Vibemovie

io.github.pooriaarab由 GitHub 账号 pooriaarab 发布v0.2.3更新于 Oct 24, 2026

Turns an agent coding session into a short recap video. Local-first.

发布方自己的服务:AI Benchmark Index

已验证STDIO仅桌面开发者工具媒体与设计

概览

AI 生成的概览

把智能体编码会话的事件记录转换成一段带解说的短片回顾,输出为自包含的动画 HTML 页面。

可以做什么
VibeMovie 接收会话事件的 JSON 数组(如 task-done、tests-pass、pr-merged 等),生成电影式回顾:故事线、章节、代码高亮、统计浮层、配乐以及可选解说。MCP 模式会暴露一个可供智能体调用的 render 工具,同一套流程也可通过命令行和库使用。默认的 Hyperframes 引擎输出完全离线渲染的动画 HTML 页面,无需 API 密钥,也不联网。
适用场景
当你想要一份可分享、有故事性的 AI 辅助编码会话总结,而不是原始录屏时使用,例如为某个拉取请求或一次构建会话制作回顾,用于社交分享或团队展示。它是展示工具而非采集工具,需要提供事件文件作为输入。
运行要求
需要本地 Node.js 运行环境;该包通过 npm 安装或用 npx 直接运行。MCP 服务器通过 mcp 子命令以 stdio 方式启动。默认离线引擎不需要 API 密钥或网络;可选的托管视频、解说或虚拟形象服务则需要你自己的凭据。
安装前请注意
离线的 Hyperframes 层级不会让数据离开本机,但 README 描述了一条级联流程,可能使用你已有的模型服务商或你自带的视频生成密钥,并提到托管解说与虚拟形象服务;选择这些会把会话数据发送给第三方。会话事件文件可能包含仓库路径、分支名和代码片段,分享输出前请先检查输入内容。

安装

在 SourceWeft 中

  1. 打开 控制台中的 Vibemovie,将其添加到工作区。
  2. 为需要使用其工具的对话启用该服务。

Desktop only,通过 STDIO。 STDIO 服务会启动本地进程,因此需要 SourceWeft 桌面宿主。

其他 MCP 客户端

参照 仓库 中的启动说明。

README

VibeMovie

[npm version] [Downloads/month] [License: MIT]

Package · Setup · API example · Compare

VibeMovie is an open-source coding-session recap generator for developers and creators. It turns structured events into a self-contained animated HTML story through its local Hyperframes engine. An optional cinematic engine creates narrated MP4 output through WaveSpeed AI and FFmpeg. VibeMovie belongs to the Vibe Suite.

Install

bash
npm install -g vibemovie

Requires Node.js 18 or later. Includes TypeScript declarations and two direct runtime dependencies. The default Hyperframes engine needs no API key, browser automation, or FFmpeg installation.

Render your first recap

Create an input file:

bash
cat > session.json <<'JSON'[  {"kind":"task-done","ts":1769500000000,"agent":"claude-code","payload":{"label":"Refactor auth","durationMin":18}},  {"kind":"tests-pass","ts":1769503600000,"payload":{"passed":42}},  {"kind":"pr-merged","ts":1769505100000,"payload":{"pr":42,"branch":"main"}}]JSONvibemovie render session.json --title "Auth recap" --out recap.html

The CLI prints recap rendered and the output path. Open recap.html in a browser to play the story and scrub its timeline. No environment variables or credentials are required. The CLI does not load .env files. The input describes events; VibeMovie does not inspect your coding-agent history automatically.

How it works

text
JSON events -> normalize and sort -> scene compiler                                       |                 +---------------------+---------------------+                 v                                           v          Hyperframes (default)                    cinematic (explicit)       embedded HTML/CSS/JavaScript             WaveSpeed images + video + voice                 |                                           |        offline browser player                         FFmpeg -> MP4

Hyperframes rendering uses a local vibe-core runner with egress disabled. Cinematic rendering sends derived prompts, narration text, and intermediate media to WaveSpeed AI.

Why it exists

A terminal recording preserves activity. A recap selects tasks, test outcomes, diff totals, and pull-request milestones into a short story. VibeMovie compiles those supplied events with fixed templates. It does not need a language model to build the default recap. Remotion and asciinema address related video and terminal workflows with different inputs.

Compare

Checked 2026-10-14 against Remotion and asciinema CLI.

