A media-generation CLI built to be driven by your coding agent

Almost every project I build needs generated assets at some point — an icon, a hero image, the occasional voiceover or short clip. Getting them has always meant the same tax: SDK wiring, polling loops, glue code, or walking my coding agent through an API it half-remembers from training data that's months stale. So I built ploof to make that tax disappear.

It's a command-line tool that turns a prompt into a file — image, video, or audio. The twist is who it's built for: not me, my coding agent. I tell Claude Code what I want, and it installs ploof, runs ploof learn (which prints a reference matched to the installed version, so the agent works from what the tool actually does today, not a guess), authenticates, and the finished file lands on disk. No SDK, no polling, no glue.

One command line covers every modality and provider. Images, video, and audio — generate, edit, extend, transcribe, translate — across OpenAI today and fal.ai's entire model marketplace. You can declare a whole batch of assets in a YAML manifest, wire one task's output into the next, and run them in parallel; every file ships with a JSON sidecar recording the exact prompt and parameters, so any asset is reproducible later.

The cloud mascot up there? That icon was generated by ploof. It's open source (MIT) — on GitHub and npm.