handdraw-style-prompter

Hand-drawn Style Prompter

This is the install entrypoint for the complete hand-drawn style package. The package root contains the images/ gallery and numbered reference assets; the operational contract is skills/handdraw-style-prompter/SKILL.md.

Before handling a request, read that Skill file completely. Resolve every gallery and reference-image path from this installed package root. Install this repository at path . so the images/ directory and the nested Skill are kept together.

Request routing index

When the user asks to illustrate an article, plan article illustrations, choose insertion points, or write prompts for article images, read and invoke article-illustration-planner before handling the request. Use the hand-drawn style rules in handdraw-style-prompter for style-number resolution and prompt construction.

When the user asks to design a poster, generate poster prompts, or create structured poster prompts, read and invoke poster-prompt-generator, actively recommending cohesive style and theme color pairings from the hand-drawn style library (#001–#312) and theme color library (C-01–C-36).

When the user asks to design an article cover, generate cover prompts, create a WeChat official account cover, or design an X/social cover for an article (e.g., "make a cover for the article", "article cover design", "official account cover", "X article header image"), read and invoke article-cover-designer before handling the request. It extracts a ~200-word summary, infers target audience, sets the business scenario, injects style (#001–#312) and theme color (C-01–C-36), and applies the fixed prompt instruction ("First design the metaphor then generate the image, the main title is obvious, there is little or no small text. You design the rest for me.").

When the user asks to blend, merge, or fuse two styles, generate a cross-media fusion prompt, or create a "character visual language × scene visual language" artwork (e.g., "[Style Fusion Design]", "style fusion design", "style fusion", "fuse two styles", "character and scene style fusion", "dual style fusion", "cross-media style fusion"), read and invoke style-fusion-prompter before handling the request. It supports fusing two hand-drawn style numbers (#001–#312), or a style number with a realistic style, optionally injects layouts (SC-/IG-) and theme colors (C-01–C-36), and applies the strict dual visual language coexistence template.

When the user asks to design an IP character, build a Character Bible or IP Design Manual, generate standard IP cards (e.g., "IP character design", "character design manual", "design three-view and merchandise", "IP-01 to IP-18 cards"), read and invoke ip-designer before handling the request. It supports 5 layers (Strategy, Character, Visual, System, Commercial) across 18 standard cards (6-card basic, 10-card standard, or 18-card full edition).

When the user asks to manage, add, query, replace, or delete assets in the custom library (e.g., "add to character library", "save to prop library/scene library", "replace character base image", "modify asset name/tag", "initialize/mount gallery directory", "view gallery inventory"), read and invoke custom-asset-manager before handling the request. It manages the full lifecycle of custom character (CH-), prop (PR-), and scene (SCN-*) assets with zero Git conflicts.

When the user asks to plan, script, or direct a knowledge or tutorial AI video, generate A/B-roll visual shot lists, or structure video production (e.g., "make a tutorial video", "video storyboard table", "knowledge video editing", "A/B-roll arrangement", "video director"), read and invoke knowledge-video-director before handling the request. It orchestrates voiceover scripts, acoustic alignment, and 7-column visual shot lists alternating between white-background IP theater (A-roll) and dark-background motion graphics/demos (B-roll).

When the user asks to plan a photo shoot, create a photography project, generate shot lists, design couple/wedding/family/portrait/cosplay photography plans, or draw card for a themed shoot (e.g., "shooting plan", "couple photos", "wedding photos", "portrait", "family portrait", "Cosplay shoot", "draw card", "8-grid", "photography storyboard"), read and invoke couple-photo-orchestrator before handling the request. It orchestrates the full pipeline from direction selection through Look design, scene micro-zones, pose-emotion planning, Share Card compilation, and 8-shot adaptive grid generation.

Default mode

Default to creating prompts only. Do not call an image-generation tool unless the user explicitly asks to generate, render, or preview an image.

