Skip to content
Back to search

oil-visual

制作统一漫画墨线风格的解释图与透明角色插图,用于概念、流程、比较和文章配图。

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/oil-oil--oil-visual/badge.svg)](/listings/oil-oil--oil-visual)

HTML

<a href="/listings/oil-oil--oil-visual"><img src="/api/listings/oil-oil--oil-visual/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

把 [oil-visual 仓库地址](https://github.com/oil-oil/oil-visual) 交给 Agent,要求安装这个 Skill;也可以使用命令:

npx skills add oil-oil/oil-visual

让当前宿主重新加载 Skill,然后在对话里点名使用:

Use $oil-visual to explain how the main Agent assigns work to subagents.
Use short, exact Chinese labels inside the image.
Use $oil-visual to draw the glasses stick figure reviewing a blueprint,
with the warm-yellow Border Collie sitting beside the desk.
用 oil-visual 画一个单独拿着铅笔的小人,透明背景,不要狗。
用 oil-visual 画一只单独坐着的暖黄色边牧,透明背景,不要人物。
用 oil-visual 画一个同风格的文件夹和几张散开的纸,透明背景,不要人物或狗。
用 oil-visual 画几颗同风格的星芒装饰元素,透明背景,不要角色。

完整规则、提示词结构和交付检查都写在 [`SKILL.md`](./SKILL.md) 里。

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 208 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. ---
  2. name: oil-visual
  3. description: "制作统一漫画墨线风格的解释图与透明插画素材,可单独绘制眼镜线条人物、暖黄色牧羊犬、两者组合,或同风格的物件与装饰元素。当用户需要概念、机制、流程、对比的完整解释画面,或可复用的排版插图时使用;不用于普通代码图表、照片处理或无关品牌视觉。"
  4. ---
  5. # Oil Visual
  6. Create raster visuals in one shared manga-ink language. Choose one output mode before generating; do not mix the two production paths.
  7. ## Choose the output mode
  8. ### Mode A — explanatory image
  9. Use when the image must explain a concept, mechanism, workflow, comparison, or tradeoff by itself.
  10. - Deliver a complete PNG or WebP with a finished off-white scene.
  11. - Generate every essential title and label directly inside the bitmap.
  12. - Make the relation visible through objects, paths, states, or repeated materials; labels identify the evidence but do not replace it.
  13. - Do not generate an unlabeled base and add essential words in a separate rendering step.
  14. ### Mode B — transparent illustration
  15. Use when a character, object, or decorative illustration will be composed into a hero, document, card, slide, or other layout.
  16. - Request a genuinely transparent PNG when the selected generator supports native alpha. Otherwise use a perfectly uniform chroma-key background that does not occur in the artwork. Default to `#00FF00`; use `#FF00FF` when the subject contains green.
  17. - Do not include explanatory labels unless the user explicitly requests text inside the illustration.
  18. - Preserve native alpha when it is clean. For a chroma-key source, remove the background with the bundled `scripts/cutout.py`. Deliver a transparent PNG in either case.
  19. - Keep the transparent artwork as a reusable visual asset; the surrounding layout supplies the title and explanatory copy.
  20. If the destination is unclear, choose Mode A when the image itself must communicate the idea and Mode B when another layout will carry the explanation.
  21. ## Shared visual language
  22. - Draw confident black manga/comic ink outlines with varied line weight and restrained circular halftone screentone.
  23. - Treat the recurring characters as optional subjects, not a required pair. When a person is selected, draw the minimal stick-figure protagonist with a round head, thin round glasses, dot eyes, a simple smile, and thin line-drawn limbs; when a dog is selected, draw the chubby warm-yellow Border Collie. Select characters only when requested or when they visibly help explain the claim; never add them merely to signal the style.
  24. - For object-only artwork, carry the same style through clear silhouettes, varied ink line weight, restrained circular halftone, and simple material details. Do not add a person, dog, or face to make an object feel on-brand.
  25. - When characters are present, keep them secondary to the subject's evidence or action unless the character itself is the requested subject.
  26. - Use black, white, and halftone gray as the base. Use warm yellow for the dog when present; without the dog, warm yellow may appear as a small light patch or sparse star accent, but is not required.
