actionow-ai/vvv

The problem

Asking a coding agent to make a short video from scratch usually means it rewrites a Remotion or FFmpeg pipeline per project, re-debugging layouts, timing and rendering every time. Vibe Vision Video (vvv) ships editable Remotion templates plus an Agent Skill so the agent fills in photos, copy and music against a fixed, described structure instead of reinventing the composition each run.

What's different

Compared with raw Remotion, where every project starts from a blank composition, vvv separates what is built once (scenes, 16:9 and 9:16 layouts, beat-planned timing) from what changes per film (media, text, music). Compared with generative video tools, it produces an editable project you can keep and hand off rather than a rendered file you cannot modify. The parts that genuinely stand out are the machine-readable template interface (vvv describe / validate) and vvv check, which measures format, audio, still stretches and beat alignment, making it clearer what the agent cannot verify. It also supports rebuilding a reference film as your own project.

Who it's for

Developers already using Claude Code, Codex, Cursor or similar agents who need repeatable short-form video — wedding reels, social clips, product promos — from a known template. It suits people comfortable with Python and Node tooling and willing to review samples. It is not for professional editors needing frame-level timeline control, nor for teams wanting a no-code web editor; the README does not claim generated video or voice.

Cost of adoption

You need Python 3.9+, Node.js 26 and npm locally; rendering runs on your machine. Install via the Claude Code plugin marketplace, npx skills add actionow-ai/vvv, or an MCP endpoint, with per-agent details in the docs. The project is at 0.1, verified only on macOS ARM64 with Claude Code and Codex; other agents and platforms are less tested. Code and templates are MIT, but fonts, soundtracks and third-party code carry their own licenses. Reuse savings are asserted, not benchmarked — the README says no comparison is published.

Verdict

The idea — one template contract shared across many coding agents — is sensible and the check/handoff tooling is more thought-through than a demo. But at two days old and 333 stars, this is a brand-new project riding an early spike, not something with a track record; the README is unusually thick, yet platform coverage is one machine and one agent pair. Treat it as an experiment: useful if you want agent-driven video with editable output, and can absorb breaking changes. Wait if you need stability or broad platform support.