Ian Xiaohei Svg Motion

Ian Xiaohei SVG Motion

把 Ian 小黑正文配图的原则，改造成可控的 HTML + SVG + GSAP timeline 动效。目标不是复刻生图像素，而是把文章里的一个认知动作重建成分层漫画舞台，方便录视频、按字幕 cue 对齐、后续改文案和元素。

Core Rule

Do not auto-vectorize a PNG and animate the resulting path soup. Use raster images only as composition references. Rebuild the scene as semantic SVG groups:

idea -> shot plan -> SVG layer plan -> HTML/SVG template -> GSAP timeline -> static + motion review

Workflow

1. Extract one cognitive anchor from the article, script, screenshot, or storyboard scene.
2. Choose one physical metaphor: sorting, carrying, bridging, leaking, catching, folding, weighing, opening, rerouting, or falling.
3. Make Xiaohei perform the core action. If removing Xiaohei does not change the meaning, redesign.
4. Design a 16:9 white canvas with one main scene, 35%+ whitespace, and 3-5 short handwritten annotations at most.
5. Build semantic SVG groups for every moving object.
6. Animate with GSAP timeline. Use x, y, rotation, scale, opacity, and SVG stroke drawing. Avoid layout animations.
7. Save both a playable HTML page and a static review screenshot.
8. If this is for a 口播成片 project, map timeline segments to subtitle/cue timing before final render.

Read When Needed

• Read references/style-rules.md before designing a new scene or judging whether the result still feels like Ian Xiaohei.
• Read references/svg-layering.md before writing or editing the SVG structure.
• Read references/motion-rules.md before writing GSAP animation.

Output Contract

Default output folder:

output/-xiaohei-svg-motion/

Required files:

index.html
README.md
preview.png
vendor/gsap.min.js

Optional files:

source.png        # reference image, if provided by the user
cue-map.md        # when aligning to spoken script or subtitles

Reusable Assets

AI model and interface icons live in:

assets/icons/

Use assets/icons/index.json as the source of truth. Keep original SVG files in subfolders and inline their complete SVG content into scene SVGs during generation, so gradients, strokes, and brand colors are preserved.

Icon rules:

• If index.json already has the exact model, provider, or semantic icon, use that icon. Do not replace ChatGPT, Claude, Flash lightning, Step, DeepSeek, Qwen, bug, table, chart, database, or web icons with generic boxes, stars, sparks, or abstract marks.
• When the user provides a new SVG icon, save the raw SVG under assets/icons//, add an entry to index.json, and then reference it by id in the scene plan.
• For comparison scenes, icon choice is part of the meaning. Keep model icons visually consistent in size and placement so the viewer can identify the actors before Xiaohei starts moving objects.

Template

Use assets/templates/xiaohei-comic-motion/ as the starter when the request resembles a hand-drawn comic scene with moving objects, arrows, annotations, or a Xiaohei action.

Use assets/templates/xiaohei-series-reference/ when the request asks for multiple variants, a reusable series, or a broader reference set for 口播 explanation scenes.

Copy the template into the project output folder, then edit:

• SVG group names.
• Handwritten labels.
• Object positions.
• Timeline labels and durations.
• README notes for the scene-specific metaphor.

Quality Gate

Before final response:

• Open the HTML in a local server.
• Check console for errors.
• Capture a static screenshot with ?static=1.
• Capture at least one playback mid-frame when the scene has multiple beats. If available, use ?frame=mid for a deterministic review frame.
• Visually inspect that text, arrows, Xiaohei, props, belts, pits, cards, and final impact points do not overlap.
• Confirm the animation has visible progression: context first, action second, problem/result third, summary last.
• Confirm annotations appear only after the object/action they explain.
• Remove decorative or semi-transparent guide arrows if the object motion already shows direction.
• Confirm Xiaohei is part of the action.
• Confirm the page is not just a PNG in an `` tag unless the user explicitly asked for PNG-only camera moves.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: Agentchengfeng
• Source: Agentchengfeng/chengfeng-videocut-skills
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.