  27. - Add at most two muted semantic colors. Common mapping: blue = input/content, orange = action/warning/cost, purple = process, green = successful result.
  28. - Avoid 3D, glossy gradients, photorealism, wobbly sketch lines, generic card grids, dashboards, unrelated clutter, and watermarks. Deliberately requested decorative elements are valid subjects.
  29. ## Mode A workflow — explanatory image
  30. ### 1. Write the visual brief
  31. ```text
  32. viewer_question: what should be understood in 10 seconds?
  33. concrete_claim: one-sentence conclusion
  34. real_objects: visible objects, interfaces, documents, tools, or states
  35. relation: comparison, transformation, causality, sequence, hierarchy, feedback, tradeoff, or pipeline
  36. visual_evidence: what must remain understandable when labels are ignored?
  37. scene: believable setting and 2–4 useful environmental cues
  38. semantic_colors: what each accent color means
  39. labels: exact short strings plus the evidence surface for each label
  40. ```
  41. Show the input, action or relation, and result. Keep one dominant focal action and no more than three major visual regions. For multiple steps, use a simple left-to-right or top-to-bottom sequence.
  42. ### 2. Design the labels
  43. - Prefer 2–6 labels. Use more only when the explanation truly needs them.
  44. - Keep each label short and concrete: role, action, state, or outcome.
  45. - Place every label on or immediately beside its evidence surface, such as a desk nameplate, task sheet, folder tab, machine, meter, lane, or result document.
  46. - Use modern Chinese sans-serif typography, medium or bold, large enough to read at the intended display size.
  47. - Do not turn body copy, commands, tables, or long paragraphs into image text. Use a deterministic layout method when dense or editable text is required.
  48. Add this block to the generation prompt:
  49. ```text
  50. Text (verbatim): Render these exact labels as part of the bitmap illustration:
  51. "<label 1>", "<label 2>", "<label 3>".
  52. Use each phrase exactly once. Do not translate, paraphrase, misspell, repeat,
  53. or add any other text. Use modern sans-serif medium/bold typography, large
  54. and readable. Place "<label 1>" on <evidence surface>; place "<label 2>" on
  55. <evidence surface>; place "<label 3>" on <evidence surface>.
  56. ```
  57. ### 3. Build the prompt
  58. Use this order:
  59. 1. State the concrete claim and shared task.
  60. 2. Describe the real setting, selected subjects, and any action. Include the protagonist or dog only when requested or when they visibly help explain the claim.
  61. 3. Describe the evidence objects and their geometry: aligned, nested, connected, split, transformed, repeated, or converging.
  62. 4. Assign semantic colors.
  63. 5. Quote the exact labels and specify each placement.
  64. 6. Add the Mode A style anchor.
  65. 7. End with exclusions.
  66. Mode A style anchor:
  67. ```text
  68. Professional editorial manga/comic ink illustration. Clean confident black ink outlines with varied line weights, expressive but controlled. Use classic circular halftone screentone for gray and shadow areas. Render only the selected characters and objects; do not add a protagonist or dog automatically. Use an off-white lightly textured real environment, not a blank white canvas. Typography is modern sans-serif, medium or bold, large and readable. Color is restrained: black, white, halftone gray, warm yellow when useful, plus at most two muted semantic accent colors. No 3D, no glossy gradients, no photorealism, no generic card grid, no dashboard, no unrelated clutter, no tiny text, no long paragraphs, no watermark.
  69. ```
  70. ### 4. Inspect and retry
  71. 1. Inspect the output at original resolution.
  72. 2. Compare every label with the brief character by character. Confirm that each appears exactly once and that no stray text was added.
  73. 3. Reject missing, duplicated, invented, or misspelled labels.
  74. 4. Regenerate with one targeted correction while repeating all scene and style invariants. Do not conceal an error with a separate text layer.
  75. Use this retry instruction:
  76. ```text
