Image: GitHub
Why it mattersA short family or wedding film has needed editing software. Vvv moves that job into a conversation with a coding agent. The agent picks the template and places the photos.
Making a short film from a pile of photos has needed editing software. It has needed a learning curve and a few hours. Vvv is a new open-source skill that moves that job into a chat with a coding agent. The repository gained 235 stars one day after launch. It is MIT-licensed.
What vvv does
The author describes vvv as six built-in video templates plus an agent skill. The skill drives the templates. Three are landscape: a hand-drawn Doodle, a clean Album and a camcorder-style DV Story. The other three are a comic opener, a beat-cut Starter reel and a vertical 9:16 Album. Each template is a Remotion project. The output is React and TypeScript code. The agent can edit the code when the brief requires it.
The user types one sentence. For example, "use the vvv skill to make a 9:16 wedding short from the photos in ./photos". The agent picks a template. It places the photos, writes the copy and renders a sample. It then runs the project's own vvv check on the result. The check measures the format, the audio levels and the still-frame stretches. It also checks the beat alignment. It then lists what only a human viewer can judge. The user watches the sample and says what to change.
How it installs
Vvv is a Claude Code plugin. It is an Agent Skill for Codex, Cursor, Gemini CLI, Copilot and more than 20 other agents. That version installs with npx. There is a WorkBuddy skill too. The author also runs an MCP server at mcp.vibevision.video for any MCP client. Rendering runs on the local machine. It needs Python 3.9 or newer, Node.js 26 and npm. The agent installs the rest.
A second feature works from a video the user already likes. The user gives the agent the file plus photos, a logo, copy and music. The agent studies the reference and rebuilds it as an editable project with the user's own material. The author says the reference is studied, never released. Video and voice are not generated.
What the author claims and does not
The project is version 0.1. The author labels this an early release. The author says reuse of a template is meant to save tokens. Those tokens would otherwise go into redesigning and debugging the same video code. The author also states that no comparative benchmark is published yet. The 235-star count is GitHub's own figure.
A small team that already uses a coding agent for day-to-day work can read vvv as one case study. The agent skill reshapes a job that lived in a different tool. The template is built once, with its scenes, layouts, timing and checks. Every new film changes only the photos, the names, the copy and the music. Whether the output is good enough to send to a client is still a human judgement. The author does not pretend otherwise. The last job of the vvv check command is to list what only a viewer can decide.
Source
Primary source: actionow-ai/vvv on GitHub.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.