Automatic Style and Theme Color Recommendation Policy (intelligent style and theme color recommendation by default)

No matter what image or prompt is generated,by default, if the user has not specified a style and theme color, the optimal combination of style and theme color must be automatically recommended, and it is strictly forbidden to interrupt or refuse generation because the user lacks a number:

[!IMPORTANT] [Core principle: each time judged dynamically in real time by AI, strictly forbidden to hardcode/mechanically apply]

• Real-time dynamic assessment: It is strictly forbidden to maintain any static, rigid mapping of "theme keyword -> fixed style number/color number" in memory, rules, or prompts (for example, it is strictly forbidden to recommend 268 whenever solar terms/tradition appear, recommend 266 whenever nature appears, recommend 011/054 whenever technology appears, or recommend 018 whenever daily life appears).
• Open-ended matching across the entire library: Each recommendation must be made by AI based on the deep context, spiritual core, visual metaphor, and image composition of the current user's specific theme, within the entire library of 312 hand-drawn styles (#001–#312) and 36 classic theme colors (C-01–C-36), performingreal-time, open-ended aesthetic reasoning and dynamic pairing。
• Stimulate diversity across the entire library：The same theme possesses multidimensional aesthetic possibilities from different design perspectives (e.g., "vegetable garden" can be ink-wash freehand, pastoral woodcut print, coarse-grained oil pastel, or Bauhaus geometric composition). Each recommendation should be conceived according to the specific entry point, fully demonstrating the rich vitality of the entire library's 288 painting styles and 36 colors.
• Concretization of aesthetic rationale：The aesthetic recommendation rationale given by AI must be closely tied to the visual metaphor and pictorial composition of the current theme, explaining why the brushstrokes/texture of that painting style and the emotion of that color can perfectly convey this theme, and eliminating boilerplate.

1. Neither style nor theme color specified：Based on the theme semantics, emotional tone, audience, and usage scenario of the user input, AI dynamically recommends in real time from 312 hand-drawn styles and 36 classic monochrome libraries the single combination with the highest degree of fit [style number (#001–#312) + theme color number (C-01–C-36)]. There is no hard limit on the number of theme colors; depending on the pictorial layers, flexibly decide on monochrome dominance, two-color pairing (primary color + accent color), or three-color harmony, briefly describe one sentence of concretized aesthetic recommendation rationale, and directly output the complete image-generation prompt (or execute image generation).
2. Only style specified, theme color not specified：Strictly retain the user-specified style, and based on that style and the pictorial theme, AI dynamically recommends in real time the most harmonious [theme color (C-01–C-36)] pairing (which may be a monochrome or multicolor combination).
3. Only theme color specified, style not specified：Strictly retain the user-specified theme color, and based on the color tonality and pictorial theme, AI dynamically recommends in real time the single best-matching [hand-drawn style (#001–#312)] pairing.
4. Both specified：Output completely in accordance with the user-specified content.

Style activation policy

Use the same capability decision for prompt-only and explicit image-generation requests. Resolve the selected/current model (or model family) against skills/handdraw-style-prompter/references/model_capabilities.json; if no model is specified, default to gpt-image-2 (the calibrated baseline model of this library). Only use the default unknown capability fallback when the user explicitly requests an uncalibrated third-party model. The decision uses the indexed author name plus the generated style name and is not based on the author's fame or life status.

• name_activation=strong: use only the indexed author name, generated style name, and theme. (Specifically for styles like #042 Beatrix Potter, #011 David Shrigley, #269 Socialist Realist Propaganda Poster, and other strong-activation styles: the model understands the aesthetic natively; output only the reference author and style name, and STRICTLY DO NOT output core traits, local image file paths, or reference-isolation blocks.)

• Otherwise, include every available positive core trait with the author/style name and theme.

• If name plus traits is not strongly activated, also require the configured reference asset. For explicit generation, pass it through referenced_image_paths. In pure-image prompt-only output, write the local asset path and reference-isolation instruction inside both prompts for the user to upload manually. In graphic-text prompt-only output, do not place a path, upload instruction, or isolation block inside either copyable prompt; show the resolved reference image to the user outside the prompts instead. Assets live under the installed package root in numbered 200-style buckets: for example, #217 uses images/individual/201-400/217_grid.webp.

• If no positive core trait exists and name is not strongly activated, use the author/style name, theme, and reference image.

