English | 中文 | 日本語

carl-motion

29 motion-graphics styles, one copy-paste prompt each. Paper collage, synthwave, ink wash, woodcut, point cloud, claymation, risograph, glass UI and more. Every style comes with an 8-second example, the prompt, the mistakes that usually break it, and for 20 of them the full source code that rendered the clip.

Install the Skill, then tell your agent make my topic in the risograph style:

npx skills add LearnPrompt/carl-motion

Claude Code also works as a plugin: /plugin marketplace add LearnPrompt/carl-motion then /plugin install carl-motion@carl-motion.

Contents

• Start here
• The 29 styles
• Skill
• Templates and source
• How the examples were made
• Upstream skills and references
• Media and licenses
• Acknowledgements
• Author

Start here

Nothing to install for the first try. Pick a look in the gallery below and open its page. Copy the prompt block and replace your topic with your own subject. Paste it into Claude, ChatGPT or any model that writes code, and ask for a 1920×1080, 30 fps, 8-second MP4. If the result looks off, read that page's Where it usually breaks section and run it again.

The 29 styles

The order follows the video we cut from them: neighbouring styles sit on roughly complementary colors, so each cut reads as a clear change. 24 of the 29 play here. The other 5 only have third-party examples we can't redistribute, so their pages link to the source instead.

01 · Retro Paper Collage 复古纸张拼贴 source included · Python	02 · Product UI 2.5D Camera 产品界面2.5D运镜 prompt + source link	03 · Neon Synthwave 霓虹复古Synthwave source included · Blender
04 · Hand-drawn Paper Notes 纸面手绘笔记 prompt + source link	05 · Watercolor 水彩手绘 source included · Python	06 · Black-and-white Woodcut 黑白木刻版画 source included · Python
07 · Soft Inflatable 3D 软体充气3D source included · Blender	08 · Point-cloud Survey 点云测绘／剥层 source included · Blender	09 · Blueprint Drawing 蓝图工程制图 source included · Python
10 · Die-cut Sticker Explainer 贴纸科普 source included · Python	11 · Feature-film 3D Character 动画电影式3D角色 prompt + source link	12 · Chinese Ink Wash 中国水墨 source included · Python
13 · ASCII Character Visuals ASCII字符视觉 prompt + source link	14 · Isometric 2.5D World 等轴2.5D世界 prompt + source link	15 · Van Gogh Impasto 梵高厚涂油画 source included · Node
16 · Iridescent Psychedelic Shader 虹彩迷幻Shader prompt + source link	17 · Claymation Stop-motion 黏土定格 source included · Blender	18 · Pixel Game 像素游戏 prompt + source link
19 · Flat Vector 扁平矢量 source included · Python	20 · Impressionist Oil (Monet) 印象派油画／莫奈 source included · Node	21 · Particle Morph 粒子聚散造型 source included · Python
22 · White-model Projection 实体白模投影 source included · Blender	23 · Ancient Egyptian Mural 古埃及壁画 source included · Node	24 · Product Exploded View 产品爆炸拆解 source included · Blender
25 · Editorial Paper Data Story 纸面编辑式数据叙事 prompt + source link	26 · Glassmorphism UI 玻璃拟态UI source included · Node	27 · Brick Toy 积木玩具 source included · Blender
28 · Risograph Print Risograph孔版套印 source included · Python	29 · Sci-fi Scan HUD 科幻扫描HUD prompt + source link

Skill

carl-motion turns this library into something your coding agent can use. Ask for a style by name or by look (flat, warm, hand-made). The skill tells the agent to run scripts/resolve_brief.py first. An exact name opens that card and, when this repo has a preview clip, measures frames from that clip; a catalog URL is not treated as watched. A look that matches none of the 29 is not forced onto one. When the style ships source code here, the agent starts from that engine instead of a blank file, on the stack that engine already uses. Every card says how far the style is proven (implemented, upstream only, or research), and the finished clip has to pass the same delivery gate.

Templates and source

Scene engines. 20 styles ship the code that rendered their B clip, all at 1920×1080, 30 fps, 240 frames. Every engine has the same entry, ./render.sh <work_dir>, writes 240 PNG frames, and hands them to scripts/deliver.py, which encodes the clip and returns a red, yellow or green verdict. scripts/doctor.py <engine> checks the machine first. Contract: scene-contract.md.

9:16 explainer layout. skills/carl-motion/pipeline/script_layout.py stacks two clips of the same style (A over B), adds a fixed title bar, and puts the result first: from frame 0 each style shows its name and a prompt card already half written, the rest streams in like model output, and a warning strip lands at 5.4 s. A presenter box on the right is a placeholder for a talking-head shot.

Pipeline. The rest of skills/carl-motion/pipeline is the tooling behind the video: encode.sh (one H.264 spec for every clip), qa.py (format, black-frame and jump-cut checks), color_order_v4.py (orders styles so neighbours sit on complementary colors), color_stitch_verify.py (re-measures every cut on the finished video) and opener_salience.py (scores candidate openers in CIELAB). See its README.

How the examples were made

Every style is a pair of 8-second clips. A is an upstream reference; B is a second example in the same style, so viewers can see the look carry over to a different subject. 20 of the B clips were written and rendered from scratch for this repo by Claude Opus 5.5, from a style prompt plus the A clip as reference, and seven of them retell one short story, how cats moved in with the first farmers, each in its own style. Every clip passed the same checks: 1920×1080, 30 fps, exactly 240 frames, full decode, no black frames, and every flagged jump frame inspected by eye.

Upstream skills and references

These are the style skills and galleries this library was built on. Their prompts are linked from each style page.

The eight-video-skill source guide (Chinese, checked 2026-11-07) maps the original roundup to this library, including the current WTT name and the archival-doc entry with no repository URL in the source post. Additional references are not installed or tested capabilities.

Media and licenses

Code, prompts and our own renders (assets/previews, assets/posters for styles marked source included, everything under skills/) are MIT. Preview clips for four styles are LemoLab demo films under CC BY 4.0 and are credited on their pages. Clips from prompt-motion.com, X, yasue.org and video-shotcraft are not copied here: their pages link out. The 9:16 sample frame uses a huashu (MIT) A clip and our B clip.

Acknowledgements

• lemomo-ai/lemo-opuscar, Vincentwei1021/mg-styles-15, alchaincyf/huashu-art-motion for the open style recipes.
• prompt-motion.com and yasue.org for public references.
• LearnPrompt/awesome-seedance for the gallery-plus-Skill format.

Author

Carl (@卡尔的AI沃茨) started GoodCase and maintains this repo. Six years of large-model R&D at ByteDance, Alibaba and Baidu before turning to AI video, then a run of AI shorts that made it onto Chinese TV. 500k+ followers across platforms. Follow along on X @aiwarts or the WeChat official account 卡尔的AI沃茨. Questions and collaborations: [email protected].