  77. Keep the scene, composition, characters, objects, colors, and all correct labels unchanged.
  78. Change only the incorrect text "<wrong>" to the exact text "<right>".
  79. Do not add, remove, translate, or repeat any other text.
  80. ```
  81. ## Mode B workflow — transparent illustration asset
  82. ### 1. Select the subject
  83. Identify exactly which subjects are requested: the person alone, the dog alone, both together, or an object or decorative element without characters. Describe one clear action when an action is needed, and include only the props needed to establish it. For a standalone object, describe its shape, material, and distinguishing details instead. Leave generous padding around the artwork so the cutout can be composed safely.
  84. ### 2. Build the prompt
  85. Describe the selected subjects first. State explicitly that no unrequested person, dog, face, or prop should appear. Then append the fixed style anchor and one background rule below.
  86. ```text
  87. Style: professional manga/comic ink illustration. Clean confident ink outlines
  88. with varying line weights, thick for contours and thin for details, not wobbly
  89. or sketchy. Use classic circular halftone screentone dot patterns for gray and
  90. shadow areas. Give characters and objects the same clear silhouette and ink
  91. language. Color usage is extremely restrained: mostly black, white, and gray
  92. halftone; warm yellow on the dog when present, or as an optional small light
  93. patch or sparse star accent when the dog is absent. Use no more than two muted
  94. semantic accent colors. Render only the requested subjects.
  95. Do not let halftone, ink, props, or the subject touch the image border. Keep
  96. generous padding. No text, no watermark. PNG format.
  97. ```
  98. Use this background rule when the generator supports genuine transparency:
  99. ```text
  100. Render on a genuinely transparent background with a clean alpha channel.
  101. No backdrop, floor plane, cast shadow, or opaque pixels outside the artwork.
  102. ```
  103. Otherwise replace `<KEY_COLOR>` with the selected hex color and use this rule:
  104. ```text
  105. The background must be a perfectly uniform flat <KEY_COLOR> rectangle with zero
  106. gradient, texture, noise, speckles, shadows, floor plane, or lighting variation.
  107. ```
  108. ### 3. Validate the source
  109. - Inspect the image before removal.
  110. - For native alpha, confirm the image is RGBA, all four corners have alpha `0`, and no opaque background or edge residue remains. Keep the alpha intact; do not run the cutout script.
  111. - For a chroma-key source, confirm all four corners are uniform and visually match the chosen key color. Reject backgrounds with gradients, texture, shadows, speckles, or artwork touching the border.
  112. - Preserve a chroma-key source alongside the transparent result until the output is approved.
  113. ### 4. Remove a chroma-key background only
  114. Skip this step for clean native-alpha output. For a uniform chroma-key source, install Pillow if the active Python environment does not have it, then run:
  115. ```bash
  116. python3 scripts/cutout.py source.png transparent.png
  117. ```
  118. Optional tuning:
  119. ```bash
  120. python3 scripts/cutout.py source.png transparent.png \
  121. --transparent-threshold 12 \
  122. --opaque-threshold 220
  123. ```
  124. The script samples the image border, builds a soft alpha matte from color distance, and removes color spill from antialiased edges. It works with any uniform key color, so the key can be chosen to avoid the subject palette.
  125. ### 5. Validate the transparent result
  126. - Confirm the output is RGBA and all four corners have alpha `0`.
  127. - Confirm every requested subject remains complete. Check glasses and thin limbs for a person, ears and tail for a dog, and narrow contours or small details for an object.
  128. - Check for a gray fringe at 100% zoom.
  129. - Confirm internal white and halftone areas were not erased.
  130. - Regenerate the source instead of forcing the algorithm when the background is visibly uneven.
  131. ## Output handling
  132. - Save approved project assets inside the current project or output directory.
