Kiln Uses Claude Code to Build Launch Video, Open-Sources Videowright for AI-Generated Video Creation
Key Takeaways
- ▸Claude Code excels at web-based video production because animated explainer videos rely heavily on HTML, CSS, SVG, and timing—all well-represented in training data
- ▸Audio-first workflow is critical: speech-to-text timestamps drive video segment timing, preventing the expensive re-cuts that plagued the v1 iteration
- ▸Modular code structure (one web component per video beat) is essential for effective AI collaboration; monolithic files cause LLM confusion during edits
Summary
Kiln used Anthropic's Claude Code to produce its v1 launch video in just 2 days, validating Claude's capabilities for creative work. The experience was so successful that the team built and open-sourced videowright, a framework that generates videos from prompts using AI agents. The tool supports any web-based animation technology (SVG, Lottie, React, Svelte) and includes AI-generated voice-overs, sound effects, music, automatic lip-sync via speech-to-text word timestamps, and pixel-perfect MP4 export via headless Playwright. Videowright is available as an MIT-licensed npm package, democratizing video creation by eliminating the need for expensive video editors or motion graphics software.
- Real app UI can be animated live by letting Claude read application code, enabling screen recording in 'storytelling mode' without mock-ups
- Videowright's open-source release signals a new category of AI-native video tools, potentially disrupting traditional motion graphics workflows
Editorial Opinion
Videowright represents a meaningful expansion of Claude Code's utility beyond software engineering into creative domains. By demonstrating that AI can handle the full video pipeline—from concept to export—Kiln has opened a door for smaller teams to compete on production quality previously reserved for well-funded studios. The emphasis on modular code structure and audio-first workflows shows that successful AI-assisted creative work isn't just about capability, but about designing the collaboration itself for clarity and iteration.
