AI-first video editor + generator in one repo. Tell Claude what you want, get an MP4 back. Open source, fully local, no accounts, no API keys for the renderer.
by ZELDOgiq & Media AI AT Status: v3.0.0-alpha.1 (foundation) · v2.0.0 generator-only release is still live and supported
aivc-dach/
├── editor/ # AI-driven video editor — fork of OpenCut (Next.js + Bun)
├── generator/ # HTML/CSS templates → MP4 (Puppeteer + ffmpeg) — formerly v2.0.0
├── mcp-server/ # Model Context Protocol stdio server, exposes editor + generator tools to AI agents
└── shared/ # brand config + TypeScript types shared across modules
A single tool that lets non-technical creators:
- Generate ready-to-use intros, outros, shorts and sponsor reads from a one-line prompt (the v2.0.0 workflow).
- Edit raw footage with AI assistance — the editor's chat sidebar can cut, trim, add clips, drop a generated intro on the timeline, all via the MCP server.
- Mix providers: Claude (default), OpenAI or local Ollama via the Vercel AI SDK. BYOLLM.
This v3.0.0-alpha.1 release is the foundation: branch + OpenCut fork + generator move + MCP skeleton. The actual AI integration (chat sidebar wired to MCP, drag-and-drop templates, bidirectional state sync) lands in subsequent v3.0.0-alpha.x and the v3.0.0 stable release.
| Module | What works today | What's coming |
|---|---|---|
editor/ |
OpenCut fork builds & runs (bun run dev:web → http://localhost:3000), AIVC DACH branding (title, primary/accent colors, dark default) |
Full branding pass, AI chat sidebar, MCP wiring, template drag-and-drop |
generator/ |
All v2.0.0 features (6 templates, renderer, install scripts, smoke test) work unchanged at the new path | None planned — feature-complete for now |
mcp-server/ |
stdio server, ping tool live, 4 stub tools defined for editor + generator |
Real implementations: editor.cut, editor.trim, editor.addClip, generator.render |
shared/ |
brand config + TypeScript types | Templates manifest (Phase 5) |
git clone https://github.com/ZeldoGiQ/aivc-dach
cd aivc-dach
generator/scripts/install.sh # macOS / Linux
# or:
generator\scripts\install.bat # WindowsThen in Claude Code:
Make a news intro about Gemini 4 with subtitle "Google's new AI model"
Output: output/news-intro-<timestamp>.mp4.
cd editor
bun install
cp apps/web/.env.example apps/web/.env.local # dummy values are fine for local UI work
bun run dev:web # opens http://localhost:3000Note: the editor inherits OpenCut's auth, DB and CMS dependencies. The dummy values in
.env.localare enough to load the page; cloud calls fail gracefully. Real production secrets land in v3.0.0.
cd mcp-server
bun install
bun run smoke # initialize → tools/list → ping → expects "pong"To wire into Claude Code, add to ~/.claude/config.json:
{
"mcpServers": {
"aivc-dach": {
"command": "bun",
"args": ["run", "/abs/path/to/aivc-dach/mcp-server/server.ts"]
}
}
}If you're running v2.0.0 today: nothing breaks immediately. v2.0.0 lives at the v2.0.0 tag and on the main branch until v3.0.0 stable replaces it.
What changes in v3.0.0:
- Generator moved from repo root to
generator/. The renderer command is nownode generator/renderer/render.js(instead ofnode renderer/render.js). brand.config.example.jsonmoved from repo root toshared/.- New
editor/andmcp-server/modules. - Top-level scripts orchestrate all modules:
bun run dev:editor,bun run smoke,bun run render. - Brand config path (
~/.aivc-dach/brand.config.json) is unchanged. No user-data migration needed.
bun run smoke # generator (Python) + mcp-server (Bun)
bun run smoke:generator # just the v2.0.0 templates / renderer check
bun run smoke:mcp # just the MCP stdio danceBoth are green for v3.0.0-alpha.1. The editor doesn't have a CI smoke yet — that's part of v3.0.0.
| Version | Phase | Scope |
|---|---|---|
| v3.0.0-alpha.1 | Foundation (this PR) | Branch + OpenCut fork + generator move + MCP skeleton |
| v3.0.0-alpha.2 | AI layer | Vercel AI SDK + Anthropic / OpenAI / Ollama providers, chat sidebar shell |
| v3.0.0-alpha.3 | MCP wiring | editor.cut, editor.trim, editor.addClip connect to OpenCut's Zustand store |
| v3.0.0-alpha.4 | Templates → editor | Drag-and-drop generator templates onto the timeline, live render preview |
| v3.0.0-alpha.5 | State sync | Bidirectional: chat ↔ timeline live updates |
| v3.0.0 | Stable | Full branding pass, real production env, end-to-end smoke (footage → cut → drop intro → export) |
| v3.1.0 | Polish | Native AIVC logo, custom Inter weights, tutorial videos |
| Command | What it does |
|---|---|
AIVC help |
Show all commands |
AIVC reset |
Reset the brand config (generator/scripts/reset.*) |
AIVC update |
git pull + reinstall renderer / editor / mcp-server deps |
Set up brand |
Run the (optional) brand wizard |
Make a <format> |
Render a generator template (works today) |
Open editor |
Start the editor dev-server (works today, no AI yet) |
| Tool | Used for | Required? |
|---|---|---|
| Claude Code | Skill integration & "one prompt" workflow | ✅ |
| Node.js (≥ 18) | Generator renderer (Puppeteer + ffmpeg) | ✅ |
| Bun (≥ 1.2) | Editor (OpenCut fork) + MCP server | ✅ |
| Git | Repo cloning + updates | ✅ |
| ffmpeg | Frame encoding (auto-fallback to ffmpeg-static) |
optional |
| Python 3.11+ | Faster Whisper subtitles (planned) | optional |
Supported systems: Windows 10/11, macOS, Linux.
See TROUBLESHOOTING.md for the 7 most common issues (Chromium download, ffmpeg, permissions, …). Editor-specific issues (env vars, OpenCut deps) will be added in v3.0.0.
MIT — see LICENSE. Third-party attributions in NOTICE.md.
The editor under editor/ is a fork of OpenCut (also MIT). OpenCut's full license text is preserved at editor/LICENSE.
- Owned and maintained by ZELDOgiq & Media AI AT
- Editor foundation: OpenCut (MIT)
- Generator: Puppeteer + ffmpeg, six templates inspired by Hyperframes-style workflows
- MCP server: Built on @modelcontextprotocol/sdk
- AI layer (planned): Vercel AI SDK — Anthropic / OpenAI / Ollama providers
- Community: Vibe Coding DACH Skool (German-speaking creators — courses, premium templates, live workflow reviews)