• If the model identifier or its capability entry is unavailable, treat it as unknown, include any available positive traits, and require the image as the safe fallback.

• Use python skills/handdraw-style-prompter/scripts/resolve_reference.py --model <model> --style <number> when a deterministic decision check is useful. The script prints JSON and never guesses an unknown model's capability.

• The resolver reports activation_source as name+style, name+style+traits, name+style+traits+reference-image, or name+style+reference-image, plus filtered prompt_traits and the local reference_path when required.

• Resolve the style number to its configured reference asset relative to the installed package root. The normal fallback is images/individual/{bucket}/{number}.webp (for example, 048 maps to images/individual/001-200/048.webp); a matching {number}_grid.webp in the same bucket takes priority.

• When a reference image is required, inject the following reference-isolation block into pure-image prompt-only output and every actual image-generation prompt. Do not inject it into a graphic-text copyable prompt; the actual image-generation prompt still receives it together with the attached asset.

  Chinese: 所附图片仅用于参考画风。只提取参考图的风格特征，例如线条、笔触、媒介、材质、色彩倾向和整体视觉语言；不要使用、复制或延续参考图中的任何主体、人物、动物、服装、道具、动作、姿态、场景、背景、构图、布局、文字或故事。最终画面内容完全以用户提供的主题为准。

  English: Use the attached image only as a style reference. Extract only its stylistic qualities, such as linework, brushwork, medium, material texture, color tendencies, and overall visual language. Do not use, copy, or carry over any subject, person, animal, clothing, prop, action, pose, setting, background, composition, layout, text, or story from the reference image. The user's written theme is the sole source for the image content.

• The user's theme is the sole source for subjects and narrative; the reference image must never override or add content to the theme.

• Include only positive visible traits; filter clauses containing 避免, 不要, 不准, or 禁止 and do not copy the traits field mechanically.

• If the numbered image is missing or cannot be passed, report that limitation and provide the normal text prompts; never invent or substitute a reference image.

• After generation, identify whether the numbered reference image was used. If used, identify its number. Do not imply generation when the user requested prompts only.

Session initialization

On the first turn in the current Codex task/thread where this Skill is invoked, perform both initialization actions before handling the user's request. This applies to Any first request, including opening the gallery, browsing the index, requesting prompts, or requesting image generation:

1. If this task has not already displayed the gallery, dynamically resolve the absolute file:// URI to skills/handdraw-style-prompter/gallery/index.html based on the current workspace or installed package root, and call mcp__codex_app__open_in_codex in a browser tab with target.type="browser" and that URL. Never open gallery/index.html as target.type="file", in an editor, or as a file preview.
2. Include the exact standalone status line 当前处于纯图模式，可切换为图文模式。 in the same final reply, even when the request produces no prompt.
3. Continue with the user's request after the browser call; opening the gallery must not block prompt generation or an explicitly requested image-generation follow-up. If the same first reply would also add the pure-image mode note, show this status line only once.
4. Do not repeat the browser call or this first-session status notice on later turns in the same task/thread. Use the conversation context (not a persistent state file) to determine whether initialization already happened.
5. If the browser call is unavailable or fails, provide the dynamically resolved local file:// fallback link, then continue normally.

This initialization applies only when this Skill is invoked for the first time in a task/thread; unrelated conversations must not open the gallery.

Inputs

For style-only work, if the style number (001–312) is omitted, automatically recommend an optimal style and theme color combination based on theme semantics. For layout work, require a layout ID (SC-001 or IG-001); if style and/or theme color are omitted, automatically recommend harmonious ones. Accept optional theme color (C-01–C-36 or color name), aspect ratio, subject constraints, text requirements, and a mode. When the user requests a prompt (e.g. specifying a style and/or layout) without providing a theme, do NOT pause or halt to ask for a theme—immediately output the complete prompt with a clear placeholder (such as 【请在此输入画面主题，或在生图模型中垫入你的照片】 / [Enter theme here, or attach your photo in the image AI]). If a supplied number, layout ID, or color ID is invalid, ask the user to choose a valid indexed value; do not invent one. Do not add an aspect ratio when none was supplied.

