OpenComputer Hosts 'Ship Video' Agent Using Claude Opus 5.5 to Generate Explainer Videos from Code

OpenComputer features Ship Video, an open-source agent that converts URLs or text prompts into short explainer videos using Anthropic's Claude Opus 5.5. The project demonstrates a novel approach to AI video generation: instead of relying on diffusion-based video models, the agent instructs the LLM to write the video as HTML, CSS, and JavaScript code, which is then rendered deterministically.

How It Works

The system operates as a serverless agent on the OpenComputer platform, utilizing a microVM runtime (Amazon Linux 2023, arm64, 4 vCPU, 8 GB RAM). The workflow involves three core tools:

• web_fetch: Extracts page text, headings, and design elements (colors, fonts) from a provided URL.
• check_scene: Loads the generated HTML in a headless browser to verify JavaScript errors and visible text at specific timestamps.
• render_video: Uses a virtual clock to replace standard browser timing functions (like requestAnimationFrame), allowing for deterministic frame-by-frame seeking. The frames are captured via headless Chromium and encoded into an MP4 using ffmpeg (libx264, crf 18, 1920x1080 at 30 fps).

Technical Specifications

• Model Usage: Each video generation consumes approximately 90k input tokens and 15k output tokens, with the majority of the output being the generated code itself.
• Infrastructure: The agent is defined in TypeScript and deployed via opencomputer deploy. It requires no external server or queue from the developer, as OpenComputer manages the session API, model gateway, and microVM lifecycle.
• Storage: Finished videos are stored as public URLs in Vercel Blob, with upload tokens scoped to specific paths for security.

Availability

The project is open-source and available on GitHub (diggerhq/shipvideo). Users can deploy the agent to their own OpenComputer account with a single click or clone the repository to build custom agents using the provided quickstart guide.