Loading tutorials…
Loading tutorials…
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.
Who this is forIn-house designers and creative ops at brands shipping 50+ AI-generated assets per month. If brand reviewers keep rejecting Midjourney work as "off-brand," the issue is system, not skill.
What you'll need
Step 1
Translate "our brand is warm and human" into concrete visual rules: color palette, lighting type, composition style, subject types.
Open your brand guidelines doc. Find the photography or imagery section.
Extract: color palette (with hex codes), lighting style (soft, hard, golden hour, studio), composition rules (rule of thirds, center, negative space), subject demographics, props/settings allowed.
If your brand guidelines do not specify imagery, do this work first with your brand stakeholder. Without explicit rules, you are guessing.
Translate each rule into Midjourney prompt language. "Warm and human" → "warm tones, golden hour light, candid editorial photography, mid-30s subjects in casual settings."
Step 2
Most brands need 3-5 visual modes: hero, product, lifestyle, illustrated, social. Each mode gets its own dedicated SREF.
Mode 1 — Hero/editorial: long-format banners, top-of-funnel content. Mood: aspirational, premium.
Mode 2 — Product clean: e-commerce, feature highlights. Mood: focused, clean, neutral.
Mode 3 — Lifestyle: customer-in-context, ads, testimonials. Mood: warm, candid, real.
Mode 4 — Illustrated/diagrammatic: educational content, blog headers. Mood: clear, simple, on-brand colors.
Mode 5 — Social square/portrait: mid-feed posts, stories. Mood: hooky, scroll-stopping.
For each mode, find or generate 3 reference images. Save the SREF codes. Document --sw and --stylize defaults per mode.
This is the foundation. Without it, every prompt re-invents the brand from scratch.
Step 3
Each mode gets a fill-in-the-blank template with locked brand elements and 2-3 swappable variables.
Hero template: "--sref [hero-mode-code] --sw 200 [SUBJECT] in [SETTING], [TIME OF DAY], cinematic editorial photography, premium feel --ar 16:9 --v 7 --stylize 200"
Product template: "--sref [product-mode-code] --sw 250 [PRODUCT] on neutral seamless background, soft studio lighting, sharp focus, commercial photography --ar 1:1 --v 7 --stylize 50"
Lifestyle template: "--sref [lifestyle-mode-code] --sw 150 [PERSONA] [DOING ACTION] in [SETTING], natural light, candid editorial, shot on Kodak Portra --ar 4:5 --v 7 --stylize 100"
Locked: --sref, --sw, --stylize, --ar, base style fragment.
Swappable: subject, setting, action, time of day.
Store templates in a shared Notion/Airtable so any team member uses the same starting point.
Step 4
Every Midjourney output goes through a 5-7 point QA before delivery. Without this, off-brand output ships.
QA point 1: Color palette matches brand. No accidental wild colors.
QA point 2: Subject demographics match brand audience definition.
QA point 3: Mood matches the mode (premium for hero, warm for lifestyle, etc.).
QA point 4: No visible AI artifacts — bad hands, weird text, floating limbs, distorted faces.
QA point 5: Composition follows brand rules (negative space, subject placement).
QA point 6: No accidental competitor or brand references in background.
QA point 7: Image is at delivery resolution (Upscaled, not raw 1024x).
Print this checklist. Tape it next to your monitor. Reject anything that fails 2+ points.
Step 5
Before signing off on a system, generate 50 images across all modes. Lay them out. Brand stakeholder reviews as one body of work.
Use your templates to generate ~10 images per mode (5 modes × 10 = 50 total).
Use varied subjects and settings to stress-test consistency.
Create a single Figma/Canva canvas with all 50 images laid out by mode.
Brand stakeholder reviews. Goal: every image feels like one brand.
If 5+ images feel off-brand, the templates need iteration. Find the common failure mode (usually a specific SREF or --sw value) and adjust.
Sign-off only after 90%+ of generated output passes brand review without edits.
Step 6
A system that lives in one person's head is not a system. Document everything; train at least one backup operator.
Build a "Brand AI Imagery Playbook" doc. Include: brand visual rules, mode definitions, SREF codes, prompt templates, --sw / --stylize defaults, QA checklist, example outputs.
Run a 1-hour training with anyone else who will generate brand imagery (designer, social manager, marketing assistant).
