LeClap: On-Device Video Composition via JSON Templates

What changed

LeClap introduces a novel approach to video creation by enabling deterministic video composition through a single JSON template. This template defines all aspects of the video, including sections, filters, music, and overlays. The key innovation is its ability to render this template consistently across multiple platforms: Node.js, web browsers using WebAssembly, and natively on-device within React Native applications. Unlike generative video tools, LeClap's output is reproducible, meaning the same template will yield the same video result on a given platform. It explicitly avoids server-side processing and generative models, focusing instead on a direct composition pipeline. The project is structured as a monorepo using pnpm workspaces, with core packages like ffmpeg-video-composer (the cross-platform composition engine), @leclap/cli (for scaffolding and rendering), and @leclap/mcp (an MCP server for agent-callable tools). The latest release mentioned is ffmpeg-video-composer@2.3.0.

Key Features and Capabilities:

• Template-Driven Composition: Videos are described using a Zod-validated JSON descriptor, simplifying the creation process and eliminating the need for imperative FFmpeg commands.
• Cross-Platform Rendering: Supports Node.js, browser (via WASM), and React Native (fully on-device), ensuring consistent results across environments.
• On-Device Processing: The React Native integration allows for video capture, editing, and rendering directly on mobile devices without server interaction.
• Agent-Callable: An MCP server exposes LeClap's engine as agent-callable tools, allowing AI agents to author and render video templates.
• Deterministic Output: Guarantees that the same input template produces the same output video, crucial for reproducible workflows.
• Creative Kit: Includes a bundled library of on-device-safe templates and assets.

Why it matters for builders

LeClap empowers builders by providing a serverless and deterministic solution for video generation. Developers can integrate video composition directly into their applications, whether web or mobile, without the overhead of cloud rendering services or the uncertainty of AI-generated content. The agent-callable nature also opens up possibilities for AI-driven content creation workflows, where agents can programmatically generate video assets based on specific requirements. This significantly lowers the barrier to entry for creating dynamic video content within applications.

Practical impact

Developers can leverage LeClap to build features such as personalized video messages, dynamic marketing content, or in-app tutorials that are composed and rendered client-side. The quick start guide provides clear instructions for setting up the development environment using mise for toolchain management (Node 24, pnpm 12, FFmpeg 8.1.1, Rust) and running the web or Expo mobile applications. For those using the CLI, commands like npx @leclap/cli init my-video for scaffolding and npx @leclap/cli render template.json for rendering are available. The project's MIT license allows for free use and modification, with a note on trademark usage for the 'LeClap' name.

Caveats and source limits

The source material indicates that while the composition logic is the same, the actual rendered pixels may differ slightly across platforms due to variations in FFmpeg encoders (e.g., libx264 on Node/browser, libopenh264 on Android, h264_videotoolbox on iOS). The project is described as having 29 stars and 1 fork on GitHub, suggesting it is relatively new. Specific pricing information for any potential commercial use or advanced features is not provided. The source also mentions that while an LLM can write React code for Remotion, LeClap is designed for AI agents to author templates directly via MCP, avoiding LLMs in the final output path.