Users can browse skills/handdraw-style-prompter/gallery/index.html for numbered style contact sheets, skills/handdraw-style-prompter/gallery/layouts.html for layout thumbnails, and skills/handdraw-style-prompter/gallery/colors.html for classic monochrome theme colors. The authoritative style content is styles_200_reorganized.md; skills/handdraw-style-prompter/references/styles.json is a generated index and must be refreshed with python skills/handdraw-style-prompter/scripts/build_library.py after Markdown changes. Layout metadata is skills/handdraw-style-prompter/references/layouts.json; each entry's bilingual prompt file under skills/handdraw-style-prompter/references/layouts/ is the authoritative layout content and the layout gallery is refreshed with python skills/handdraw-style-prompter/scripts/build_layout_gallery.py. Monochrome color metadata is skills/handdraw-style-prompter/references/colors.json and refreshed with python skills/handdraw-style-prompter/scripts/build_color_gallery.py.

Layout workflows

• A layout ID selects composition and text structure, not illustration style. Use its prompt file verbatim in the selected output language, then append the user's theme as the only source of subject matter and copy.
• Dual Architecture of Layout Patterns： All 161 layout patterns in this library are clearly divided into two major categories in terms of underlying design and execution mechanism:
  1. Deterministic Static Templates (direct plain-text stitching type · Static Templates)：
     • Includes the vast majority of conventional patterns (such as SC-001SC-022 social media cards,IG-001IG-035 infographics,SB-001SB-068 comic panels,IP-001IP-013 IP design,EC-001~EC-024 e-commerce).
     • The compositional topology is single and fixed, and the image-generation model can understand it plainly and unambiguously. The handling method is ready-to-use: directly extract its prompt template and stitch it with the specified painting style and theme for output.
  2. High-dimensional Dynamic Parsing Layouts (Skill-level dynamic parsing decision type · Generative Frameworks)：
     • The representative pattern is SC-021(adaptive dual-panel photo translation social media card).
     • This type of pattern is not a fixed wireframe of a single route, but a highly abstract "visual composition and spatial grammar system." The image-generation model does not have the ability to autonomously parse a multiple-choice menu; if the candidate list is directly laid out flat, the model will degrade due to semantic conflicts into mechanically copying the original photo's panoramic background.
     • Execution mechanism：The AI assistant must act as an Art Director to perform upfront parsing and dynamic decision-making:
       • Analyze the visual features and emotional tone of the photo or theme uploaded by the user;
       • Strict rotation of composition grammar (Round-Robin mechanism): If the user has not explicitly specified a composition, a strict sequential rotation must be performed among [Out-of-Bounds Frame-Breaking Type], [Miniature Floating Island Type], [Memory Map Tiling Type], [Classic Floating Landing Type], and [Positive-Negative Shape Interlocking Type], and it is strictly forbidden to output the same style consecutively, ensuring visual diversity and high-level aesthetics;
       • Dynamically generate a strong background purification instruction for the current image (resolutely remove environmental impurities such as original stone walls, roads, and trees, ensuring 40%–80% pure negative space);
       • Strict canvas aspect ratio preservation and dual-panel size rules: When splicing, it is strictly forbidden to modify or stretch the original image ratio; after adding the art area, the overall image size must be strictly twice the original image width (side-by-side) or twice the original image height (stacked top and bottom), with each area occupying 50%, joined seamlessly and aligned at the center line;
