Comic Panel Generator · Open Skill 公開技能
Comic Panel Generator
Comic Panel Generator
Manhua
2 0 0
https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/3949831/2026-10-13/9hbshz7j.jpg
# Comic Panel Generator Creates **comic-book / manga-style pages**: one image containing **4–7 story panels** with **irregular, editorial layouts** (not forced equal grids). Supports pure text-to-image, typography in panels, and **image-to-image** when the user uploads references. ## When to use - User asks for comic, manga, manhwa, graphic-novel, or **美式漫画** style art. - User wants a **multi-panel page** (4, 5, 6, or 7 modules) in one shot. - User uploads photo(s) or art to **blend into** a comic page. - User wants to **change panel count**, layout, style, dialogue, or redo one page. ## Progressive disclosure Do **not** load all reference files at once. After `read_skill(comic-panel-generator)`, open **only** the file for the current phase: | Phase | File | Tools | |-------|------|-------| | Plan | `references/01_plan_comic_page.md` | `recognize_image` (if uploads) | | Generate page | `references/02_generate_comic_page.md` | `generate_image` | | Upload / I2I | `references/03_image_to_comic.md` | `generate_image`, `recognize_image` | | Optional uniform split | `references/04_optional_split.md` | `split_image` | | Edits | `references/05_edit_comic_page.md` | `generate_image` | Shared rules (models, prompts, canvas): **`references/00_shared_rules.md`** — read once when starting generation or I2I. ## Collaboration - **First turn:** Confirm panel count (4–7), story beat or theme, orientation (`portrait` / `landscape` / `square`), and whether panels need **readable text** (speech bubbles, SFX, signs). If the user uploaded images, read `03_image_to_comic.md` before generating. - **One-click:** `write_todos` → plan → single comic page on canvas → `set_guidance` with next options (variant page, HD text pass, split if regular grid). - **Partial edit:** Read `05_edit_comic_page.md` only; do not replan unless the user changes panel count or story scope. - Never fabricate media URLs. On tool failure, report honestly. ## Canvas deliverables - **Primary:** One canvas asset per page — `name` like `Comic page 1` or user title; `entity_type='storyboard_images'`. - **Work-in-progress** contact sheets for splitting: `entity_type='work_in_progress'`, `add_to_canvas=false` (see `04_optional_split.md`). - **Optional:** Individual panel tiles only when the user chose a **regular grid** and confirmed split (see phase 4). ## Model channels (summary) Hard-code **`model=` channel names** only (never physical model names). Full rules in `00_shared_rules.md`. | Scenario | Channel | |----------|---------| | T2I, no readable text in image | `t2i` | | T2I, speech bubbles / captions / signage | `t2i.hd`, `t2i.hd.lite`, or `t2i.hd.gpt` | | I2I with user upload(s) | `i2i.hd.lite`, `i2i.hd`, or `i2i.hd.gpt` | Use **`list_models`** only to verify channels if a call fails; defaults in shared rules are authoritative for this skill.