Loading tutorials…
Loading tutorials…
--sref is the difference between "I got a great image once" and "I can ship 100 on-brand images per campaign." This is the workflow that production AI image teams actually use.
Who this is forAnyone shipping multi-image content for the same brand or campaign — paid social ads, blog hero libraries, e-commerce product backdrops, brand lifestyle libraries. If consistency matters, --sref matters.
What you'll need
Step 1
--sref controls visual style. --cref controls character/face. --iref controls full composition. Pick the right one for the job.
--sref (Style Reference): inherits color palette, lighting, texture, mood. Does NOT lock subject or composition. Best for brand visual consistency across different scenes.
--cref (Character Reference): locks a specific face/character across generations. Use for recurring brand characters or talent likenesses.
--iref (Image Reference): inherits BOTH style and composition. Strongest match to source image. Use when you want near-recreation with controlled variation.
For brand-consistent ad creative, --sref is what you want 90% of the time. The other two are for specific edge cases.
All three accept image URLs OR numeric codes (we cover codes in step 4).
Step 2
Reference URLs must be direct image links Midjourney can fetch. Upload to your Midjourney library, use Discord uploads, or host on a public image URL.
Easiest path: in the Midjourney web app, drag any image into the prompt bar. It uploads and generates a Midjourney-hosted URL automatically.
Alternative: use any direct .jpg or .png URL on the public web (your CDN, Imgur direct link, etc.). Note that some hosts block hotlinking.
For brand work, upload reference images to a private CDN or your Midjourney library — do not rely on third-party hosts that may break.
Image quality matters: use 1024x1024+ references. Tiny thumbnail references produce muddy style transfer.
Build a "reference library" folder. 5-10 references per visual mode, named by mode ("hero-editorial-warm", "product-clean-studio").
Step 3
--sref goes at the START of the prompt, before subject text. You can chain multiple URLs and use --sw to control influence strength.
Single reference: "--sref https://yourcdn.com/ref1.jpg [your subject + context + style + parameters]"
Multiple references (blends styles): "--sref https://ref1.jpg https://ref2.jpg [prompt]"
Style weight (--sw): controls how strongly the reference influences output. Range 0-1000. Default 100.
--sw 25-50: subtle style inheritance. Subject prompt dominates.
--sw 100 (default): balanced. Style is clearly present but subject prompt is honored.
--sw 250-500: strong inheritance. Reference dominates; subject prompt becomes secondary.
--sw 1000: maximum match. Output looks nearly identical in style to reference.
For brand consistency, --sw 100-200 is usually right. Start at 100 and adjust.
Step 4
Midjourney has billions of numeric style codes. --sref random generates a random style; the resulting code can be saved and reused for that exact look.
Run any prompt with "--sref random" appended. The output uses a randomly-generated style.
Each generated image shows its SREF code in the parameters. Copy it — it looks like "--sref 3847291056" (a long number).
Save the code with a name and what it produces ("3847291056 = warm editorial film, golden hour bias, slight grain").
Re-use the code in any future prompt: "[your prompt] --sref 3847291056 --sw 100"
Numeric codes are MORE consistent than image URLs because they reference a fixed style vector, not an interpretation of a specific image.
Build a personal "style code library" — 10-20 codes covering your brand modes. This is the production AI image team standard.
Step 5
Advanced workflow: stack a numeric code (for color/lighting style) with an image URL (for compositional reference). Two-axis control.
Pattern: "--sref [numeric code] [image URL] --sw 100 [your prompt]"
The numeric code anchors color palette and lighting mood.
The image URL nudges composition and texture.
Adjust --sw to balance how much each contributes. --sw 150 gives more weight to the combo; --sw 50 lets your text prompt dominate.
This is how pros get "the brand look" reliably across 50+ generations. Hard to replicate without the right reference combinations.
Step 6
Document which codes/URLs you use for each visual mode. This becomes your team's prompt template library.
Create a doc (Notion, Airtable, even a Google Sheet). Columns: Mode, SREF code/URL, Use case, Stylize value, Example image link.
Modes to define: hero/editorial, product-clean, product-lifestyle, brand-mood, ad-creative-warm, ad-creative-cool, illustrated-flat.
Per mode, document the exact prompt fragment: e.g., "Hero editorial: --sref 3847291056 --sw 150 --stylize 100"
Train anyone else on your team to use this playbook. Now Midjourney output is reproducible across operators.
Update quarterly. As campaigns evolve, retire old modes and add new ones.