       • Final compiled outputunambiguous, exclusive, strongly constrainedexecution prompt words, ensuring the image generation model accurately presents rich and varied artistic compositions.
• Image upload and image editing rules (image upload is not mandatory; user image upload and editing are fully supported): Regardless of whether using SC-021(adaptive dual-panel photo translation social media card) or any other style/image type, user image upload is anoptional action：
  • When the user has not uploaded an image: Plain text direct output is fully supported. The AI assistant directly concatenates the layout rules and specified style features into a complete prompt (if no theme is provided, provide intuitive placeholders), and must never block the conversation or force the user to upload an image; the image generation model natively and autonomously generates the realistic image and style translation.
  • When the user uploads an image / edits an image (realistic to hand-drawn / image-guided style transfer): Fully supported! When the user uploads a photo or reference image requesting 'image editing,' 'style change,' or 'hand-drawn conversion,' in order to prevent the image generation model from mechanically overlaying filters on the surface of the realistic photo, the prompt mustuse the following fixed conversion/redraw instruction： 【请只提取人物的五官特征和姿态，场景轮廓，在画风上严格按 选择的风格重新画，不要在原照片上加质感。】 (English equivalent:[Extract only the character's facial features and posture, and the scene contours. Strictly redraw from scratch according to the selected style; do not add texture or filters onto the original photo.]) The AI assistant extracts the core features of the reference image (character appearance features, pose, clothing silhouette, overall scene) as the theme description, and unconditionally appends the above fixed instruction, forcing the image generation model to redraw from scratch in the hand-drawn dimension, completely eliminating realistic filters and photo micro-surface retention.
  • The core principle is: the final pixel-level visual rendering and art style translation are always natively completed by the image generation model (such as GPT Image, Midjourney, Flux, etc.).
• When a layout ID is selected, automatically use graphic-text mode even if the user did not name a mode (except for E-commerce layouts EC-*: e-commerce layouts do not append the generic graphic-text suffix, and do not stitch a visual style unless explicitly requested by the user). For standard layouts, the layout prompt and the fixed graphic-text suffix are additive (stacked): append the fixed generic graphic-text suffix verbatim at the end of the prompt to combine concrete layout rules with graphic-text semantic guidance. In layout requests, never append the pure-image mode notice 当前处于纯图模式，可切换为图文模式。, and never include reference-image file paths or upload instructions inside the copyable prompt.
• If the user also supplies a style number, combine the selected style's author/style label, permitted positive traits, and reference-image policy with the layout prompt (for e-commerce layouts, only stitch style if the user explicitly specified a style number). A style reference image may influence only rendering style; it must not override the selected layout, text structure, or theme.
• Return a single complete layout-combination prompt in the language of the user's theme. Chinese characters select Chinese; otherwise use English. Treat mixed input as Chinese. Existing style-only requests retain their normal bilingual output.
• The layout gallery is browse-only: it has no prompt input form. It shows a numbered thumbnail, opens a large preview on click, and copies the canonical Chinese layout prompt on request.

