--- id: wiki-2026-0508-style-reference-sref title: Style Reference (--sref) category: 10_Wiki/Topics status: verified canonical_id: self aliases: [Style Reference, Midjourney sref, --sref] duplicate_of: none source_trust_level: A confidence_score: 0.9 verification_status: applied tags: [midjourney, style-reference, prompt-engineering, image-generation] raw_sources: [] last_reinforced: 2026-05-10 github_commit: pending tech_stack: language: prompt framework: midjourney --- # Style Reference (--sref) ## 매 한 줄 > **"매 Midjourney 의 reference-image style transfer flag"**. V6 (2024-01) 부터 도입 — 매 image URL 또는 sref code (numeric ID) 를 prompt 에 attach → 매 referenced image 의 visual style 을 generation 에 적용. 매 --cref (character) 와 더불어 controllable creative 의 핵심 tool. ## 매 핵심 ### 매 syntax ``` /imagine prompt: --sref [--sw <0-1000>] ``` - **`--sref `**: 매 reference image URL. - **`--sref `**: 매 numeric "style code" (e.g., `1234567`) — Midjourney 의 internal style ID. - **`--sref random`**: 매 random style code (V6.1+) — 매 creative exploration. - **`--sw <0-1000>`**: style weight (default 100). higher = stronger style adherence. ### 매 multiple references - `--sref url1 url2 url3` — 매 multiple style 의 blend. - 매 weight per ref: `--sref url1::2 url2::1` (relative). ### 매 --sref vs --cref vs --iw - **--sref**: 매 style only (color, texture, mood). - **--cref**: 매 character / face identity preservation. - **--iw <0-3>**: 매 image weight 의 prompt vs reference balance. ### 매 응용 1. Brand asset variation (매 일관된 visual style). 2. Mood board → final art (매 creative pipeline). 3. Style code library (매 reproducible aesthetic). 4. Series consistency (매 book illustration, comic). ## 💻 패턴 ### Basic usage ``` /imagine prompt: a serene mountain lake at dawn --sref https://s.mj.run/abc123 --sw 100 ``` ### Style code (numeric) ``` /imagine prompt: cyberpunk samurai --sref 2589374610 --sw 200 --v 6 ``` ### Multiple references blended ``` /imagine prompt: a cozy library --sref https://img1.jpg https://img2.jpg --sw 150 ``` ### Random style exploration ``` /imagine prompt: portrait of a wizard --sref random --v 6.1 ``` ### sref code 추출 (after generation) ``` # Reply to a generated image with /describe — 매 sref code 를 capture /describe # → outputs prompt + potential sref signature ``` ### Weighted multi-ref ``` /imagine prompt: futuristic city --sref url1::3 url2::1 --sw 200 # url1 의 style 의 3x weight relative to url2 ``` ### Series consistency (book illustration) ``` # Page 1 /imagine prompt: a fox in a snowy forest, page 1 --sref 7891234560 --sw 150 --ar 3:4 # Page 2 (same sref) /imagine prompt: the fox finds a glowing stone, page 2 --sref 7891234560 --sw 150 --ar 3:4 # Page 3 /imagine prompt: the fox returns home, page 3 --sref 7891234560 --sw 150 --ar 3:4 ``` ### --sref + --cref combo ``` /imagine prompt: knight in armor --cref https://character.jpg --cw 100 \ --sref https://style.jpg --sw 150 # character identity preserved + style applied ``` ## 매 결정 기준 | 상황 | Approach | |---|---| | 매 brand consistency | sref code (saved) + --sw 150-300 | | 매 character series | --cref + --sref combo | | 매 exploration | --sref random | | 매 strong style enforcement | --sw 500-1000 (warning: prompt 이 약화) | | 매 subtle hint | --sw 50-100 | **기본값**: 매 --sw 100-200 + 매 saved sref code. ## 🔗 Graph - 부모: [[Midjourney]] · [[Prompt_Engineering|Prompt-Engineering]] - 응용: [[Style-Transfer]] - Adjacent: [[IP-Adapter]] · [[ControlNet]] ## 🤖 LLM 활용 **언제**: 매 creative brief → Midjourney prompt 의 LLM 작성, 매 sref code library 의 tagging / search, 매 series consistency planning. **언제 X**: 매 open-source self-hosted (use IP-Adapter / FLUX), 매 strict face identity (use --cref or PuLID), 매 photo retouching. ## ❌ 안티패턴 - **--sw 너무 높음 (>500)**: 매 prompt 이 무시됨 — 매 reference image 의 copy 에 가까워짐. - **Multiple sref 너무 많이**: 매 5+ refs 는 muddy. 매 2-3 의 sweet spot. - **sref + style emoji 충돌**: "--sref X cinematic dark" — 매 둘 다 style → conflict. 매 하나 선택. ## 🧪 검증 / 중복 - Verified (Midjourney official docs, V6 release notes 2024-01, V6.1 sref random 2024-07). - 신뢰도 A. ## 🕓 Changelog | 날짜 | 변경 | |---|---| | 2026-05-08 | Phase 1 | | 2026-05-10 | Manual cleanup — V6 syntax + --sw + multi-ref + cref combo |