Sign in Get started
Blog

Changing Midjourney aspect ratios can still break your freelance brand pack

By the Havincy team

Share
Changing Midjourney aspect ratios can still break your freelance brand pack
Midjourney style weight --sw runs from 0 to 1000 with a default of 100, and the retainer folder already holding story, feed, and landscape stills for the same client still shows palette, grain, and lighting drifting crop-to-crop because Style Reference was never pinned when each --ar hop re-rolled the look.

Three formats delivered does not mean one brand pack.


Style weight runs from 0 to 1000, and most packs never write the number down


Midjourney’s Style Reference docs set a concrete control freelancers treat as optional: --sw ranges from 0 to 1000, default 100, and it decides how hard the Style Reference pulls colors, medium, textures, and lighting into the new image (https://docs.midjourney.com/hc/en-us/articles/32180011136653-Style-Reference). Style Reference copies vibe, not people or objects. On Discord you attach it with --sref plus image URL(s); on the web Imagine bar you drop the reference into the Style Reference slot and can pin or lock it for reuse across prompts. Compatible with Midjourney v6 and later.


Discord can weight multiple refs, for example --sref URL1::2 URL2::1. On V7 image srefs, --sv picks the sref model version (default --sv 6; --sv 4 is the older V7 sref model). Most freelancers change --ar, keep the text prompt, and assume default 100 still holds. The folder fills. The weight was never written down, so the next crop invents a different strength of the same look.

Three crops delivered; the client still sees three freelancers

Monday night the CapCut timeline already has a 9:16 story, a 1:1 feed still, and a 16:9 landscape hero for the same DTC retainer. Figma holds the type overlays. The brief is closed on product, headline, and crop list.


Side by side, the client does not see one vendor. The story frame runs cooler and flatter; the square pops warmer with heavier grain; the landscape softens the lighting and thickens the shadows. Same product words. Three visual authorships. CapCut and Figma can assemble the crops. Neither is the style system of record. Assembly sits after the look is locked.

Aspect hops without a pinned Style Reference

An aspect hop is not a harmless export. Changing --ar without a locked Style Reference asks Midjourney to invent adherence from text alone. Prompt adherence then follows adjectives instead of a pinned vibe: soft film grain, warm key, muted teal rewrite themselves every crop because nothing visual is pinned.


On Discord, lock --sref with the approved image URL(s) or Style Creator code before you change --ar. On the web, pin the Style Reference slot so the next prompt reuses it. Midjourney’s docs state that Style Reference applies colors, medium, textures, and lighting without copying objects or people (https://docs.midjourney.com/hc/en-us/articles/32180011136653-Style-Reference). That is the brand-pack contract: same vibe across 9:16, 1:1, and 16:9. If the reference is missing, each aspect hop is a new freelancer.

Prompt adjectives fighting the reference, and the Moodboard trap

Two failure modes show up in the same week.


First, the text prompt fights the Style Reference. You pin a cool, low-contrast pack, then type golden hour, high contrast, glossy product sheen. Prompt adherence splits. When Style Reference is pinned, strip style adjectives and leave content: product, pose, crop intent, negative space. Let --sref carry the vibe.

Second, Moodboards and --sw do not mix. The same Style Reference article states that --sw is not compatible with Moodboards. Another moodboard dump will not hold the pack when weight notes stop applying. If you need a reusable code instead of an image URL, that is Style Creator’s job, not another board.

Style lock before the batch: code, weight, content-only prompts, rejected drift

Write the lock before the three-crop batch opens.


Style source: one approved still URL on --sref, or a Style Creator code you will reuse. Midjourney’s Style Creator builds reusable custom --sref codes by selecting sample styles; preview rounds use GPU time; most styles stabilize after five to ten rounds; stacked codes from reopening a session do not merge — use both codes together to match what you saw (https://docs.midjourney.com/hc/en-us/articles/41308374558221-Style-Creator).

Weight note: the --sw you will ship with. Default is 100. Raise it when the vibe is too weak across crops; lower it when the reference is crushing content. Write the number in the pack.

Content-only prompts per crop: product, angle, crop intent, negative space. No competing style adjectives once --sref is pinned.

Rejected drift: two or three killed stills with reasons (palette warmed, grain thickened, lighting softened). That pile teaches the coverage pass what off-pack looks like before CapCut opens.

Coverage pass across 9:16, 1:1, and 16:9

One pass before assembly. Continuity of vibe, not a prettier single frame.


Check palette, medium, lighting, and texture across the three crops with the same --sref and --sw. Style Reference should copy those four, not the product object itself. If the 9:16 story is cooler than the 1:1 feed, re-run the weak crop with the lock intact — do not hide drift in Figma with a global grade. CapCut pastes approved stills. It does not invent the brand pack.

Where characters, image edit, and multi-format sit in the loop

Havincy’s production loop is Idea, Create, Produce, Publish, Measure, Improve. Persistent characters and universes live in Create when the retainer needs the same people or worlds across stills and later video or talking avatars. Image generation and image edit lock hero frames and style continuity before any multi-format hop. Director sits in Produce once the frames are approved. Social distribution and analytics close the loop after the crops ship. CapCut and Figma remain assembly. The system of record for Midjourney style reference and style weight stays upstream of the aspect hop.


FAQ — Is a random style reference safe for retainers?

No. A one-off --sref random or an unpinned Imagine slot is exploration, not a retainer contract. Retainers need a named still URL or Style Creator code, a written --sw, and rejected drift examples so the next batch fails the same way on purpose instead of inventing a fourth freelancer. Random is fine for pitch R&D. It is not fine once the client has signed the look.

Style weight runs from 0 to 1000, and most packs never write the number down

FAQ — When to raise style weight versus rewrite the text prompt?

Raise --sw when the pinned Style Reference is correct but too weak across crops. Rewrite the text prompt when content is wrong: product angle, crop, or props. Do not raise weight to paper over style adjectives that fight the reference; strip those adjectives first. Documented range: 0 to 1000, default 100, and --sw does not apply on Moodboards (https://docs.midjourney.com/hc/en-us/articles/32180011136653-Style-Reference).


FAQ — Style Creator rounds versus another moodboard dump?

Spend the GPU rounds. Style Creator preview rounds cost GPU time, and most styles stabilize after five to ten rounds; if you reopen a session and get a second code, the codes do not merge — use both together to recreate what you approved (https://docs.midjourney.com/hc/en-us/articles/41308374558221-Style-Creator). Another moodboard dump does not give a reusable --sref code, does not document --sw, and does not survive the next aspect hop. For a freelance brand pack across 9:16, 1:1, and 16:9, a stabilized Style Creator code plus a written weight beats a larger board.

← All articles