← Reddit

Built a tool to export Claude Design animations as MP4

Reddit · fermatf · April 26, 2026
A developer created claude2video.com, a tool that exports Claude Design animations as lossless MP4 files by opening animation URLs in a headless browser and capturing individual frames rather than screen recording. The tool, built with Next.js, Playwright, and ffmpeg, is offered free and eliminates quality loss inherent in screen recording methods. The creator requested feedback from users encountering edge cases with the application.

Detailed Analysis

Claude Design, Anthropic's feature for generating live browser-rendered animations via React code, lacks a native video export capability — a gap that developer Tomas Pavlin has addressed by building claude2video.com, a free tool that converts Claude Design animation artifacts into lossless MP4 files. The tool operates by opening the animation's share URL in a headless Playwright browser, taking deterministic control of the animation clock, and capturing individual frames at 60fps before stitching them into a standard MP4 using FFmpeg. Built on a Next.js stack, it requires no local installation and supports two workflows: pasting a Claude Design share URL directly into the site, or swapping the `claude.ai` domain in the share link with `claude2video.com` for a faster one-step redirect.

The technical approach solves a meaningful quality problem inherent to the prevailing workaround of screen recording. Because Claude Design animations are rendered as live React components rather than pre-baked video assets, screen capture introduces compression artifacts such as color banding and dropped frames. Pavlin's tool sidesteps this by stubbing `requestAnimationFrame` and advancing the animation clock in precise increments — for instance, 16.6ms per frame — then waiting for full rendering cycles before screenshotting. This deterministic frame-capture method works across both standard React Stage components and canvas or WebGL-based animations, producing output that accurately represents the original rendered quality.

The emergence of claude2video.com reflects a broader pattern in the AI tooling ecosystem: community developers rapidly filling functional gaps left by fast-moving platform launches. Claude Design is a recently introduced feature, and its lack of export infrastructure has already prompted multiple independent solutions, including Claude Code handoff workflows (where the animation is exported as standalone HTML and then converted to video via a separate prompt), and HyperFrames, an open-source Node.js tool that uses FFmpeg to render animations locally across modular runtimes like CSS and Lottie. Each approach involves meaningfully different tradeoffs in quality, setup complexity, and portability.

This dynamic illustrates a recurring tension in generative AI product development: capabilities that allow users to create sophisticated outputs — in this case, polished motion graphics — often outpace the infrastructure needed to make those outputs practically portable. Anthropic's Claude Design enables non-engineers to produce animation-quality visuals through natural language prompting, but without export tooling, those visuals remain trapped in the browser environment. The community response, led by tools like claude2video.com, effectively extends the platform's utility into real-world production pipelines, bridging Claude's generative output with standard media formats that downstream video editing and publishing workflows expect.

The project also signals growing interest in Claude Design as a serious creative instrument, not merely a demo surface. That a developer has invested in building, hosting, and offering a free conversion service — and is actively soliciting edge-case URLs to improve robustness — suggests meaningful usage volume and creative ambition among Claude's user base. As Anthropic continues developing Claude Design, the existence of these third-party export tools creates both a template and implicit pressure for native export functionality to eventually be incorporated into the platform itself.

Read original article →