Set up a shared inbox or Notion page for new images pending QA — central choke point ensures nothing ships unreviewed.
Update the playbook quarterly. As brand evolves, modes evolve too.
Step 7
Even with systems, output drifts over time. Quarterly audit catches it early.
Each quarter, pull the last 30 days of Midjourney output.
Compare to the baseline 50-image consistency test from setup.
Common drift modes: stylize getting lowered over time for "more realism," SREF stopping being applied to "save time," new operators introducing personal taste.
Reset the system: re-share playbook, re-run a small consistency test, fix any drifted templates.
This 1-hour quarterly audit prevents the slow degradation that kills most AI imagery systems within 6 months.
Common mistakes
Skipping the brand-rule extraction step
What goes wrong: Without explicit brand rules, every generation is a guess. Brand stakeholder rejects 60% of output. Trust in AI imagery erodes.
How to avoid: Extract explicit visual rules from brand guidelines first. If guidelines do not have them, build them with the brand stakeholder before any prompt work.
One SREF for the whole brand
What goes wrong: You force product shots, hero banners, and social content through one style code. Each one gets close but never quite right. Output is consistent in mood but wrong for context.
How to avoid: Build a multi-mode SREF library. 3-5 modes is the right floor for most brands. Each mode = its own SREF.
No QA pass
What goes wrong: Output ships with AI artifacts (bad hands, weird text, accidental competitor logos in backgrounds). Brand looks unprofessional. Stakeholder rejects after launch.
How to avoid: Mandatory 5-7 point QA checklist on every image. No exceptions.
No documented playbook
What goes wrong: The system lives in one person's head. They leave, system collapses. New operator generates rogue brand-inconsistent output for 3 months before anyone notices.
How to avoid: Document the playbook. Train at least one backup operator. Update quarterly.
Generating without Stealth Mode for brand work
What goes wrong: Brand campaign visuals leak in the public Midjourney feed pre-launch. Competitors see your direction. Brand stakeholder calls it a confidentiality breach.
How to avoid: Pro Plan + Stealth Mode for ALL brand work. Non-negotiable. Anything client-visible defaults to private generation.
Not auditing for drift
What goes wrong: After 6 months, your "brand consistency system" is a doc nobody reads and prompts that have drifted from the playbook. Output looks generic again.
How to avoid: Quarterly audit. 1 hour. Pull last 30 days, compare to baseline, fix drift early.
Recap
Done — what's next
How to use Midjourney style references and SREF codes
Read the next tutorial
Hand it off
Building and maintaining a brand-consistent AI imagery system is a creative ops job. Most brands underestimate the setup time and skip the maintenance. EverestX matches you with a graphic designer or creative strategist who builds the system, trains your team, and maintains it ongoing. Setup typically $1,000-2,500; ongoing maintenance $400-1,000/mo.
See specialist rates
Most B2B SaaS brands need 3 modes (hero, lifestyle, illustrated). E-commerce brands typically need 4-5 (hero, product, lifestyle, social, sometimes illustrated). Service businesses often only need 2-3. Start with what you actually publish; add modes only when a real use case demands them.
Midjourney does not offer custom model training the way Stable Diffusion does. The closest equivalent is using --sref with consistent references and --p (personalization) trained on your own liked images. For true custom-trained brand models, Stable Diffusion + DreamBooth or LoRA is the alternative — but adds significant complexity.
For most brands: 6-8 hours for initial setup if you have clear brand guidelines. 16-20 hours if you have to build brand visual rules from scratch first. Maintenance: ~1 hour per quarter for audits, plus ongoing per-batch QA.
Three usual causes: (1) stylize too high, making images feel artistic instead of real. (2) Using stock-photo SREF references that have a generic vibe. (3) No post-processing — output ships raw from Midjourney instead of through your real photo grading workflow. Address all three before next review.
No. Real people deserve real photos. AI-generated "team headshots" feel uncanny and dishonest, especially when discovered. Use Midjourney for: scenes, settings, backgrounds, illustrated content, abstract concepts. Use real photography for: real people in your brand.
Midjourney
--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.
Canva
Most teams skip Brand Kit and then wonder why their creative looks 40% less consistent than competitors. This walks through the full Brand Kit setup — colors, fonts, logos, voice, templates — and the per-tier limits that trip people up.
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.