MidjourneyWorkflow & organisation
Midjourney Marketing Campaign Workflow: Brief to Final Set
To make a consistent Midjourney campaign, write a short brief, lock a style block of fixed wording, --sref and parameters, vary only the subject and framing, generate each channel at its own --ar, save and name every prompt, then review the whole set side by side against a checklist.
Start with a one-page brief
Consistency in a campaign is decided before the first prompt. If you start generating straight away, every good image pulls the look in a new direction and you end up with twenty nice pictures that do not belong together. Write a short brief first, even if it is only for yourself.
- Message: the one thing the campaign says.
- Audience: who should stop scrolling, and what they care about.
- Visual direction: three to five words for mood, a colour palette named in plain words, light quality and a photographic or illustrative treatment.
- Subjects: the recurring people, products or places.
- Deliverables: every channel and format you need, with counts.
- Constraints: what must never appear, brand rules, and where real product shots or text will be added later.
Remember that Midjourney is weak at long exact text, so headlines, logos and legal lines belong in your design tool. Plan clean space in the images for them.
Lock the style before you make anything else
Next, turn the visual direction into a fixed "style block": the part of every prompt that never changes. It has three layers.
- A written treatment. For example:
soft overcast daylight, muted sage and cream palette, gentle film grain, calm documentary photography. Copy it word for word into every prompt. - A style reference. Add
--srefwith an image URL or a numeric style code that captures the grade you want. Adjust how strongly it applies with--sw, the style weight. If you have no reference yet,--sref randomgives you a random style code to explore, and you can reuse the code you like. The style references guide covers this in depth. - Fixed parameters. Decide on
--raw, your--svalue and the model version once, and keep them. Note that style codes can render differently across model versions, so do not switch versions halfway through a campaign.
A --seed can help when you want to compare small wording changes, because the same prompt, seed and settings give very similar results. It does not lock a style once the subject changes, so rely on the style block and --sref for consistency. See seeds for what they can and cannot do.
Test the style block on two or three different subjects before going further. If it only works for one, it is not locked yet.
Build variations without breaking the look
Think of each prompt as fixed slots and variable slots. The style block and parameters are fixed. Only the subject, action, setting detail and framing change.
a woman in a linen apron arranging tulips at a market stall, medium shot, soft overcast daylight, muted sage and cream palette, gentle film grain, calm documentary photography --ar 4:5 --raw --sref [your code] --sw 200
close detail of hands wrapping tulips in brown paper, soft overcast daylight, muted sage and cream palette, gentle film grain, calm documentary photography --ar 4:5 --raw --sref [your code] --sw 200
Plan a shot list as a real shoot would: an establishing wide, a hero image, people and product in use, close details and a few simple backgrounds for text-heavy formats. When an image is nearly right, use Vary (Subtle) to refine it, or Vary (Strong) to explore alternatives with the same overall idea. Keep --chaos low for campaign work; high values are for exploring, not for matching a set.
Aspect ratios per channel
Generate each format at its own ratio with --ar rather than cropping one master image. Midjourney recomposes the scene for the frame, so a 9:16 version is not simply a cut-down 16:9. Platform specifications change, so confirm current sizes before you deliver.
| Channel or format | Typical ratio | Parameter |
|---|---|---|
| Square feed posts, marketplace tiles | 1:1 | --ar 1:1 |
| Portrait social feed posts | 4:5 | --ar 4:5 |
| Stories, Reels, short vertical video covers | 9:16 | --ar 9:16 |
| Website hero, video thumbnails, presentations | 16:9 | --ar 16:9 |
| Wide banners and email headers | 3:1 or wider | --ar 3:1 |
| Print posters | 2:3 or 3:4 | --ar 2:3 |
For more on how ratios affect composition, see aspect ratios.
Organise and name your prompts
A campaign can easily run to hundreds of prompts. Without a system, you will not be able to find the exact wording behind the approved hero image three weeks later. Store the prompt text, not just the image.
- One folder per campaign, with sub-groups per channel or shot type.
- A naming pattern such as
spring24_hero_market-stall_v03: campaign, role, subject, version. - A master style block saved separately, so every new prompt starts from the approved version.
- Notes on what changed between versions, especially parameter values.
- Status labels such as exploring, shortlisted, approved and rejected.
The guide to organising prompts goes further on folder structures.
Review against a checklist
Review images side by side, not one at a time. Lay out the whole set for each channel and look for the odd one out.
| Check | What to look for | Fix |
|---|---|---|
| Style match | Same palette, grade, grain and light across the set | Reapply the style block and --sref; adjust --sw |
| Subject accuracy | Products, people and props look right | Composite real product shots; Vary (Subtle) |
| Artefacts | Hands, faces, garbled text, merged objects | Vary, reroll or retouch |
| Text space | Clean area for headline and logo | Describe empty space in the prompt |
| Formats | Every deliverable generated at its ratio | Regenerate with the right --ar |
| Resolution | Big enough for the largest placement | Upscale the approved image |
| Rights and claims | No real brands, people or misleading product claims | Rewrite prompt; check with legal |
| Records | Final prompt and parameters saved and named | Save before handover |
Doing this faster with Dreamdrive
Dreamdrive keeps the full prompt with every capture, --sref code included, so a look you liked is a search away in your Library. Dreamdrive works inside midjourney.com in desktop Chrome.