video-shotcraft:让 AI Agent 直接制作院线级产品宣传片
video-shotcraft: Let an AI Agent Produce Cinema-Grade Product Films
by Mycelium Protocol
产品宣传片的制作流程通常是:设计师画分镜 → 动效工程师实现动画 → 音效师配音 → 反复审片迭代。即使是一个 30 秒的产品 Demo,从零到完成也要几天。
video-shotcraft 把这个流程交给了 AI Agent:给 Agent 一个产品 URL,它负责分镜设计、动画实现、音效配置,最终输出一个可以直接发布的 1920×1080 视频。
GitHub: https://github.com/Vincentwei1021/video-shotcraft | ⭐ 3832
核心内容
104 张镜头配方卡
每张卡片记录一个具体的视觉镜头的完整信息:
- 用途(Purpose):这个镜头适合表达什么
- 能量感(Energy):节奏快慢、紧张程度
- 建议时长
- 参数说明:缓动曲线、运动幅度等
- 实现注意事项
- 已知坑
104 张卡分布在 10 个功能类别中,覆盖从产品特写到场景过渡的常见镜头需求。Gallery 在线可搜索、可过滤,复制镜头卡名称交给 Agent 即可使用:
161 种运动风格
每种风格都有对应的动态预览(mp4),不用猜效果,直接在 Gallery 里选好,告诉 Agent 要哪几张卡的组合。
完整视频模板:Ink Press
开箱即用的产品宣传片模板:
- 时长:36.2 秒
- 规格:1920×1080,30fps
- 镜头数:10 个
- 风格:纸墨琥珀色,2.5D 真实页面摄像机运动,标题卡,转场,完整电影级音效通道
Agent 把你产品的截图、文案、品牌色换进去,直接产出同等质量的成片。
149 个音效 + 5 套背景音乐
音效按 16 个场景/材质类别组织:
transition impact riser camera ui text paper film light data scifi mech glass fluid crowd counter
选用逻辑:先选类别(对应场景语气),再选音色(粗糙/细腻/金属感等)。配有专门的 sound-design.md 说明每个文件的用法。
快速开始
最直接的方式:把仓库链接直接给 Agent:
Install this skill for me: https://github.com/Vincentwei1021/video-shotcraft
Agent 会自动 clone 并链接到技能目录。或手动安装:
# 用 skills CLI
npx skills add Vincentwei1021/video-shotcraft
# 手动 clone + 链接
git clone https://github.com/Vincentwei1021/video-shotcraft.git
cd video-shotcraft
ln -s "$(pwd)" ~/.claude/skills/video-shotcraft # Claude Code
# 或
ln -s "$(pwd)" ~/.codex/skills/video-shotcraft # Codex
安装后,直接告诉 Agent 要做什么:
Use video-shotcraft to create a promo for my desktop product.
Use the deck-deal-flyin and row-embed shot cards to present this feature.
Use video-shotcraft to make a promo with the Ink Press template.