PropertyVibeMovieRemotionasciinema
InputStructured coding eventsReact video compositionsTerminal session output
ScopeTemplate-based recap storiesProgrammatic video and motion graphicsTerminal recording and streaming
Default artifact hereSelf-contained HTMLRendered compositions or embedded playerAsciicast recording
AuthoringJSON, Node.js API, agent toolReact componentsRecorder CLI

Coverage

Integration or surfaceStatusSource behavior
HyperframesImplemented; localEmbedded HTML, CSS, JavaScript, and scene data
WaveSpeed AIImplemented; live API pendingHosted generation orchestration
ByteDance Seedream v4Implemented via WaveSpeed; live API pendingHero and face reference images
FLUX Kontext MaxImplemented via WaveSpeed; live API pendingKeyframe image edits
Kling v2.5 Turbo ProImplemented via WaveSpeed; live API pendingImage-to-video clips
ElevenLabs Eleven v3Implemented via WaveSpeed; live API pendingNarration synthesis
WaveSpeed video face swapImplemented; best-effortOptional intermediate face consistency step
FFmpegImplemented; external executableNormalize, concatenate, grade, and mux cinematic output
Claude Code / OpenAI Codex / Google Gemini CLIEvent input onlySupply normalized JSON yourself; no harness-specific log reader
MCPImplementedOne stdio render tool
OpenAI Sora / Replicate—No provider adapter shipped

The source declares two engines, three ratios, three templates, six scene kinds, and five hosted model endpoints through one API service. Verification here covers source and local tests with injected cinematic dependencies. No paid API requests or credential checks were performed. An implemented endpoint is not a guarantee that a provider model remains available.

Common use cases

  1. Claude Code developers: export task and test events into an offline HTML recap.
  2. Codex maintainers: describe pull-request milestones in JSON for a release story.
  3. Gemini CLI creators: choose portrait Hyperframes output with --ratio 9:16.
  4. Node.js teams: call buildScenes and renderHyperframes inside an existing event pipeline.
  5. MCP agents: render supplied events with the single render tool.
  6. WaveSpeed users: opt into hosted cinematic generation with their own key and FFmpeg installation.

Features

  • Scenes: title, tasks, diff, terminal, merge, and end; optional scenes depend on supplied data.
  • Templates: documentary, speedrun, and meme control captions and pacing.
  • HTML ratios: 16:9, 9:16, and 1:1.
  • Player: play/pause, timeline scrubbing, captions, and embedded scene data.
  • Deterministic compilation: the same events and scene options produce the same local scenes.
  • Cinematic pipeline: generated reference images, keyframes, clips, narration, and local MP4 assembly.
  • Shot recipes: exported recipe data and deterministic selectors for custom callers; not an automatic CLI storyboard selector.
  • Suite notification: rendering emits a best-effort render-done event through vibe-core.

Runtime surfaces

text
vibemovie render [file.json|-] [--ratio 16:9|9:16|1:1]  [--template documentary|speedrun|meme] [--engine hyperframes|cinematic]  [--out path] [--title name]vibemovie mcpvibemovie helpvibemovie --help / -hvibemovie --version / -v / -V

render reads a file or stdin. Input accepts an event array or { "events": [...] }. Defaults are hyperframes, documentary, 16:9, and ./vibe-recap.html. Cinematic output defaults to ./vibe-recap.mp4.

vibemovie mcp starts the stdio Model Context Protocol (MCP) server with one tool:

ToolRequired inputOptional inputResult
renderevents arrayratio, template, engine, title, outHTML text, or a written output path

Library example

Install with npm install vibemovie. Save this as recap.mjs and run node recap.mjs:

js
import { renderMovie } from 'vibemovie';const result = await renderMovie([  { kind: 'task-done', ts: 1769500000000, payload: { label: 'Ship auth' } }], { title: 'Auth recap', out: 'recap.html' });console.log(result.engine, result.path); // hyperframes recap.html

buildScenes(events, options) and renderHyperframes(scenes, options) are also exported for custom pipelines.

