Glonce

launchvideo.io has Claude Opus 5.5 write launch videos as code

toolsThe Glonce editorsHacker News🔥75Glonce rating75Impact55Novelty65Relevance95Surprise55Players70Credibility85

launchvideo.io is a demo product that turns a pasted URL or a short product description into a finished MP4 launch video with no human in the loop. Claude Opus 5.5, accessed through OpenComputer's model gateway as anthropic/claude-opus-5.5, writes the film as HTML, CSS and Web Animations code rather than generating pixels directly; a serverless agent then renders that code into video. The site says the whole process takes about four minutes and roughly 100,000 tokens per video, split into roughly 90,000 input and 15,000 output tokens, most of the input being the HTML itself. There is no video model involved: the page's clocks, including requestAnimationFrame, timers, Date, and CSS and Web Animations, are replaced with a virtual clock, so every frame is a deterministic seek rather than a live playback capture. Each job runs in a fresh microVM on Amazon Linux 2023 (arm64) with 4 vCPU and 8 GB RAM running Node 22; the first tool call installs Playwright's headless Chromium and a static ffmpeg, which takes about a minute, and the VM is discarded afterward. The agent has three tools: web_fetch pulls page text, headings, dominant hex colors and Google Fonts from the target site; check_scene loads the generated film and reports JavaScript errors plus the visible text at sample timestamps; render_video renders the final clip and uploads it. Output is 1920x1080 at 30 fps, with JPEG frames piped into libx264 at crf 18. For storage, the agent itself holds no secrets: the form mints a Vercel Blob upload token scoped to a single path and valid for three hours, and the finished MP4 becomes a public Blob URL. The examples shown on the page are described as single, untouched runs: a URL or prompt in, an MP4 out, no edits. The whole product is built as one OpenComputer serverless agent file, three tools and the web form; OpenComputer runs the agent, the rendering microVM, the model gateway and the session API the page polls for progress. The code is open in the repo diggerhq/shipvideo, and the site offers a one-click deploy plus a quickstart it describes as a TypeScript file, a deploy and a session, doable in about ten minutes via npx opencomputer template deploy. The idea is credited to a post by Deedy on Opus 5.5 and instructional video, summarized on the site as: the model writes the film as code, and code renders the same every time.

Key facts

  • launchvideo.io generates an MP4 launch video from a URL or product description using Claude Opus 5.5 (anthropic/claude-opus-5.5) and a serverless OpenComputer agent, with no human edits.
  • One video takes about four minutes and roughly 100,000 tokens, split into roughly 90,000 input and 15,000 output tokens.
  • There is no video model: the model writes the film as HTML/CSS/Web Animations code, and a virtual clock replaces the page's real clocks so every frame is a deterministic seek, rendered at 1920x1080, 30 fps, via libx264 at crf 18.
  • Each job runs in a fresh microVM (Amazon Linux 2023, arm64, 4 vCPU, 8 GB RAM, Node 22) using three tools: web_fetch, check_scene and render_video; the agent holds no secrets, relying on a Vercel Blob upload token scoped to one path for three hours.
  • The agent and renderer are open source at diggerhq/shipvideo, with a one-click deploy and a quickstart the site says takes about ten minutes; the idea is credited to a post by Deedy on Opus 5.5 and instructional video.

Why it matters

The project is a concrete example of an LLM coordinating a multi-stage creative pipeline by writing deterministic code rather than generating video frames directly. Because the output is HTML/CSS animation rendered through a virtual clock instead of a video model, the same input reliably produces the same result, which is the core claim the site borrows from Deedy's original post: code renders the same every time.

Who it affects

It is aimed at developers and small teams who want a quick launch or explainer video for a product without hiring an editor, and at people building agents on OpenComputer who want a worked example combining a model, tools and a rendering microVM.

How to use it

A user pastes a URL or types a product description into the form on launchvideo.io and gets an MP4 back after about four minutes. The underlying agent can be deployed to an OpenComputer account in one click, or cloned from the diggerhq/shipvideo repository; the site describes a quickstart of a TypeScript file, a deploy and a session, doable via npx opencomputer template deploy in about ten minutes. No pricing for using the hosted tool is stated on the page.

How solid is it

The account comes directly from the product's own site, which lays out specific technical details: token counts, microVM specs, render settings and the Vercel Blob storage scheme. The example videos are described as untouched single runs. No independent outlet has verified the claims, and the page does not name a company or individual behind launchvideo.io beyond the linked repository.

Risks and caveats

The process is fully automated with no human edits, so any error the agent's own check_scene tool misses would ship straight through. No claim is made about video length or content variety beyond the showcased examples, and no publication date is given for the page or for Deedy's original post.

“No video model. The page's clocks (requestAnimationFrame, timers, Date, CSS and Web Animations) are replaced with a virtual clock, so every frame is a deterministic seek.”

— launchvideo.io