Midjourney Sref Explained: --sref Style Reference Guide

Midjourney sref is the style reference flag that copies look from an image URL or a numeric style code onto a new scene. You write it as --sref after your text prompt, then tune strength with --sw (0–1000, default 100). On Midjourney v7 you can also pick how the reference is read with --sv. You leave with paste-ready --sref syntax, a practical --sw map, how style reference codes and public libraries fit a team workflow, and a clean split from --stylize and --cref. Work a few grids with the same reference before you change subjects. Soft path when you need Midjourney-shaped scene text from a photo first: https://promptmake.net/image

What midjourney sref is

Midjourney sref (style reference) pulls palette, texture, lighting grade, and medium feel from a reference you attach. The model keeps your text as the subject and place brief. The reference steers how the frame should look: film grain vs clean catalog light, gouache vs chrome product finish, muted editorial vs high-contrast poster color. You attach either a public image URL after --sref or a numeric style code that Midjourney maps to a saved look.

You reach for midjourney sref when one approved hero must set the visual language for a whole series: campaign variants, book pages, product set shots that share grade, or concept boards that must match a locked mood. Text adjectives alone drift across grids. A style reference gives the model a concrete look target without forcing the same subject into every tile.

Who this guide is for: Discord and web Imagine users on Midjourney v7 (and V8 Alpha where you have access) who paste random community codes and get chaos; teams that need a shared style reference library with weights and notes; people who confuse --sref with --stylize or --cref. Our stylize article covers the house-taste dial. This page stays on style reference: URLs, codes, --sw, --sv, and a repeatable lock workflow.

How Midjourney `--sref` works

Treat midjourney sref as a look lock on top of a full text prompt. You still name subject, place, light, and medium in words. --sref tells Midjourney which visual family those words should inhabit. Official docs require text with the reference; a bare image URL without scene language is the wrong tool for this job. Image prompts (composition references) sit before --sref when you use both. File URLs should end in common image types such as .png, .jpg, .jpeg, .webp, or .gif. Discord and the web prompt bar both accept --sref. Wrong spacing or a dead URL wastes a credit the same as a bad subject line, so lock syntax before you chase taste. The subsections below cover URL and code forms, style weight, and style-reference versions on v7.

Syntax: image URLs, codes, and `--sref random`

Put --sref at the end of the prompt with other parameters. Space before the double dashes. After the flag, pass one or more references separated by spaces:

fox librarian in a treehouse library, warm lantern light, children's book scene --ar 3:2 --sref https://example.com/hero-style.png --v 7

Numeric style codes use the same flag: --sref 1234567890 (replace with a real code from your library). Multiple references blend: --sref URL1 URL2 or --sref CODE1 CODE2. On Discord you can weight individual references the way multi-prompts work: --sref URL1::2 URL2::1. --sref random asks Midjourney for a random style; useful for exploration, weak for brand lock. You can combine image URLs with codes in one --sref list when you blend a hero file with a known code look.

Style weight `--sw` (0–1000)

--sw sets how hard the style reference pulls. Range is 0–1000. Default is 100. Low values let your text and Midjourney's default habits win more often. High values push the reference look harder: stronger palette match, tighter texture and grade. As of mid-2026 on Midjourney v7, Midjourney notes that style weight often hits harder on numeric style codes than on image URLs, so the same number is not a 1:1 feel across both input types. Start at 100 on a new hero. Step to 50–80 when the reference smothers subject clarity. Step to 150–250 when the series still drifts from the approved look. Save explore jumps (400–1000) for mood boards, not for SKU-faithful catalog sets.

Style reference versions with `--sv`

Midjourney v7 exposes several style-reference algorithms through --sv. Docs and release notes around mid-2025 through mid-2026 treat --sv 6 as a common default for current image-based sref behavior, with --sv 4 as the fallback when you need the older V7 style-reference model (useful for legacy codes from before the June 2025 sref refresh). Exact defaults can move with account and build; check the suffix Midjourney shows after a run. --sref random and many numeric codes are documented as compatible with specific --sv values (often 4 and 6). If an old community code looks wrong on a fresh v7 job, re-test with --sv 4 before you discard the code. When you move onto V8 Alpha or a later V8 default, re-check --sv once the same way you re-check other parameters.

