twodogegg/shuliu-skills

multi-image-consistency

generate coordinated prompts for tasks that require multiple images with strong cross-image consistency.

소스 보기

사용 방법

  1. 설치 명령을 복사하고 자신의 에이전트 환경에 설치하기 전에 소스 패키지를 확인하세요.
  2. 필요한 도구, 인증 정보, 지원 환경은 원본 문서에서 확인하세요.
  3. 입력과 원하는 산출물을 설명하고 출처와 결과를 대조하세요.
npx skills add https://github.com/twodogegg/shuliu-skills --skill multi-image-consistency

시작 전 확인

  • 설치에는 모델 구독, API 한도, 파일 접근 권한이 포함되지 않습니다.
  • 별점 수와 설치 수는 활동 지표이며 독립적인 품질·보안 검증이 아닙니다.
원본 Skill 문서

원본 저장소의 제목, 예시, 코드, 표, 링크, 이미지를 유지해 표시합니다.

Build prompt sets for multi-image generation with consistency as the main goal.

Core rule

Always create a style anchor first. Do not treat each requested image as an independent generation.

The style anchor is the canonical source for the shared visual language of the whole set. Every later prompt must either:

  • reference the style anchor explicitly when the image tool supports reference images, or
  • restate the same anchor facts in text when reference images are not supported.

Workflow

1) Read and classify the source material

Extract the project's:

  • medium or artifact type
  • audience and purpose
  • number of images required
  • sequence or page/shot order
  • fixed entities: characters, products, props, locations, UI shell, brand assets
  • style cues: genre, rendering style, mood, palette, era, lighting
  • continuity constraints

Classify the project into one of these broad types:

  • product/ui screens
  • comic, manga, manhua, or webtoon
  • storyboard or animatic frames
  • poster or campaign series
  • illustration series
  • character or environment design pack
  • other sequential image project

If image count is missing but the source material clearly implies sections, infer a reasonable count and say so briefly.

2) Choose the consistency dimensions based on the project type

Do not use the same consistency checklist for every project. Adapt it.

Product or UI screens

Prioritize:

  • brand name and wording
  • layout shell
  • navigation structure
  • color palette
  • typography feel
  • card, table, and button style
  • interaction language
  • density level
  • recurring metrics, labels, and UI modules

Comic / manga / manhua / webtoon

Prioritize:

  • character face, hair, body type, age cues
  • outfit and signature accessories
  • environment and world details
  • art style and line quality
  • panel rhythm and reading flow when relevant
  • camera distance and shot variety
  • emotional continuity
  • prop continuity
  • time-of-day continuity

Storyboard / cinematic frames

Prioritize:

  • subject identity
  • location continuity
  • lens and camera language
  • lighting continuity
  • blocking and action progression
  • shot order
  • key props and spatial relationships

Poster / campaign / illustration series

Prioritize:

  • visual motif
  • palette
  • composition system
  • brand or campaign lockups
  • recurring subjects
  • lighting and mood
  • typography treatment when present

3) Build the consistency bible

Before writing prompts, create a compact internal “consistency bible” with the facts that must persist.

Include only the facts that matter. Typical fields:

  • project title or working label
  • visual style summary
  • subject roster
  • world / setting summary
  • palette and mood
  • recurring objects
  • non-negotiable continuity rules
  • aspect ratio and framing rule
  • what is allowed to vary from image to image

Use this bible to write every prompt.

4) Create the style anchor prompt

Write one prompt for the first image that establishes the whole project's visual language.

The style anchor should:

  • be image 1 of the set
  • clearly establish style, characters or system, setting, lighting, palette, and composition language
  • feel like the definitive reference image for the rest of the sequence
  • include enough detail that later prompts can point back to it

If useful, label it Master Prompt / Style Anchor.

5) Create the follow-on prompts

For each remaining image:

  • keep the anchor facts stable
  • change only what is needed for that specific frame, page, screen, or scene
  • preserve continuity from the previous image when the set is sequential
  • keep terminology stable across prompts
  • avoid accidentally redesigning characters, UI, props, or environments

When the target system supports reference images, explicitly start follow-on prompts with wording similar to:

  • Image A is the style anchor... preserve the same visual language...
  • Use the first image as the continuity reference...

When the target system does not support reference images, start with a short anchor recap that repeats the crucial facts in the same order each time.

6) Output structure

Default output format:

  1. Project Type
  2. Consistency Priorities
  3. Style Anchor Summary
  4. Prompt 1: Style Anchor
  5. Prompt 2...N
  6. Reference-image version if applicable
  7. Text-only continuity version if applicable

The prompts themselves are the primary deliverable.

Prompt-writing rules

  • Prefer concrete visual language over vague adjectives.
  • Keep names, descriptors, and labels consistent across prompts.
  • Reuse exact wording for important persistent details.
  • State what changes in the current image and what remains fixed.
  • For sequential stories, maintain narrative continuity from one image to the next.
  • For UI, keep repeated modules and shell structure stable.
  • For comics and storyboards, specify camera angle, shot size, action beat, and emotional beat.
  • For poster or illustration series, specify the repeating motif and what variant is introduced in the current image.

Genre-specific output hints

For comics, manga, manhua, and storyboards, use a compact prompt formula: shared style + character continuity + environment continuity + shot description + action beat + mood + composition notes

For product or UI work, use a compact prompt formula: shared design system + fixed shell + page purpose + required modules + realistic content + implementation-ready fidelity

For character packs, use a compact prompt formula: shared style + subject identity + pose/outfit variation + stable facial features + stable palette + sheet purpose

When the source material is under-specified

If the user provides a full plan but some visual details are missing, infer reasonable defaults that fit the medium and state the assumptions briefly before the prompts. Do not block on minor missing details.

Avoid

  • generating all images as if they are unrelated one-off prompts
  • changing style adjectives casually from prompt to prompt
  • redesigning the main subject between images
  • mixing multiple incompatible aesthetics unless the brief explicitly asks for that
  • outputting only a page list without the actual prompts

Supporting references

For output format guidance, see references/output-template.md. For project-type consistency heuristics, see references/consistency-playbook.md.

관련 작업