Files
2nd/10_Wiki/Topics/AI_and_ML/Style_Reference_(--sref).md
T
Antigravity Agent f8b21af4be Wiki cleanup: error-doc removal, dedup merge, link normalization
10_Wiki/Topics 대규모 정리:
- 오류 캡처/미완성 stub 문서 227개 제거
- 교차폴더 중복 43클러스터 병합 (63파일 → redirect)
- 링크명 정규화: 깨진 링크 수정·redirect 직결·개념 매핑 ~2,400건
- 카테고리 MOC 6개 신규 생성
- Graph 섹션 미해결 related-keyword 링크 10,058건 제거

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-20 23:52:15 +09:00

4.5 KiB

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack
id title category status canonical_id aliases duplicate_of source_trust_level confidence_score verification_status tags raw_sources last_reinforced github_commit tech_stack
wiki-2026-0508-style-reference-sref Style Reference (--sref) 10_Wiki/Topics verified self
Style Reference
Midjourney sref
--sref
none A 0.9 applied
midjourney
style-reference
prompt-engineering
image-generation
2026-05-10 pending
language framework
prompt 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: <description> --sref <url-or-code> [--sw <0-1000>]
  • --sref <URL>: 매 reference image URL.
  • --sref <code>: 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 <upload generated image>
# → 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

🤖 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