Midjourney style reference codes and libraries

A style reference code is a numeric ID Midjourney treats as a reusable look. Creators share codes in Discord, blogs, and public style reference libraries so teams can paste a number instead of hosting a hero PNG. Codes travel light in a spreadsheet. They also age: Midjourney rebuilt parts of the sref system between v6 and v7, and again inside v7, so a viral code from 2024 may need --sv 4 or may map to a different feel on current defaults. Treat every imported code as a hypothesis until you run one control grid on your account. Build your own library from approved campaign heroes when brand risk is high; use public libraries when you are exploring or matching a known community aesthetic. The subsections below cover how to store codes and how to mix them with URLs without melting the brief.

Build a small team library (not a thousand random codes)

Store each midjourney sref entry with four fields: code or URL, --sw ship value, --sv if required, and a one-line job note ("muted film product," "gouache kids book," "chrome hard-surface"). Keep a PNG of the hero next to the row so newcomers see the look, not only the number. Cap the active set per project at a handful of entries. A bloated style reference library invites people to stack five codes on one prompt and call the mess "brand." Review the library when you change Midjourney version or when a campaign closes; archive codes that no longer match.

Mix codes, URLs, and relative weights

Blending two soft codes can work when both share the same medium family. Blending a photoreal product URL with a painterly illustration code often fights the subject. Prefer one primary midjourney sref plus optional light secondary weight: --sref PRIMARY::2 SECONDARY::1 --sw 120. If the grid still splits into two looks across tiles, drop the secondary and raise --sw on the primary instead. When a public library code is close but not right, use it as exploration, then lock a new hero from your own winning tile and switch the project to that URL. Owned heroes beat anonymous codes for client work.

Step-by-step: lock a look with midjourney sref

Change one variable per grid. If you rewrite the scene, add --cref, jump --sw, and swap three style codes in the same run, you learn nothing. Lock subject text and aspect ratio first. Pick a single style reference second. Then move --sw in measured steps. PromptMake /image can draft Midjourney-formatted scene text from a photo when you need a clean starting line; you still attach --sref and --sw by hand for the look lock. Soft CTA: https://promptmake.net/image (about 3 image runs per day as a guest, about 5 with a free account). The steps below assume Midjourney v7 syntax you can paste into Discord or the web app.

1. Approve one hero and get a stable URL or code

Generate or photograph the look you want to repeat. Upscale or save the hero so the URL stays reachable. Prefer a frame that shows medium and grade with a simple subject; busy scenes leak props into later jobs even when v7 sref reduced subject bleed. If you use a numeric code from a library, paste it into a notes doc with the source link and date. Run one throwaway prompt that only tests the reference on a neutral subject ("ceramic cup on a table") so you see the look before you burn campaign credits.

2. Baseline at `--sw 100` with fixed scene text

Write the production scene once. Add --sref with your hero URL or code, --sw 100, and your usual --ar and --v 7. Generate a four-up. Write the failure in one line: too weak (look drifts), too strong (subject melts into texture), wrong medium, color cast off brand. If the failure is subject or composition, fix the prose before you touch --sw. Style reference will not invent a missing product feature or fix a vague noun stack.

3. Tune `--sw`, then save the ship recipe

From a weak match, step --sw to 150 or 200, keep text and --sref identical. From an overpowering match, step down to 50–80. Stop when the series holds grade without eating the brief. Save the full suffix next to the hero PNG: --sref … --sw … plus any --sv you needed. Reuse that recipe on every production prompt in the set. Use --sref random or high --sw only for mood exploration outside the ship lane. When you switch to V8 Alpha or a later V8 build, re-test the ship --sw once; do not assume the same number feels identical.

Common midjourney sref mistakes

Stacking five style reference codes on one line is the classic failure. The model averages conflicting looks and your subject becomes a texture experiment. Pick one primary midjourney sref. Add a second only when you have a reason and a weight plan.

Other traps that show up in Discord every week:

• Treating --sref as a character lock (that job belongs to --cref / omni-style character tools on current builds)
• Expecting midjourney sref to replace a clear subject sentence
• Copying a legacy style code without testing --sv 4 when the look no longer matches the screenshot in the library post
• Raising --sw to 800 on catalog work and wondering why labels and materials dissolve
• Changing scene text, chaos, and --sw in the same regenerate
• Using --sref and a heavy house polish dial at once without a plan (see the stylize split below)
• Hosting the hero on an expiring link so tomorrow's batch fails

Fix pattern: one hero → baseline at --sw 100 → name the visible defect → move weight one band → save the ship suffix. Reach for character reference only when identity of a person or mascot must hold across scenes.

