VibeFrame

AI-native video editing. CLI-first. MCP-ready.

Edit videos with natural language. Every edit is a command. Every workflow is scriptable.

Demo

CLI walkthrough → https://youtu.be/EJUUpPp2d_8

Claude Code integration → https://youtu.be/sdf930sZ7co

Why VibeFrame?

Traditional video editors are built for clicking buttons. VibeFrame is built for AI-powered workflows.

Traditional Editor	VibeFrame
Import → Drag → Trim → Export	`vibe edit silence-cut interview.mp4 -o clean.mp4`
Manual scene detection	`vibe detect scenes video.mp4`
Export for each platform	`vibe pipeline viral project.vibe.json`
Click through menus	Natural language → CLI → done

Design Principles: Headless First — AI-Native — MCP Compatible — Provider Agnostic

Quick Start (CLI)

Prerequisites: Node.js 20+, FFmpeg

CLI-first. Every video edit is a command.

# Install curl -fsSL https://vibeframe.ai/install.sh | bash Remove silence from an interview vibe edit silence-cut interview.mp4 -o clean.mp4 Add captions with auto-transcription vibe edit caption video.mp4 -o captioned.mp4 Create a TikTok from a script vibe pipeline script-to-video "A day in the life of a developer..." -a 9:16 -o ./tiktok/ Export to MP4

vibe export project.vibe.json -o output.mp4

For development:

git clone https://github.com/vericontext/vibeframe.git
cd vibeframe
pnpm install && pnpm build

Use with Claude Code

Already have the CLI installed? Claude Code runs vibe commands for you — just describe what you want in natural language.

You say	Claude Code runs
"Remove silence from interview.mp4"	`vibe edit silence-cut interview.mp4 -o clean.mp4`
"Extract 3 best moments from podcast.mp4"	`vibe pipeline highlights podcast.mp4 -c 3`
"Add Korean subtitles to video.mp4"	`vibe edit caption video.mp4 -o captioned.mp4`
"Create a TikTok from this script"	`vibe pipeline script-to-video "..." -a 9:16`
"Remove background noise"	`vibe edit noise-reduce noisy.mp4 -o clean.mp4`
"Make a 60-second highlight reel"	`vibe pipeline highlights long-video.mp4 -d 60`

No setup needed beyond installing the CLI. Claude Code discovers and runs vibe commands directly.

MCP Integration

Works with Claude Desktop and Cursor via MCP. No clone needed — just add to your config and restart:

{
  "mcpServers": {
    "vibeframe": {
      "command": "npx",
      "args": ["-y", "@vibeframe/mcp-server"]
    }
  }
}

Config file locations:

Claude Desktop (macOS): ~/Library/Application Support/Claude/claude_desktop_config.json
Claude Desktop (Windows): %APPDATA%\Claude\claude_desktop_config.json
Cursor: .cursor/mcp.json in your workspace

See packages/mcp-server/README.md for full tool, resource, and prompt reference.

AI Pipelines

End-to-end workflows powered by multiple AI providers (Claude + ElevenLabs + Gemini + Kling/Runway):

vibe pipeline script-to-video "A morning routine of a startup founder..." \ -d 60 -a 9:16 -g kling -o startup.vibe.json

vibe pipeline highlights interview.mp4 -d 90 --criteria emotional vibe pipeline auto-shorts podcast.mp4 vibe pipeline animated-caption video.mp4 -s bounce -o captioned.mp4

CLI Reference

Every command supports --help. Run vibe --help for a full list.

Group	Commands	Example
`vibe generate`	`image`, `video`, `speech`, `sound-effect`, `music`, `motion`, `storyboard`, `thumbnail`, `background`	`vibe generate image "prompt" -o img.png`
`vibe edit`	`silence-cut`, `jump-cut`, `caption`, `grade`, `reframe`, `speed-ramp`, `text-overlay`, `fade`, `noise-reduce`, `image`, `fill-gaps`	`vibe edit caption video.mp4 -o out.mp4`
`vibe analyze`	`media`, `video`, `review`, `suggest`	`vibe analyze media video.mp4 "summarize"`
`vibe audio`	`transcribe`, `voices`, `isolate`, `voice-clone`, `dub`, `duck`	`vibe audio transcribe audio.mp3`
`vibe pipeline`	`script-to-video`, `highlights`, `auto-shorts`, `regenerate-scene`, `animated-caption`	`vibe pipeline script-to-video "..." -a 9:16`
`vibe project`	`create`, `info`, `set`	`vibe project create "name"`
`vibe timeline`	`add-source`, `add-clip`, `split`, `trim`, `move`, `delete`, `list`	`vibe timeline add-source project file`
`vibe batch`	`import`, `concat`, `apply-effect`	`vibe batch import project dir/`
`vibe detect`	`scenes`, `silence`, `beats`	`vibe detect scenes video.mp4`
`vibe export`	-	`vibe export project.vibe.json -o out.mp4`

Every command supports --help, --json, --dry-run, and --stdin. Run vibe schema --list for a full machine-readable command index.

Agent Mode (Standalone)

For environments without Claude Code, Codex, or MCP — a built-in interactive session:

vibe agent                     # Start (default: OpenAI)
vibe agent -p claude           # Use Claude
vibe agent -p ollama           # Free, local, no API key

Best used for onboarding and quick experiments. For production workflows, use CLI commands directly or via Claude Code / MCP.

AI Providers

See MODELS.md for detailed model information (SSOT).

Category	Providers	Default
Agent LLM	OpenAI, Claude, Gemini, xAI, OpenRouter, Ollama	GPT-5-mini
Image	Gemini, OpenAI, xAI Grok	Gemini Nano Banana
Video	xAI Grok, Kling, Runway, Veo	Grok Imagine
Audio	ElevenLabs, Whisper	-

Required API Keys: OPENAI_API_KEY, ANTHROPIC_API_KEY, GOOGLE_API_KEY, ELEVENLABS_API_KEY, RUNWAY_API_SECRET, KLING_API_KEY, XAI_API_KEY

Project Structure

vibeframe/
├── packages/
│   ├── cli/               # CLI + Agent mode
│   ├── core/              # Timeline engine (Zustand + Immer + FFmpeg)
│   ├── ai-providers/      # Pluggable AI providers
│   ├── mcp-server/        # MCP server (npm: @vibeframe/mcp-server)
│   └── ui/                # Shared React components
├── apps/web/              # Next.js landing & preview UI
├── MODELS.md              # AI models reference (SSOT)
└── ROADMAP.md             # Development roadmap

Roadmap

Phase	Status	Description
1. Foundation	Done	Core CLI, FFmpeg.wasm export
2. AI Providers	Done	Multi-provider integrated
3. MCP Integration	Done	Claude Desktop & Cursor support
4. AI Pipelines	Done	Script-to-Video, Highlights, B-Roll, Viral
5. Server Infrastructure	Planned	Hybrid rendering, chunked uploads
6. Collaboration	Planned	CRDT-based local-first sync

See ROADMAP.md for details.

Open Core Model

VibeFrame Core is 100% open source (MIT License). Core features will always remain free and open source.

Contributing

pnpm build     # Build all packages
pnpm test      # Run all tests
pnpm lint      # Lint code

Contributions welcome — AI provider integrations, CLI improvements, docs, bug fixes & tests. See CONTRIBUTING.md.

Contributors

License

MIT - see LICENSE

Built for the AI age. Ship videos, not clicks.

NAME

SYNOPSIS

INFO

DESCRIPTION

README