如果不指定镜头卡,Agent 会先介绍内置模板并询问是否使用——推荐先从 Ink Press 开始,替换产品资产后就能得到一个完整的成片。
仓库结构
video-shotcraft/
├── SKILL.md # Agent 入口,核心制作规则
├── references/
│ ├── pipeline.md # 端到端生产流程
│ ├── shots/ # 104 张镜头配方卡(10 个功能类别)
│ ├── sequences/ # 可复用的完整视频结构和序列模式
│ ├── aesthetic-rules.md # 视觉 QA 标准
│ ├── music-beat-sync.md # 背景音乐分析和节拍同步方法论
│ └── sound-design.md # 音效指南和示例
├── demos/ # Remotion 参考实现(同 shots 分类)
├── gallery/ # 静态运动预览 Gallery
├── template/ # 可运行的完整视频模板
└── assets/
├── lib/ # 可复用 Remotion 组件
├── scripts/ # 页面资产抓取脚本
└── audio/
├── bgm/ # 5 套背景音乐
└── sfx/<category>/ # 149 个音效,16 个场景类别
技术细节:Remotion 渲染
video-shotcraft 用 Remotion 做视频渲染——本质上是把 React 组件渲染成视频帧,每帧都是一个确定性的 React 状态,方便 Agent 精确控制每一帧的内容和动画参数。
无头服务器渲染注意事项(2 核 Linux,Node 22):
| 问题 | 表现 | 解决 |
|---|---|---|
| 并发上限 | ”Maximum for —concurrency is 2” | 传 --concurrency=1 |
| Headless Chrome | 新版 Chrome 移除了旧 headless 模式,直接调 chromium 会失败 | 改用 chrome-headless-shell 二进制 |
| CDN 不可达 | remotion.media 被墙,无法自动下载 headless-shell | 传 --browser-executable=<本地路径> |
镜头卡的制作背景
104 张镜头配方卡通过研究一批产品宣传片提炼而来,参考来源包括 ClickUp、Perplexity、Slack、Notion、Figma、Framer、Bear、Raycast、Pitch、Miro、Superhuman、Loom 的官方产品片。
记录的是这些视频的运动语言(时序、缓动曲线、镜头编排)——所有 Remotion 实现都从头重写,不含任何原始视频/图像/品牌素材。
这个项目本身也是用 Claude Code + 这套方法论迭代开发出来的,包括 QA 过程。
Mycelium Protocol — 追踪 AI 系统的底层演化
关于 Mycelium
菌丝协议。持续追踪 AI 工具、系统和实验的内容节点。
video-shotcraft: Let an AI Agent Make Cinematic Product Videos
by Mycelium Protocol
Product video production typically goes: designer storyboards → motion engineer animates → sound designer scores → multiple review rounds. Even a 30-second product demo can take days from zero to finished.
video-shotcraft hands this entire pipeline to an AI agent: give it a product URL, and it handles storyboarding, animation, and sound design — the output is a ready-to-publish 1920×1080 video.
GitHub: https://github.com/Vincentwei1021/video-shotcraft | ⭐ 3832
What’s Inside
104 Shot Recipe Cards
Each card captures a specific visual shot in full detail:
- Purpose: what emotional or narrative goal this shot serves
- Energy: pacing, tension level
- Suggested duration
- Parameters: easing curves, motion scale
- Implementation notes
- Known pitfalls
The 104 cards span 10 functional categories covering product close-ups through scene transitions. Browse and filter them in the live Gallery, copy card names, and hand them to your agent:
161 Motion Style Previews
Every style has a live mp4 preview — no guessing at the effect. Pick what you want in the Gallery, then tell the agent which cards to combine.
Complete Video Template: Ink Press
A validated, production-ready promo template:
- Duration: 36.2 seconds
- Spec: 1920×1080, 30fps
- Shots: 10
- Style: paper-ink-amber, 2.5D real-page camera moves, title cards, transitions, full cinematic SFX pass
The agent swaps in your product’s screenshots, copy, and branding to reproduce the same quality. Fastest path to a finished film.
149 SFX + 5 BGM Tracks
SFX organized into 16 scene/material categories:
transition impact riser camera ui text paper film light data scifi mech glass fluid crowd counter
Pick category first (matches scene mood), then timbre. See sound-design.md for per-file usage guidance.
Quick Start
Most direct path — hand the repo link to your agent:
Install this skill for me: https://github.com/Vincentwei1021/video-shotcraft
The agent clones and links it to the skills directory. Or install manually:
npx skills add Vincentwei1021/video-shotcraft
# Or manually:
git clone https://github.com/Vincentwei1021/video-shotcraft.git
ln -s "$(pwd)/video-shotcraft" ~/.claude/skills/video-shotcraft # Claude Code
# or
ln -s "$(pwd)/video-shotcraft" ~/.codex/skills/video-shotcraft # Codex
Then tell the agent what you want:
Use video-shotcraft to create a promo for my desktop product.
Use the deck-deal-flyin and row-embed shot cards.
