A new service called Ship Video demonstrates Claude Opus 5.5’s capability to automatically generate explainer videos from product URLs or text descriptions.
According to the site, the tool takes a URL or product description as input and uses Opus 5.5 to write a film script, which a serverless agent then renders into a finished MP4 video. Each video takes roughly four minutes to produce and uses approximately 100,000 tokens per generation.
The system operates through a serverless architecture on OpenComputer. When a job is submitted, it runs in a fresh microVM on Amazon Linux 2023 with 4 vCPU and 8 GB RAM. The agent uses three main tools: web_fetch (which extracts page text, titles, headings, colors, and fonts), check_scene (which loads the film and reports errors and visible text), and render_video (which handles final rendering and storage).
Rendering occurs without a dedicated video model. Instead, the system replaces standard browser timing functions with a virtual clock that enables deterministic frame seeking. Videos are rendered at 1920x1080 resolution, 30 fps, with JPEG frames piped into libx264 compression.
The entire product consists of one OpenComputer agent defined in TypeScript, three tool functions, and a web form. According to the documentation, Opus 5.5 typically consumes roughly 90,000 input tokens and 15,000 output tokens per film, with most token usage from the HTML itself.
Storage is handled through Vercel Blob, with upload tokens scoped to individual jobs for three-hour windows. The finished MP4 becomes a public URL. The control plane uses the same API as OpenComputer’s CLI, allowing the web interface to create sessions, send turns, and poll event streams to display progress.
The project is open-source and available on GitHub at diggerhq/shipvideo. The site notes that the concept builds on earlier work by Deedy exploring how Opus 5.5 can write films as code that render deterministically.
Key facts
- Each video takes about four minutes to generate and uses roughly 100,000 tokens
- The system runs in fresh microVMs with 4 vCPU and 8 GB RAM, rendered at 1920x1080 and 30 fps
- Opus 5.5 uses approximately 90,000 input tokens and 15,000 output tokens per video
- No manual editing is required; videos are generated in a single run from URL or text prompt
- The project is open-source and deployable through OpenComputer with one click