Midjourney sref vs `--stylize` vs `--cref`

These three controls solve different problems. Midjourney sref pulls look from a reference URL or style code. --stylize (or --s) scales Midjourney's own aesthetic taste on the current text; it does not attach a file. --cref (and related character / omni reference tools on your build) aims at who appears in the frame, not the grade of the whole image. Teams burn grids when they raise stylize to "match the brand PDF" or paste a face URL into --sref and hope for character consistency.

Practical pairings as of mid-2026 on Midjourney v7:

• Locked campaign look, new subjects: stable scene grammar + midjourney sref on the approved hero + moderate --sw (often 80–200).
• More or less Midjourney polish without a file: adjust --stylize / --s only. Full value maps and photo vs illustration bands live in the Midjourney stylize article on this blog (midjourney-stylize-explained).
• Same character, new places: character reference path for identity; optional separate --sref if the world look must match a style hero too.

Keep stylize out of deep focus here on purpose. If the brief is "more Midjourney beauty" or "more literal," open the stylize guide. If the brief is "match this file or this style code," stay on this page and tune --sref / --sw.

Draft scene text with PromptMake, then add `--sref`

Style reference works best when the text already names subject, materials, and light with enough clarity that the model has something solid to dress. Thin prompts plus a strong midjourney sref produce pretty emptiness. Dense, contradictory prompts plus a strong sref produce muddy blends. A clean Midjourney-formatted draft from a photo gives you a better base than a one-word subject and a hope.

PromptMake /image turns a photo into a Midjourney-ready scene you can paste, then you append your ship --sref and --sw from the library row. Goal modes help when the photo is a recreate, a style change, a lighting adjust, or a variation brief. Soft CTA: https://promptmake.net/image. Guests get about 3 image generations per day; registered free accounts get about 5. PromptMake formats for Midjourney, FLUX, DALL·E, Stable Diffusion, and Leonardo; for this workflow pick Midjourney, copy the draft, and add style reference by hand so the look lock stays under your control.

FAQ

What is midjourney sref?

Midjourney sref is Midjourney's style reference parameter. You add --sref plus an image URL or a numeric style code so new generations inherit palette, texture, and grade from that reference while your text sets the subject and scene. It is the look-lock tool for series work. Pair it with --sw when you need stronger or softer pull from the reference.

How do I write the `--sref` style reference syntax?

End the prompt with --sref followed by a space and your URL or code, then other flags such as --ar and --v 7. Example shape: scene text, then --sref https://example.com/style.png --sw 100 --v 7. Multiple references go in one list after the flag. On Discord you can weight them with :: relative weights. Keep a live URL; broken links fail the reference.

What does `--sw` do with midjourney sref?

--sw is style weight. It sets how hard the style reference pulls on the image, from 0 to 1000, with default 100. Lower values favor your text and default habits. Higher values push the reference look harder. On v7, weight often feels stronger on numeric style codes than on image URLs, so re-test when you switch input type.

What are midjourney style reference codes and libraries?

Style reference codes are numeric IDs you pass to --sref instead of (or beside) an image URL. Public style reference libraries and Discord shares collect codes for reuse. Codes can shift when Midjourney updates the sref system, so run a control grid and note any required --sv value. For brand work, prefer a library of your own approved heroes with saved --sw values.

How is midjourney sref different from `--stylize`?

Midjourney sref attaches an external look from a URL or code. --stylize scales Midjourney's built-in aesthetic taste on the current prompt text without a reference file. Use sref to match a hero or code. Use stylize to raise or lower house polish. Deep stylize bands and examples live in the separate Midjourney stylize article (midjourney-stylize-explained) on this blog.

How is midjourney sref different from `--cref`?

--sref targets overall style: color, medium, texture, photographic grade. --cref (and related character reference tools on your Midjourney build) targets who or what character appears. Use sref when the world must match a look. Use character reference when a person or mascot must stay recognizable across scenes. You can combine both when a series needs a locked face and a locked grade.

Where can I draft Midjourney prompts before I add `--sref`?

PromptMake /image turns a photo into a Midjourney-formatted draft you can paste into Discord or the web app. Guests get about 3 image generations per day; registered free accounts get about 5. Add your midjourney sref URL or style code and --sw from this guide after you copy. Start at https://promptmake.net/image