rossark-motion-skill README

项目的 README,来自 github.com/rossark0/rossark-motion-skill。合并为一个 markdown 文件:README.md。

A blue silhouette floats in with dangling feet, then a kid in a striped kit drops in, lands, kicks a ball that rolls in from the right, waves, grins and hops

Motion For Anything You Draw

For designers and engineers who want to bring their apps and drawings to life.

Making a drawing move is easy. Making it feel alive is hard: it has to breathe, look around, blink, shift its weight and act, without ever freezing or jumping. This skill helps your agent get there.

It started as the method I used to bring the illustrations in my app to life, and grew into a way to animate anything drawn: characters and creatures, plants, cloth, fire, smoke, water, weather, lights, machines, icons, logos, charts and whole scenes.

The animation above is drawn and animated entirely in code, with the templates in this repo.

Install

npx skills@latest add rossark0/rossark-motion-skill

Why use it?

Agents don't animate like animators.

Ask an agent to bring a character to life and it will often swap one drawing for another, so the face teleports from one expression to the next. It moves through each pose with its own eased tween, so the body stops dead at every pose. Everything moves in sync, like a machine. Blinks are so short they flicker, everything bounces, and a thin line shows across the face where two shapes meet.

All these small things add up, and the result looks like a slideshow instead of something alive.

This skill lists those mistakes and how to fix them: one clock for the whole scene, parts that turn around their real pivots, trailing parts that follow a beat later, idle life that never stops, smooth paths through key poses, small amplitudes and strong curves. When you have a reference video, it measures the motion frame by frame before building, because what makes a reference feel alive is often too small to see.

For motion in an interface, it follows Emil Kowalski's rules and Benji Taylor's Family values.

Reference

Demos

Reveal (source, MP4), at the top of this page. Full-body acting in a phone frame, on one clock, in a 9-second loop at 60 fps:

Garden (source, MP4), a 10-second loop:

A girl in a sun hat waters a sprout; it grows and blooms, a butterfly lands on the flower, she hops with joy, then the petals fall and the sprout returns

Chef (source, MP4), a 9.6-second loop:

A chef flips a pancake; the second flip goes too high, lands on his hat, wobbles, slides off and he catches it in the pan

Orbit (source, MP4):

A round bird with closed eyes floats while a tilted ring of cards orbits around it, with sparkles twinkling

To render them again on macOS 14+:

swiftc -parse-as-library -target arm64-apple-macos14 \
  templates/swiftui/Motion.swift templates/swiftui/LiveFace.swift demo/*.swift -o /tmp/render-demo
/tmp/render-demo reveal /tmp/frames phone fps=60   # or: garden, chef, orbit
ffmpeg -framerate 60 -i /tmp/frames/%04d.png -c:v libx264 -pix_fmt yuv420p -crf 20 demo/reveal.mp4

rossark-motion-skill (rossark0/rossark-motion-skill)

正在搜索数据库…

这是什么?这是一个 Claude Code 技能,教 AI 智能体如何把任意 SVG 图画做成逼真的 60fps 动画,涵盖动态面部、全身表演以及完整场景,适用于 React Native 和 SwiftUI。
这是一个应用,还是我的智能体要读取的技能?它是一个 Claude Code 技能:包含一个讲解原则和做法的 SKILL.md 文件,外加配套的引擎代码(React Native 用 motion.ts,SwiftUI 用 Motion.swift)和辅助脚本,而不是独立的应用。
我该如何开始和我的智能体一起使用?把 SKILL.md 放在你的项目旁边,再把 MOTION.md 和你的美术素材放在一起,并填写你的代码位置和规则;该技能的设计是让智能体在开始做动画之前先通读它。
我的项目里是否必须已经有 React Native 或 SwiftUI?是的。引擎提供两种实现:React Native 项目用 motion.ts/liveFace.tsx,SwiftUI 项目用 Motion.swift/LiveFace.swift,所以你需要其中一种技术栈。
它能用于任何编程智能体,还是只能用于 Claude Code?它是专门按 Claude Code 技能来构建和描述的,设计上围绕 Claude Code 智能体如何读取和应用技能指令。
这个技能是给谁用的?想为自己的绘画和应用插图做动画的设计师和工程师,而不必去请动画师或者逐帧手动制作。
演示是在什么平台上渲染的?随附的离线渲染器可将演示场景转换为 MP4 和 GIF 文件,它运行在 macOS 14 或更高版本上。
为什么智能体做出的动画通常看起来很僵硬?智能体往往会直接替换图画而不是移动各个部件,并且为每个姿势分别使用独立的缓动补间,导致身体在姿势之间停滞不动;另外眨眼动作短到会闪烁。这些问题加在一起,会让动作看起来像幻灯片,而不是有生命感的画面。
trackedges.py 是做什么的?它会逐帧测量参考视频中主体的边缘,以捕捉肉眼难以察觉的细微运动,比如 2% 的抖动、漂浮或摇摆,然后你再尝试用代码还原这些动作。
演示文件夹里有什么?项目页面上展示的示例场景、一个手机外框样机,以及一个可将场景转换为 MP4 和 GIF 文件的离线渲染器。
liveFace 开箱即用提供了什么?现成的动态眼睛和嘴巴,能够平滑滑动、眨眼和微笑;你只需提供角色的几何形状和颜色即可。
作者
rossark0
大小
15.1 MB
星标
850
更新
提交Oct 22
时长
27天
提交 · 90 天