machinepulse-ai/shotwright

Shotwright is a tool that lets a chat-based AI agent control Adobe After Effects, the well-known motion graphics program from Adobe, and produce a finished video file. The unusual part is how it runs After Effects. Instead of installing the program on the designer's own machine and steering it by hand, Shotwright runs After Effects inside a Windows container, which is a sealed, repeatable environment that always has the same versions of every piece of software. The usage flow goes like this. A designer drops in a reference video, types a chat message describing what they want, and the agent does the rest. It plans the shots, prepares the source assets, writes JSX scripts (After Effects' built-in scripting language), hands them to a tool called nexrender that renders the project without opening the UI, and finally streams the resulting mp4 back into the browser. The point is that the designer stays in charge of creative judgment while the agent handles file management, scripting, and the render queue. The project is opinionated about what it is. The README says it is not a generic AI video generator; After Effects stays at the center, and the agent layer sits on top. The loop is intent, agent, JSX, render, human review. The author argues most AI video products narrow the creative options to a few templates, while Shotwright keeps the full surface area of After Effects available. The stack mixes several technologies. A FastAPI backend and a React 18 frontend handle the chat UI. The container image comes from a multi-stage Dockerfile with shared tooling such as Node 20, Python 3.13, ffmpeg, and the Visual C++ runtime, and the After Effects 2026 installer is baked in from a separate GitHub Container Registry image. nexrender and aerender.exe do the actual rendering inside the container. A configuration file named shotwright-config.json holds shared defaults for host and container paths, image tags, and nexrender versions. The intended audience is After Effects designers who want help from an AI agent but do not want to become Windows infrastructure operators. The host requirements are a Windows 11 or Windows Server LTSC 2025 machine with at least 4 cores and 16 GB of RAM, Docker Desktop in Windows-container mode, and either a GitHub Copilot subscription or an OpenAI API key to back the agent.