FableCut is a browser-based video editor that AI agents can control, with zero external dependencies. It is designed to be driven programmatically by AI systems, enabling automated video editing workflows. The project is open-source on GitHub.
SynthePulse Insight · AI deep reading
Version 1 · 1 source
Traditional AI video tools hide the editing process behind APIs, but FableCut does the opposite—it exposes the entire timeline as a JSON document, enabling real-time collaborative editing between humans and AI agents.
Most AI video tools encapsulate the editing process behind APIs, leaving users to generate results through black-box calls. FableCut overturns this model: its core is the `project.json` file, which fully describes media, clips, tracks, effects, and keyframes—it is both the project storage format and the interface for AI interaction. Any program capable of generating or modifying JSON, whether an AI agent or a manual script, can drive the editor.
This design brings two key advantages: transparency—every editing step is a readable, auditable JSON data structure; and collaboration—humans modify the timeline through the browser UI while AI sends patch operations via MCP or REST, both working on the same project simultaneously, with the UI hot-reloading via server-sent events in about 150ms.
FableCut offers three equivalent control surfaces, optimized for AI agent usage scenarios. MCP (Model Context Protocol) is the preferred method: after registering the MCP server once, Claude Code or Claude Desktop can directly call tools like `fablecut_patch_project` for incremental modifications, avoiding the need to transmit the entire document. The toolkit also includes a compact project summary (one clip per line) and on-demand document fragments, designed to save tokens.
For environments without MCP, you can directly read/write the `project.json` file (UI auto-reloads) or interact via REST API (GET/PUT /api/project, POST /api/upload, etc.). The `CLAUDE.md` file contains complete patterns, semantics, and recipes, guiding any code-capable large model to operate the editor.
Despite being AI-driven, FableCut offers professional features comparable to traditional editors. It supports 4 video tracks and 3 audio tracks, including drag-and-drop trimming, splitting, snapping, undo/redo, and other basic operations. Effects include 12 one-click filter presets (cinematic, teal/orange, black-and-white, etc.), 17 transitions (fade, slide, zoom, glitch, etc.), keyframe animations (curve speed control for about 25 properties), speed ramps, and chroma keying.
AI integration is a highlight: it has built-in browser-based person background removal using MediaPipe, no cloud processing needed. The reference video analysis feature (`analyze.js`) extracts shot boundaries, music beats (BPM), loudness curves, and per-shot energy levels from any video, generating an editing blueprint—it can even extract the reference video's music track, helping users replicate the same rhythm with their own footage.
Export supports two modes: primarily uses ffmpeg encoding for CRF-18 MP4 (frame-accurate), falling back to real-time MediaRecorder. Export runs in the background and continues even if you switch tabs.
FableCut's most unique value lies in opening up the possibility of human-machine collaborative editing. In traditional workflows, after AI generates a video, humans cannot fine-tune details; but in FableCut, AI can edit any part of the timeline, and humans can intervene at any time through the intuitive UI. For example, a user can ask Claude Code to 'edit my footage according to this reference video's style' and then manually fine-tune transitions and subtitles.
The project's zero-dependency nature lowers the barrier to entry: only Node.js 18+ and a Chromium browser are required, no npm install needed. The server binds to localhost by default, but can be exposed on the local network via environment variables. Although still in early stages (410 stars on GitHub, 38 commits), its design philosophy and implementation quality show potential to replace traditional AI video tools.
This article is based on the README of the open-source project FableCut (released July 2026). Features, control protocols, and performance metrics come from developer claims; actual usage may be affected by environmental factors. The reliability of AI-related features (e.g., MCP-driven editing) depends on the underlying model's capabilities and has not been independently evaluated by third parties.
By unifying the project file with the interface, FableCut pioneers a new paradigm of real-time collaborative video editing between humans and AI agents, offering a more open and controllable alternative for AI content creation tools.
Primary report
Primary source