Delivery and behavior limits

Hyperframes produces animated HTML, not an MP4 or GIF. It does not record your screen or stream live agent activity. The CLI has no --session or --repo importer. Convert external recordings into the event schema before rendering. The scene compiler displays at most five task rows, four diff filenames, and five terminal lines. Terminal lines summarize supplied events; they are not a captured terminal transcript. Diff totals and test results reflect input claims, not an independent Git or test-run verification. Cinematic output is normalized to 1280×720 at 24 frames per second. HTML ratio options do not change that cinematic output size. Cinematic prompts use a fixed character and setting. There is no user-avatar selection or dynamic soundtrack engine. Missing WAVESPEED_API_KEY or FFmpeg triggers HTML fallback; recognized video filename extensions become .html. Runtime provider failures do not trigger a general HTML fallback. Failed clips may be skipped; all clips failing aborts the render. Prediction polling is bounded: normally 95 polls, narration 60, and face swap 140. Requests have no explicit network timeout. The face-swap step can fall back to the raw cut. Temporary cinematic work files remain on disk. Review input before rendering: hosted narration can include task labels, session names, and summary values.

Integration guides

  • Portrait HTML recap: vibemovie render session.json --ratio 9:16 --template speedrun --out portrait.html.
  • Event pipeline: pipe a JSON array into vibemovie render --out recap.html.
  • Hosted cinematic recap: install FFmpeg, export WAVESPEED_API_KEY, then run vibemovie render session.json --engine cinematic.
  • Environment file: load variables in your shell or launcher before starting the CLI; VibeMovie does not read .env automatically.
  • Agent setup: configure your MCP host to launch vibemovie mcp over stdio.

Hosted rendering uses https://api.wavespeed.ai/api/v3. There is no separate ElevenLabs key in this pipeline.

Status

Version 0.2.4 is pre-1.0; APIs and generation behavior can change. Public case studies are not yet published. No public documentation site is configured; setup guidance is self-contained here. Source-only prototypes illustrate additional controls; they are not shipped web applications. Use GitHub issues for support; repository access may be required. For contributions, run npm ci, npm run build, npm run typecheck, npm test, and git diff --check.

FAQ

Which integrations are verified?

VibeMovie tests local rendering and cinematic orchestration with injected dependencies. Hosted endpoints have not been credential-checked in this rewrite.

Does VibeMovie provide a managed video service?

VibeMovie runs as a local package. Optional hosted generation uses your WaveSpeed API key and provider account.

How is VibeMovie different from Remotion?

VibeMovie compiles structured session events with preset scenes. Remotion provides React-based tools for custom video compositions.

How is VibeMovie different from asciinema?

VibeMovie renders a summary from supplied events. asciinema records and streams terminal output.

Can VibeMovie work offline?

VibeMovie renders Hyperframes HTML without remote assets or API keys. Its cinematic engine requires hosted API access.

Does VibeMovie export MP4 without credentials?

VibeMovie defaults to HTML when no cinematic credentials are available. MP4 generation requires a WaveSpeed key and FFmpeg.

Does VibeMovie read Claude Code or Codex logs automatically?

VibeMovie accepts JSON event input rather than scanning harness logs. Supply an event array through a file, stdin, the library, or MCP.

Can VibeMovie use my existing Sora or Replicate account?

VibeMovie has no Sora or Replicate adapter in this version. The cinematic pipeline calls fixed WaveSpeed model endpoints.

Are all output ratios available for cinematic MP4?

VibeMovie supports three ratios in Hyperframes HTML. Cinematic assembly currently normalizes video to 1280×720.

License

MIT.

来源:README.md,提交 90ff559

工具

0
SourceWeft 尚未在沙箱中启动过该软件包,因此其工具未知。

版本历史

1
  1. v0.2.3最新Sep 16, 2026

评分

已满足适用检查项的 50%

该评分衡量此收录条目的完整程度,依据注册中心条目、端点与公开来源计算得出。它不是对服务器本身的评价,也不表示服务器安全或当前可用:可连通性是带日期的检查结果,评分也不包含安装量或用户评价。

  • 近 30 天内可连通25 分

    请保持端点能够响应 MCP 握手;SourceWeft 每 30 天探测一次。

  • 已声明认证方式10 分

    请在注册中心条目中声明凭据请求头,或保持端点能够响应探测,以便判定其认证方式。

  • 已列出工具15 分

    请让服务器在无凭据时响应 tools/list,或发布可在沙箱中无凭据启动的软件包。

  • 描述必需15 分
  • 图标10 分
  • 文档15 分
  • 许可证10 分

其他 Aibenchmarkindex 集成