Motion skill
Skill LunaAI519/motion-skill
Claude Code skill:一张参考图 → 复古剪贴报风格动画短视频(Nano Banana 关键帧 + Kling 图生视频),内置版权红线自检
npx -y skills add LunaAI519/motion-skillAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 25 days oldThe repository was created 25 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Motion风格动画生成:参考图 → Nano Banana 生成关键帧 → Seedance 2.0 生成视频。触发词:/motion、动画、把图动起来、motion动画、图生视频。
SKILL.md
4.2 KB, ~1.5k tokens by cl100k_base, as published. Nobody here has run it
Motion 动画流水线
把一张参考图变成 Motion 风格短视频。全程中文交流,步骤简短。
输入
用户需提供:
- 参考图:本地图片路径(拖进终端即可得到路径)
- 动作描述:想让画面怎么动(如"镜头缓慢推近,头发被风吹动")
- 可选:比例(默认 9:16 竖屏)、时长(默认模型默认值)
缺参考图时也可以纯文字:让 Nano Banana 直接按描述生成首帧。
流程
风格预设:复古剪贴报(默认)
不要 AI 电影感写实风(一眼假、无记忆点)。默认用复古拼贴 Motion Design,配方:
- 高饱和纯色纸纹背景:砖红/钴蓝/墨绿
- 主体 = 黑白半调网点(halftone)老照片剪贴,白色贴纸描边+投影
- 一幕一个概念:用视觉双关表达要讲的那个概念词
- 撕纸标签 + 粗黑中文概念词(生图用 pro 模型保证文字)
- 少量几何装饰:黑三角/米白圆/锯齿线
- 动效克制:元素依次贴上、定格动画式微抖、镜头完全固定
关键帧 prompt 骨架:Retro mid-century collage art, 1950s magazine cut-out style. Solid [颜色] paper background with subtle paper grain. Black-and-white halftone photograph cutouts with white sticker borders and drop shadows: [概念隐喻场景]. Small geometric accents. Torn paper label with bold black condensed Chinese characters "[概念词]". Flat graphic composition, stop-motion paper collage aesthetic, no photorealism.
视频 prompt 骨架:Stop-motion paper collage animation, handmade cutout style. [主体动作]. All paper cutout elements wobble subtly with stop-motion jitter. Camera completely static. Flat 2D collage aesthetic, no 3D depth, no camera movement.
第 1 步:生成关键帧(Nano Banana)
用 mcp__nanobanana__gemini_generate_image:
reference_images: [用户参考图路径],保持风格/人物一致prompt: 描述首帧画面(英文写 prompt 效果更好)aspect_ratio: 与最终视频一致(默认 "9:16")- 需要更高质量时先
set_model为 "pro"
如需首尾帧两张(动作幅度大的场景),再生成一张尾帧,use_image_history: true 保持一致性。简单动作只要首帧一张即可,别多生成。
生成后把图片路径发给用户确认,满意再进入下一步(视频要花积分,别浪费)。
第 2 步:上传帧到 Higgsfield
本地图片不能直接传路径,先用 higgsfield 的 media 上传工具(ToolSearch 搜 media_upload)拿到 media_id。
第 3 步:生成视频
用 mcp__higgsfield__generate_video:
model: "kling3_0_turbo"(默认。seedance_2_0 效果类似但需要 Higgsfield Pro/Ultimate 套餐且贵 3 倍,套餐不够会报 403,报错就换回默认)medias: [{ "value": 首帧media_id, "role": "start_image" }],有尾帧再加 end_imageprompt: 动作/运镜描述(英文)aspect_ratio与帧一致- 先带
get_cost: true预查积分,告诉用户花费,确认后再真正提交
提交后轮询结果,完成后把视频链接给用户。如返回 recovery_tool 立即调用。
第 4 步(可选):增强
用户嫌糊时用 mcp__higgsfield__upscale_video(bytedance provider,需传源视频宽高)。
版权红线(每次生成前自检,一条都不许碰)
- 不生成真人肖像:马斯克、明星、任何可辨识的真实人物。讲公司画产品/场景隐喻,不画老板的脸
- 不出现品牌 logo/商标:prompt 里写 "no logos, no brand marks",生成后肉眼复查
- 不生成受版权保护的角色:Labubu、迪士尼、动漫角色等一律用抽象隐喻替代
- 参考作品只学配方不搬画面:分析别人的视频可以,其中任何一帧都不得发布
- 配乐提醒用户:用免版权音乐或无声,别配流行歌
规则
- 每一步产出先给用户看,确认再花积分
- prompt 用英文写给模型,跟用户交流用中文
- 一次只生成 1 个视频(count=1),不满意再迭代
Gives 0 of the 12 instructions most css styling skills give in ~1.5k tokens
Counted across 586 of the 596 authors here whose files we hold, read 2026-08-06
- avoid excessive centered layoutsin 55 of 586, across 12 files
- bundle code into single HTML filein 54 of 586, across 14 files
- Respect prefers-reduced-motion user settingsin 52 of 586, across 35 files
- avoid purple gradientsin 51 of 586, across 11 files
- avoid uniform rounded cornersin 51 of 586, across 11 files
- avoid Inter fontin 51 of 586, across 11 files
- edit generated files to develop artifactin 50 of 586, across 10 files
- animate only transform and opacity propertiesin 43 of 586
- Make touch targets at least 44x44 pixelsin 41 of 586, across 15 files
- Ensure minimum color contrast of 4.5:1in 39 of 586, across 10 files
- use tailwind cssin 39 of 586, across 24 files
- Use SVG icons instead of emojisin 38 of 586, across 11 files
Said here and by no other author read
- communicate with the user in Chinese
- write image and video prompts in English
- use default retro collage style without photorealism
- set model to pro for higher quality text
- ask user to confirm before spending credits
- poll video generation result and return link
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.