Step 7
Generate 3 different subjects, each in 4 variations, using the same SREF. If outputs look like one brand, your playbook is working.
Pick 3 different subjects from your campaign (a product shot, a person, a workspace).
Generate each with the same SREF + stylize + --sw values.
Lay all 12 images out side by side.
They should look like they came from one photographer/illustrator working on one campaign.
If they look like 12 random images, your reference is too weak or your subject prompts are dominating. Lower stylize or raise --sw.
Common mistakes
Using --sref but treating it as a magic style word
What goes wrong: You add --sref to one prompt, see slight improvement, do not understand why, and skip it on the next 10 prompts. Brand consistency stays poor; you blame Midjourney instead of your workflow.
How to avoid: --sref every prompt for brand work. Treat reference URLs/codes as a non-negotiable part of your prompt template, not a bonus.
Picking reference images that are not brand-aligned
What goes wrong: You pull "cool looking" Pinterest references that have nothing to do with your actual brand. Output looks polished but off-brand. Client rejects.
How to avoid: Reference images must be approved by your brand guidelines. Use your own past photoshoots, your brand mood board, or competitor references your brand actively wants to match.
Stacking too many references
What goes wrong: You add 4 reference URLs hoping to blend the best of each. Midjourney averages them all and produces a muddy mid-point of nothing. Output is worse than any single reference would have produced.
How to avoid: Two references maximum. If you want variety, run separate batches with different single references, not one batch with stacked references.
Ignoring --sw (style weight)
What goes wrong: You use default --sw 100 for everything. For some modes it under-influences, for others it over-dominates. Output inconsistent across campaigns.
How to avoid: Document --sw per mode in your playbook. Strong brand looks: --sw 150-300. Subject-led shots: --sw 50-100. Tune per mode, not globally.
Not saving numeric codes
What goes wrong: You generate an image you love with --sref random, do not note the code, lose it forever. You spend the next week trying to recreate it via prompt variations and never get back to the same look.
How to avoid: Every image with --sref random: immediately copy the code into your library doc with a description. Five seconds of admin saves hours of regret.
Using broken or expired reference URLs
What goes wrong: You batch-generate 50 images using a Discord-hosted URL that expires mid-job. Half your images use the reference; half do not. Brand consistency collapses silently.
How to avoid: For production work, always reference images hosted in stable locations (Midjourney library, your CDN). Test the URL is live before kicking off a batch.
Recap
Done — what's next
Midjourney prompt engineering basics (V7, 2026)
Read the next tutorial
Hand it off
Building a real brand SREF library is a one-time project that pays back for years. Most founders cannot dedicate 2 weeks to nail it themselves. An AI image specialist on EverestX builds your library in 4-6 hours for $60-100, then can run ongoing campaign batches at $14-16/hr.
See specialist rates
--sref inherits style only (color, light, mood) — your subject prompt controls what is in the image. --iref inherits style AND composition — output looks much closer to the reference image. Use --sref for brand consistency across varied scenes; --iref for near-recreation with controlled changes.
Run prompts with "--sref random" to discover new codes. Many designers also share their favorite codes on Twitter, Reddit, and dedicated SREF galleries (search "Midjourney SREF library"). Always test a code with your own prompts before adding to your playbook.
Numeric SREF codes were introduced in V6 and refined in V7. Image URL references via --sref work back to V5 with limitations. For full --sref capability, stay on V7.
Technically up to ~10. Practically, two is the maximum for production work. Beyond two, Midjourney averages all references into a muddy mid-point that rarely matches any of them well.
Most common cause: --sw is too low. Try raising to 150-300. Also check that your subject prompt is not actively fighting the reference (e.g., reference is warm-toned but you wrote "cool blue lighting" in the prompt).
Midjourney
Most "ultimate Midjourney prompt guides" are V5-era cargo cult. This walks through the V7-native prompt structure professionals actually use — and the prompt patterns that consistently produce on-brand commercial imagery.
Midjourney
Brand consistency in Midjourney is not luck. It is a system: SREF library, prompt templates, color/lighting rules, and a QA pass. This is the production system pros use.
Midjourney
Generating one good ad image is easy. Generating 50 ad variations that are all on-brand, on-message, and ready for Meta or TikTok is a workflow problem, not a prompting problem. This is how production teams do it.
Midjourney
DIY Midjourney is a great idea — until it is not. This is the honest framework: when the cost of self-managing your AI imagery workflow exceeds the cost of hiring help, and what to look for in a specialist.