on your layout).

• Click the upload icon

• Select your reference image from your computer

• Wait for it to finish uploading

Once uploaded, the image will appear above or inside your prompt area as an attached reference.

💡 Tip: Use a clean, clear image with minimal background and neutral lighting for best results.

Step 4: Write Your Prompt Normally

After the image is attached, type your prompt as usual.

The uploaded image automatically becomes part of the prompt, so you don’t need to paste a URL or do anything special beyond that.

Example structure:

• Reference image attached

• Prompt text describing the scene, character, and style

Midjourney will use the uploaded image as visual guidance when generating your results.

Step 5: Adjust Reference Influence (Optional but Important)

Once your image is attached, Midjourney gives you an option to adjust how strongly it should follow the reference image.

Use this when:

• The face starts drifting

• The character doesn’t look close enough to the reference

• You want tighter consistency

Start with a moderate setting and increase only if needed. Too much influence can make images feel stiff or overly constrained.

Step 6: Generate and Review

Submit your prompt and review the results.

If the character isn’t consistent enough:

• Increase reference influence slightly

• Simplify your prompt

• Make sure your descriptive language matches the reference

Consistency usually improves in small adjustments, not big jumps.

Quick Notes From Experience

• Uploaded images are ideal for ongoing projects

• They’re especially helpful for character-driven work

• You’ll still want to repeat key character traits in text — the image helps, but language still matters

Reference Strength:

When I’m generating images in Midjourney’s Discord, I use --cref (character reference) and --cw (character weight) directly in my prompts to keep characters consistent across different scenes. These parameters let me control how closely Midjourney follows a reference image, which is especially useful for storytelling, age progression, or recurring characters.

If you’re using the Midjourney website, you won’t type --cref or --cw into the prompt. Instead, character consistency is handled through the reference slider (shown below, the annoyingly small redline in the lower right hand corner. I've circled it in yellow).

This lets you control:

• How closely Midjourney sticks to the reference image

• How much freedom it has to reinterpret the character

• There's also --oref = Omni-Reference (for discord users)

  • More of a general visual influence tool.

  • Can reference style, vibe, composition, or a subject, but it’s not as character-locked as --cref.

  • Think of it as: “Use this image as guidance” rather than “this is the same person.”

How I think about it:

• Lower influence → more variation, looser resemblance

• Higher influence → stronger facial structure and identity lock

For character consistency:

• Start moderate

• Increase only if the face starts drifting

• Avoid maxing it out unless you want very stiff results

You’re aiming for recognizable, not frozen.

What Makes a “Good” Character Reference Image

This matters more than people think.

Strong references usually have:

• Neutral expression

• Straight-on or slight angle

• Natural lighting

• No heavy stylization

• No dramatic shadows or color casts

Bad references (for consistency):

• Extreme emotion

• Heavy cinematic lighting

• Over-the-top styling

• Busy backgrounds

• Filters or painterly effects

Think “ID photo energy” — boring but effective 😅

My Personal Workflow:

1. Generate a clean base character image

2. Lock that image in as my main reference

3. Use consistent descriptive language alongside the image

4. Slowly introduce new environments, lighting, and emotion

5. Adjust reference influence only when drift starts happening

This is what lets me move a character through multiple scenes — and even through time — without losing who they are.

A Quick Reality Check (Because Honesty)

Even with perfect references:

• You’ll still get some variation

• Minor facial differences are normal

• Consistency improves through selection, not perfection

Midjourney is amazing, but it still needs guidance and sometimes a gentle hand holding.

Universal Character Consistency Formula:

[this character], [clear action or pose], [specific setting],

[environmental detail], [lighting], [emotional state], [style or medium]

Some Examples:

Everyday / Modern Settings

1. this woman, standing at a kitchen counter mid-movement, morning light spilling through a window, steam rising from a coffee mug, soft natural lighting, calm and focused, cinematic photo

2. this man, leaning against a city bus stop, checking the street ahead, light rain falling, cool overcast light, thoughtful and reserved, cinematic realism

Nature / Outdoor Settings

1. this man, standing at the edge of a forest trail, leaves shifting in the breeze, filtered sunlight through trees, alert but peaceful, cinematic nature photo

2. this woman, kneeling beside a riverbank, fingertips brushing the water, gentle mist rising, soft early morning light, serene and grounded, cinematic landscape portrait