  133. - Do not leave project-referenced images only in the generator's default storage.
  134. - Use versioned filenames instead of overwriting an approved asset unless the user explicitly requests replacement.
  135. - Report the final prompt, output mode, final image path, and the transparency path used. Include the chroma-key source path and any non-default cutout options when cutout was needed.
  136. ## Quality gate
  137. For every output:
  138. - The subject is recognizable in about 3 seconds.
  139. - When the output explains something, the main action or relation is clear in about 10 seconds.
  140. - Only requested characters appear; when present, they serve the subject rather than becoming generic decoration.
  141. - Line work and halftone remain consistent; warm yellow and semantic accents follow the selected subjects and meaning.
  142. For Mode A:
  143. - One claim, one focal action, and no more than three major visual regions.
  144. - The visual evidence still shows the relation when labels are ignored.
  145. - Every required label is exact, appears once, and is integrated into the correct evidence surface.
  146. - All labels remain readable at the intended display size.
  147. For Mode B:
  148. - The output has real transparency, whether supplied natively or produced by clean background removal.
  149. - Thin details and internal halftone regions remain intact.
  150. - No source background, fringe, shadow, or border artifact remains.
README.md · README · 82 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. <p align="center">
  2. <img src="./assets/readme/hero.svg" width="100%" alt="oil-visual:漫画墨线、半调网点、小人和黄色边牧组成的配图风格">
  3. </p>
  4. 用同一套漫画墨线风格,制作两类可直接使用的图片:一类是带短标签、能独立讲清概念或流程的完整解释图;另一类是可放进演示稿、文章或卡片的透明插画素材。
  5. 透明素材可以只画戴眼镜的小人、暖黄色边牧、两者组合,也可以只画文件夹、工具等物件。角色按画面需要出现,物件不必靠加一个人或一只狗来维持风格。
  6. ## 配图风格
  7. - 用粗细分明的黑色墨线画轮廓和细节。
  8. - 用圆形半调网点表现灰面和阴影,不使用柔和渐变。
  9. - 戴圆框眼镜的极简小人和胖胖的暖黄色边牧是可选角色,可以单独出现、一起出现,或都不出现。
  10. - 物件与装饰元素沿用同一套轮廓、墨线、网点和克制配色,不自动添加角色。
  11. - 完整解释图使用米白色纸张质感,只保留少量真正帮助理解的环境物件;独立插画交付透明背景。
  12. - 狗出现时使用暖黄色;没有狗时,暖黄色可作为少量点缀,也可以不用。每张图最多再使用两种语义色。
  13. - 图片需要文字时,使用短而准确的无衬线标签,并直接放在对应物体上。
  14. ## 示例
  15. ### 完整解释图
  16. <p align="center">
  17. <img src="./examples/agent-workflow-square.png" width="32%" alt="主 Agent 分配任务并统一验收">
  18. <img src="./examples/explainer-cost-comparison.png" width="32%" alt="分工清楚与重复处理的成本对比">
  19. <img src="./examples/feedback-loop.png" width="32%" alt="输入、处理、结果与反馈组成的循环">
  20. </p>
  21. <p align="center">
  22. <img src="./examples/pipeline-bottleneck.png" width="32%" alt="并行处理疏通任务瓶颈">
  23. <img src="./examples/from-complex-to-clear.png" width="32%" alt="把复杂输入整理成清楚结果">
  24. </p>
  25. 同一套墨线、网点和语义色,可以表现机制、对比、循环、瓶颈和信息整理。
  26. ### 透明插画素材
  27. <p align="center">
  28. <img src="./examples/explore-folder-spotlight.png" width="49%" alt="透明插画素材:放大镜从一排文件夹中找出黄色文件夹,没有人物和狗">
  29. <img src="./examples/simplify-trim.png" width="49%" alt="透明插画素材:剪刀剪去杂乱纸条,旁边留下整齐文件,没有人物和狗">
  30. </p>
  31. 两张都是原始透明 PNG:左图用放大镜和文件夹表现“定位”,右图用剪刀和纸张表现“精简”。它们只包含物件,没有人物、狗、文字或整页背景。
  32. ## 安装
  33. 把 [oil-visual 仓库地址](https://github.com/oil-oil/oil-visual) 交给 Agent,要求安装这个 Skill;也可以使用命令:
  34. ```bash
  35. npx skills add oil-oil/oil-visual
  36. ```
  37. 让当前宿主重新加载 Skill,然后在对话里点名使用:
  38. ```text
  39. Use $oil-visual to explain how the main Agent assigns work to subagents.
  40. Use short, exact Chinese labels inside the image.
  41. ```
  42. ```text
  43. Use $oil-visual to draw the glasses stick figure reviewing a blueprint,
  44. with the warm-yellow Border Collie sitting beside the desk.
  45. ```
  46. ```text
  47. 用 oil-visual 画一个单独拿着铅笔的小人,透明背景,不要狗。
  48. 用 oil-visual 画一只单独坐着的暖黄色边牧,透明背景,不要人物。
  49. 用 oil-visual 画一个同风格的文件夹和几张散开的纸,透明背景,不要人物或狗。
  50. 用 oil-visual 画几颗同风格的星芒装饰元素,透明背景,不要角色。
  51. ```
  52. 完整规则、提示词结构和交付检查都写在 [`SKILL.md`](./SKILL.md) 里。
  53. ## License
  54. [MIT](./LICENSE)
  55. ## 配置、依赖与使用边界
  56. 无需额外配置,但需要宿主具备生图和读图能力。生图工具能直接输出透明 PNG 时保留原生透明通道;只能输出纯色背景时,用附带的 `cutout.py` 抠图,这条备用路径需要 Python 和 Pillow。费用与凭据取决于所选生图工具。
  57. 提示词与参考图会交给所选生图工具。交付前应检查图片中的短标签和透明边缘;密集正文、可编辑图表应使用确定性排版工具。

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.