Use video-shotcraft with the Ink Press template.
If no shot cards are specified, the agent introduces the built-in template first and asks whether to use it. Starting with Ink Press and swapping in your product assets is the fastest path.
Repository Structure
video-shotcraft/
├── SKILL.md # Agent entry point + core production rules
├── references/
│ ├── pipeline.md # End-to-end production workflow
│ ├── shots/ # 104 shot recipe cards (10 functional categories)
│ ├── sequences/ # Reusable full-video structures
│ ├── aesthetic-rules.md # Visual QA criteria
│ ├── music-beat-sync.md # BGM analysis + beat-sync methodology
│ └── sound-design.md # Sound design guidance
├── demos/ # Remotion reference implementations
├── gallery/ # Static motion-preview Gallery
├── template/ # Runnable complete video template
└── assets/
├── lib/ # Reusable Remotion components
├── scripts/ # Page-asset capture scripts
└── audio/
├── bgm/ # 5 BGM tracks
└── sfx/<category>/ # 149 SFX, 16 scene categories
How Remotion Rendering Works
video-shotcraft uses Remotion to render video — React components rendered to video frames. Every frame is a deterministic React state, which lets the agent precisely control every frame’s content and animation parameters.
Headless server rendering notes (2-core Linux, Node 22):
| Issue | Symptom | Fix |
|---|---|---|
| Concurrency cap | ”Maximum for —concurrency is 2” | Pass --concurrency=1 |
| Old headless removed | Recent Chrome dropped old headless; system chromium fails | Use chrome-headless-shell binary instead |
| CDN blocked | remotion.media unreachable, auto-download rejected | Pass --browser-executable=<local path> |
Where the Shot Cards Come From
The 104 cards were distilled from studying outstanding product films from ClickUp, Perplexity, Slack, Notion, Figma, Framer, Bear, Raycast, Pitch, Miro, Superhuman, and Loom. The cards document motion language — timing, easing, choreography — re-implemented from scratch in Remotion TSX. No footage, artwork, or brand assets from the original films are included.
The toolkit itself was built and iterated with Claude Code using the same workflow it teaches.
Mycelium Protocol — tracking the deep evolution of AI systems
© 2026 Mycelium Protocol. All rights reserved.
关于本站 · 免责声明
🍄 Mushroom Research Blog 是非营利、免费公开的个人科技观察博客与公众号 XStack18,不接受商业合作、不代表任何企业或机构立场,也不谋求商业利益。我们以个人视角客观中立地记录和分析 AI、Web3 等领域的最新模型发布与技术动态——不止转述新闻标题或二手信息,而是给出有独立思考的深入分析,希望帮更多人获得有价值的一手科技认知。
⚠️ 文中介绍的开源代码与模型,仅供学习交流与技术借鉴。它们大多仍处于早期阶段,有待进一步研究和验证,请勿直接用于工作或生产环境;如需采用,请先自行充分测试,并核实其许可证与安全性。
Open-source code and models featured here are shared for learning and reference only. Most are early-stage and still need further study and verification — please don't use them directly in your work or in production. Test them thoroughly and check their licenses and security first.
- 本站文章均为作者基于公开信息的个人研究与观点整理,不代表文中提及的任何公司、产品、模型的官方立场,未与其构成商业关联或合作关系。
- 科技行业信息更新极快,我们尽力保证内容准确、及时,但不对完整性、实时性做绝对保证,具体请以相关企业/项目官方公告为准。
- 文中引用的第三方商标、产品名称、图片、数据等版权归原权利人所有,我们会尽量注明来源;如你认为存在版权疑问或侵权,请通过下方邮箱联系我们,收到通知后会尽快核实处理(更正、加注来源或删除)。
- 文章内容仅为技术科普与个人观点,不构成投资、法律或其他专业建议,据此进行任何决策的后果需自行判断和承担。
📮 侵权 / 勘误 / 合作咨询:[email protected]
💬 评论与讨论
使用 GitHub 账号登录后发表评论