SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsstory-webtoon-art

story-webtoon-art

Design & creation

View repository

Updated 10-15 21:40

About this skill

story-webtoon-art: art prompt assembly

Takes the cut script and assembles the art promptIt is a skill that does this. It focuses on "what to instruct to be drawn and how" — locking the art style anchor, copy-paste prompts per cut, and attached image ordinal mapping. It does not do generation execution, QC, or cost: generation is handled by moai-media, QC by story-webtoon-qc.

1. Overview

This skill is only responsible up to creating the prompt . It locks the art style with an anchor so it does not waver from cut to cut, assembles cut instructions into copy-pasteable prompts, and pins down the order of reference images to attach by ordinal. Actual image generation, model selection, and credit notices are moai-mediahandled by live lookup.

2. What this skill does not do

  • No image generation, credit notices, or model selection — moai-media:media-higgsfield-core delegated to the contract (§4-Generation Execution).
  • No QC — defect judgment and consistency violation QC of output images is story-webtoon-qc under the jurisdiction of.
  • Does not create speech bubbles, fonts, or SFX snippets — story-webtoon-letteringis owned by. This skill only inserts those snippets into the prompt.
  • Does not own model prompt syntax — vendor-specific syntax such as Midjourney sref and model parameters, Gemini, GPT-image, etc. is moai-mediaowned by. This skill does not import it (conflicts with the media family when porting).

3. Pre-checks

  • ${CLAUDE_PLUGIN_ROOT}/skills/story-webtoon-spec/references/manuscript-specs.md — vertical scroll canvas specifications and color space (sRGB). Reference when reflecting manuscript specifications in the prompt.
  • ${CLAUDE_PLUGIN_ROOT}/skills/story-webtoon-lettering/references/balloon-dictionary.md — If it is a branch with dialogue, bring the speech bubble snippet per cut from here and insert it.

4. Workflow

Step 1 — Lock the art style anchor

Decompose the series art style into 7 tokens (palette, technique, outline, color plane, texture, mood, font mood) and make it into an English STYLE block. Never omit the content guard sentence — Without it, it copies and draws even faces, houses, and landscapes from the style reference. For detailed and copy-paste templates, see references/style-anchor.md.

The art style anchor (7-token instance) is held by this skill per series. The art style token registry itself is moai-designerowned by, and this skill only has series anchors through the extraction procedure.

Step 2 — Assemble prompts per cut

Cut script (story-webtoon-episode) into copy-paste prompts.

  • Korean prompt body + English STYLE block mixed — if the body describes the scene, camera angle, characters, and actions in Korean, and only the STYLE block is written in English, the art style is captured clearly.
  • Character fixed appearance, color, and consistency anchors Explicitly repeat for every cut(to prevent color drift).
  • If dialogue is included, below the cut description 💬 [lettering 스니펫] + 위치 + 한국어 대사insert.

Step 3 — Attachment ordinal 1:1 mapping

Attach the reference images (character ref, style grid, previous cut) in order, and in the first paragraph of the prompt map them 1:1 with "the Nth image is X."

  • Header attachment order ↔ code block ordinals must match exactly in count, order, and name, character for charactermust.
  • Characters each get their own individual ordinal — no lumping together ("the first images are the person and the cat" ❌).
  • The number of attachments differs per cut — recount only what was actually attached, in the actual order. Dynamic ordinal calculation and the 4-line self-verification are references/reference-ordinal.md.

Step 4 — Multi-image session context management

Save context when generating multiple cuts consecutively in one session.

  • Text checkpoint handoff — hand off confirmed cuts as text summaries to reduce vision re-analysis.
  • Crop only suspicious cuts — instead of vision-analyzing everything every time, crop and check only suspicious cuts.
  • The ceiling is not the number of pages but the number of vision analyses. For details, see references/context-budget.md.

Step 4-Generation execution — delegate to moai-media

The actual generation, cost preflight, and model selection of the assembled prompt are moai-media:media-higgsfield-core contract (models_explore live lookup + get_cost advance notice).

If moai-media is not installed: output the completed cut prompt as text and guide the user to generate directly on the Higgsfield web (https://higgsfield.ai)에서 (including the attachment order banner).

5. Output format

Each cut prompt is assembled in the order: attachment order banner → ordinal mapping paragraph → body → STYLE block.

⚠️ 이 프롬프트는 참조 이미지를 함께 첨부해야 작동합니다.
   첨부 순서: ① 캐릭터A ref → ② 스타일 그리드 → (연속 시) ③ 직전 컷 결과

첫 번째 이미지는 [캐릭터A] 인물 참조입니다.
두 번째 이미지는 스타일 참조입니다 — 색·질감·화풍만 참조, 그 속 인물·사물·풍경은 베끼지 마세요.
[캐릭터A 고정 외형·색·앵커 한 줄]을 매 컷 같게 유지.

[한국어 본문 — 장면 / 화각 / 인물 / 동작 / 방향 고정]
💬 [lettering 스니펫] (대사 포함 분기면)

=== STYLE (match the ATTACHED reference image; use ONLY its palette/texture/technique,
and IGNORE its subject matter — faces, houses, landscapes, animals. Do NOT copy any of those objects) ===
- Palette / Technique / Linework / Color fill / Texture / Mood: [7토큰]

→ Generation execution: moai-media:media-higgsfield-core delegate.

6. Precautions

  • Verify the obligation to label AI-generated content — when submitting or posting AI-generated images to commercial platforms, verify the obligation to label AI-generated content (for details, see moai-lawyer's AI-related checklist).
  • Never omit the content guard — the STYLE block's IGNORE its subject matter if the phrase is missing, the subject in the style reference gets copied.
  • Attachment ordinals are calculated every time. As people are added, later ordinals shift — do not hardcode fixed ordinals.
  • Colors and appearance are explicitly repeated for every cut. If they are only in the reference and not in the prompt, the color changes from cut to cut.
  • Do not make up credits or models — moai-media looks them up.
  • Review is not done here — after output, hand off to story-webtoon-qc.

7. Related skills

Before

  • story-webtoon-episode — cut script
  • story-character-sheet — character reference and consistency anchor
  • story-webtoon-lettering — per-cut speech bubbles and SFX snippets

After

  • story-webtoon-qc — output image defect and consistency review
  • story-series-bible — update completed episode status

Delegation

  • moai-media:media-higgsfield-core — Generation execution, cost preflight, model selection

8. References

File Load conditions
references/style-anchor.md When style 7-token decomposition, English anchor blocks, and content guards are needed
references/reference-ordinal.md When attachment ordinal 1:1 mapping, dynamic calculation, and self-verification are needed
references/context-budget.md When multi-image session context saving and crop review are needed
../story-webtoon-spec/references/manuscript-specs.md Canvas specifications and color space check
../story-webtoon-lettering/references/balloon-dictionary.md Insert speech bubble snippet when dialogue is included

9. Sources

  • Style 7 tokens, content guards, attachment ordinal mapping, multi-image session management: aitoon-comic (internal asset port).
  • Generation, cost, model: moai-media:media-higgsfield-core live contract.
  • Vendor prompt grammar: moai-media owned (this skill only references it).
  • Original copywriter: Jo Nam-kyung (https://www.facebook.com/Bmisty)