Prompt modes

Default to pure-image. Accept 纯图模式 / 图文模式 in conversation and --mode pure-image|graphic-text in the CLI.

• pure-image: preserve the current prompt workflow. After the normal reply, add the small note 当前处于纯图模式，可切换为图文模式。.

• graphic-text: preserve the user's theme exactly after 主题： / Theme:. Do not expand, paraphrase, interpret, or add scene elements, characters, actions, metaphors, emotional explanations, or theme commentary. Append the following text verbatim to the end of both the Chinese and English prompts. Do not translate, trim, normalize, rewrite, label, or reorder any character in it. The exact final prompt, including this suffix, is sent unchanged to the image AI when generation is requested. When a reference asset is required, display that image outside the copyable prompts; do not expose its local path, upload instruction, or reference-isolation block in either prompt.

  【如果主题直白包含画面元素那就按主题出图，文案由你来升华，但是不要直接描述画面。 如果主题比较概念化，那么文案和主题尽量保持一致，如果文案较长由你提炼，由你先设计画面隐喻（人类和非人类都行）再出图 。 文字参与构图，图文一体】

Output

For a valid request, return these parts:

1. (If intelligent recommendation is triggered) Recommended combination description:
   • Label the recommended style and theme color (for example:💡 智能推荐组合：风格 #{编号} · {generation_name} + 主题色 {id} · {name_zh} ({name_en})）；
   • Give 1 concise aesthetic recommendation reason;
   • Prompt the user that if they have other preferred numbers, they can state a replacement at any time.
2. Selected style: number and generated style name. Include core traits inside the two copyable prompts when the resolved activation policy requires them; do not create a separate core-visual-traits section.
3. Chinese prompt: begin the copyable prompt itself with 风格名称：{generation_name}。(Important rule: The generated prompt content must strictly not contain any numbers, including style numbers such as #018, image type numbers such as SC-021 / SC-001, color numbers such as C-01 , etc.; directly use natural language names to avoid the image generation model mistakenly drawing number codes as text in the image); when a theme color is applied, include 主题色：{name_zh}（{name_en}）。(excluding the color number); describe only the user's theme and constraints explicitly provided by the user, and always include the indexed author/style name as a short 参考作者/风格名称 label. When required, append filtered positive traits as 核心风格特征：...; when the user requests image editing/image-guided style transfer, after the image requirements must append:【请只提取人物的五官特征和姿态，场景轮廓，在画风上严格按 选择的风格重新画，不要在原照片上加质感。】(“Selected style” can be instantiated specifically according to the name of the chosen style). In pure-image mode, append a required reference image's local path and reference-isolation instruction. In graphic-text mode, preserve the theme verbatim, append the exact fixed suffix from Prompt modes at the end, and show any required reference image outside the prompt instead. Traits describe rendering style only and must not replace or alter the user's subjects, actions, setting, or story.
4. English prompt: begin the copyable prompt itself with Style name: {generation_name}. (Strictly omit any numeric or alphanumeric IDs such as #018, SC-021, or C-01 from the prompt text); when a theme color is applied, include Theme color: {name_en}.; describe only the same theme and user-provided constraints, and always include the indexed author/style name as Reference author/style name. When required, append the same traits as Core style traits: ...; when the user requests an image redraw/style transfer from an uploaded photo, append: [Extract only the character's facial features and posture, and the scene contours. Strictly redraw from scratch according to the selected style; do not add texture or filters onto the original photo.] (where "selected style" may be replaced by the chosen style's English name). In pure-image mode, append a required reference image's local path and reference-isolation instruction. In graphic-text mode, preserve the theme verbatim, append the exact fixed Chinese suffix at the end, and show any required reference image outside the prompt instead. Do not append generic composition advice, quality claims, negative prompts, or the fixed style anchor.
5. A brief note: the prompt can be pasted into any image AI; generation is controlled by that AI.

Do not invent visual traits, extra style descriptions, generic quality/composition language, or default avoid-list wording. Include each entry's original reference author/style name from the index in both prompts as requested; this is an index label, not a claim about the person or an instruction to imitate them. Never use fame or life status as a proxy for model capability.

In pure-image mode, describe only concrete visible content implied by the theme—subjects, actions, objects, environment, and mood when needed. In graphic-text mode, use the user's theme verbatim and leave all semantic expansion to the fixed suffix. In both modes, leave composition, layout, visual richness, quality, and rendering decisions to the image AI unless a layout ID was explicitly selected. Respect a user-specified text requirement but do not invent copy.

Utilities

• From the installed package root, rebuild the derived index and gallery: python skills/handdraw-style-prompter/scripts/build_library.py
• Rebuild the layout gallery: python skills/handdraw-style-prompter/scripts/build_layout_gallery.py
• Rebuild the color gallery: python skills/handdraw-style-prompter/scripts/build_color_gallery.py
• Split contact sheets into numbered single images: python skills/handdraw-style-prompter/scripts/split_contact_sheets.py
• Validate all source/index/gallery invariants: python skills/handdraw-style-prompter/scripts/validate_library.py
• Produce a deterministic CLI prompt draft: python skills/handdraw-style-prompter/scripts/prompt_style.py --style 18 --theme "秋天的第一杯奶茶"
• Produce a layout-combination draft: python skills/handdraw-style-prompter/scripts/prompt_style.py --layout SC-001 --style 18 --theme "秋天的第一杯奶茶"
• Produce a color-combination draft: python skills/handdraw-style-prompter/scripts/prompt_style.py --style 18 --color C-01 --theme "秋天的第一杯奶茶"
• Resolve image-reference policy: python skills/handdraw-style-prompter/scripts/resolve_reference.py --model <model> --style 18

The CLI is a convenience check. For normal conversational use, write natural bilingual prompts rather than echoing its template mechanically.