中文版

huashu-art-motion（GitHub：https://github.com/alchaincyf/huashu-art-motion ，2026-10-06 创建，今晨抓取时已 5,826★）是中文社区出品的 Agent 技能：让你的 coding agent 把艺术风格写成会动的画——36 种艺术风格、9 种解说语法、8 种参数化片段，npx skills add alchaincyf/huashu-art-motion 一条命令装上。

它的做法是纯代码动画：先跑拆解脚本量出参考动画的转场、节拍网格和运动热图，再按机制用代码一层层把画面画出来、让它动起来，而不是拼贴生成图。风格配方卡有 53 张（洞穴壁画、埃及壁画、梵高、莫奈、包豪斯、8-bit、蒸汽波、新海诚……），解说语法覆盖 Kurzgesagt、Vox、白板、3Blue1Brown、storytime、动态文字、发布会、财经图表等 8 种可运行示范；还有“长卷穿越片”骨架：主角一路向右走，跨过边界的那一刻画风也跟着换。交付前还有数字验收（qa.py 量稳定、效率、动感、流畅）和“反 PPT 门禁”（整片几乎没有镜头内快动作就红灯、不出片）。样片包括 65 秒的《花叔闯进超级玛丽》和 98 秒的《Clawd 博物馆奇妙夜》，后者用的都是大都会博物馆开放获取的 CC0 原作。

为什么值得了解

• 它把“AI 做视频”从“拼生成图”拉回到了“代码驱动动画”：画面可复现、可调参、可验收，对想稳定出片的人比纯生成更可靠
• 对做内容、做演示、做作品集的人很实用：口播 + JSON 规格就能出一段时长精确到帧的解说动画段，横竖屏、可透明底
• 中文社区少见的、文档和方法论都很扎实的技能项目；作者此前同一系列的 huashu-design 也曾上过热门，喜欢这类“让 Agent 交出成品”的路数可以一起关注

English Version

huashu-art-motion (GitHub: https://github.com/alchaincyf/huashu-art-motion , created 2026-10-06, 5,826 stars at this morning’s fetch) is an agent skill from the Chinese community: it lets your coding agent write art styles into moving pictures — 36 art styles, 9 explainer grammars and 8 parameterised clips, installed with one command: npx skills add alchaincyf/huashu-art-motion.

Its approach is code-driven animation: first run a breakdown script that measures a reference animation’s cuts, beat grid and motion heatmaps, then redraw each layer in code so the picture actually moves, instead of stitching generated images together. There are 53 style recipe cards (cave painting, Egyptian murals, Van Gogh, Monet, Bauhaus, 8-bit, vaporwave, Shinkai and more), and explainer grammars covering Kurzgesagt, Vox, whiteboard, 3Blue1Brown, storytime, kinetic type, keynote UI and finance charts, each with a runnable demo; plus a “long-scroll” skeleton where a character walks rightward and the art style changes the moment they cross a boundary. Delivery passes a numeric acceptance check (qa.py measures stability, efficiency, motion and smoothness) and an “anti-PPT gate” (a film with almost no in-shot motion gets a red light and is not delivered). Demo films include the 65-second “Huashu breaks into Super Mario” and the 98-second “Clawd: Night at the Museum”, the latter redrawn from CC0 originals from The Met Open Access.

Why it’s worth knowing

• It pulls “AI video” back from “stitching generated images” to “code-driven animation”: frames are reproducible, parameterised and checkable, which is more reliable than pure generation for anyone who needs consistent output
• Very practical for content creators, presenters and portfolio builders: narration plus a JSON spec yields an explainer clip timed to the frame, in landscape or portrait, optionally with an alpha background
• A rare skill project from the Chinese community with both solid documentation and methodology; the author’s earlier huashu-design from the same series also trended, so if you like this “agent delivers a finished piece” approach, the series is worth following together