Opus 5.5 is good at explainer videos

原始链接: https://launchvideo.io

该项目通过将整部电影视为代码,实现了人工智能驱动的视频生成。系统由 **OpenComputer** 提供支持,使用单个 TypeScript 智能体来调度整个工作流程,无需定制服务器、队列或框架。 **工作原理:** * **智能:** 智能体使用 Claude-Opus 5.5 将电影编写为确定性的 HTML/JS 代码。 * **基础设施:** 每个任务都会触发一个全新的临时微型虚拟机(Node 22),并安装 Playwright 和 FFmpeg。 * **渲染:** 系统不使用视频生成模型,而是通过虚拟时钟来替代实时浏览器计时器。这实现了 1080p/30fps 的确定性逐帧渲染,并将其转换为 MP4 格式。 * **工具:** 三个核心工具分别负责网络数据抓取、JS/视觉错误检查以及视频渲染/上传。 * **安全:** 系统使用作用域受限的 Vercel Blob 令牌,确保智能体保持无状态且不涉及密钥。 整个技术栈均为开源,只需一键即可部署到您自己的账户中。通过利用基于代码的渲染,该系统确保了模型在每次“拍摄”场景时都能获得一致且高质量的结果。

最近在 Hacker News 上的一场讨论突显了 Claude 3.5 Opus 在编排技术工作流方面的惊人能力,特别是在制作“解说视频”方面。用户指出,Opus 并非依赖生成式视频模型,而是通过编写和执行代码(通常是 HTML/JS)来渲染视觉效果,再将其捕获为视频文件。许多参与者成功利用这种方法,在几分钟而非几小时内制作出了专业且极具传播力的内容。 然而,讨论很快转向了这些能力带来的更深层影响: * **“SaaS 外壳”困境**:开发者对 SaaS 的未来表示担忧。如果用户可以使用大语言模型随需“通过直觉编码”出定制工具或解决方案,那么简单软件产品的价值主张正日益受到质疑。 * **关于“垃圾内容”(Slop)的争论**:批评者认为这些 AI 生成的视频是缺乏人类意图和深度的低成本“垃圾”;另一方则反驳称,这仅仅是数字内容创作的演变,高水准制作的准入门槛已被大幅降低。 * **社区的未来**:参与者质疑在 AI 智能体遍布的世界中,以人为本的平台能否生存。有人提出,未来的价值在于人类的观点、社区连接以及 AI 无法复制的真实“现实世界”体验。
相关文章

原文

The whole product is one agent file, three tools, and this form. OpenComputer runs the agent, the microVM it renders in, the model gateway, and the session API the page polls.

Agent
One OpenComputer serverless agent, defined in TypeScript and deployed with opencomputer deploy. No framework, no queue, no server of ours.

Model
anthropic/claude-opus-5.5 through OpenComputer's model gateway. Roughly 90k input and 15k output tokens per film, most of it the HTML itself.

Runtime
Every job is one session in a fresh microVM: Amazon Linux 2023 on arm64, 4 vCPU, 8 GB RAM, Node 22. The first tool call installs Playwright's headless Chromium and a static ffmpeg (about a minute); the VM is thrown away after.

Tools
Three defineTool functions. web_fetch returns page text plus title, headings, the most used hex colors, and Google Fonts. check_scene loads the film and reports JS errors and the visible text at sample timestamps. render_video renders and uploads.

Rendering
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. 1920x1080 at 30 fps, JPEG frames piped into libx264, crf 18.

Storage
The agent holds no secrets. The form mints a Vercel Blob upload token scoped to one path for three hours, parks it in a per-job manifest, and the tool fetches it by job id. The finished MP4 is a public Blob URL.

Control plane
This page uses the same API the CLI does: create a session, send one turn, poll the event stream (tool.started, tool.completed, turn.completed) to show progress, and treat the MP4 appearing in Blob as done.

Everything above is in the repo, and one click deploys it to your account. The idea comes from Deedy's post on Opus 5.5 and instructional video: the model writes the film as code, and code renders the same every time.

联系我们 contact @ memedata.com