Wedding videos as code: templates your coding agent can render
An open-source kit of editable video templates plus an Agent Skill, so Claude Code or Codex can turn a folder of photos into a beat-synced wedding MV.

What it does
vvv ships a library of editable video templates (React/TypeScript under the hood) and a skill that teaches coding agents to use them. You tell the agent the style, aspect ratio, and song; it fetches your photos, groups them by outfit and mood, aligns the music’s chorus to the emotional peak, and renders a 1080p film locally. Nine templates ship in the box, from a hand-drawn “Doodle” to a black-screen “Cinema” opener, each in 16:9 and 9:16.
The interesting bit
The core idea is separation of concerns for video: scene structure, layouts, beat-timed timelines, and render checks are built once per template; each new film only swaps photos, names, dates, and copy. vvv.py describe tells the agent which inputs a template exposes, and validate catches bad input before rendering — so the agent edits declared knobs instead of regenerating video code from scratch every time. There’s also a reference-video mode: give it a Douyin/Xiaohongshu link and it reverse-engineers the structure (dense frame grabs, per-page shot lists) and rebuilds it with your own assets.
Key highlights
- Works with 25+ agents via Agent Skills, plus an MCP server for browsing templates without installing anything
vvv checkmeasures aspect ratio, loudness, static frames, and beat alignment, flagging the worst offenders- Projects are portable: a
.vvvfile bundles code, assets, and render-verification records so another agent can pick up where the last left off - Photos are analyzed per-image (people, emotion, outfit) and sequenced along each template’s emotional arc
- Local FFmpeg tooling for media probing, trimming, audio extraction, and frame sampling
Caveats
- Rendering is local and needs Python 3.9+, Node.js 26, and npm — a fairly specific toolchain
- The token-saving claim for template reuse is explicitly unbenchmarked; the comparison baseline is still on the roadmap
- Recreating reference videos works best with agents that can generate images; it does not generate video or voiceover, and the project itself is study/research-only with clear warnings about not redistributing others’ work
Verdict
Worth a look if you’re curious about agent skills applied to a concrete, non-code domain — the template-as-contract design is genuinely instructive. If you wanted a one-click consumer video app, this is decidedly not that; it’s a developer’s toolkit wearing a wedding suit.
Frequently asked
- What is actionow-ai/vvv?
- An open-source kit of editable video templates plus an Agent Skill, so Claude Code or Codex can turn a folder of photos into a beat-synced wedding MV.
- Is vvv open source?
- Yes — actionow-ai/vvv is open source, released under the MIT license.
- What language is vvv written in?
- actionow-ai/vvv is primarily written in Python.
- How popular is vvv?
- actionow-ai/vvv has 507 stars on GitHub.
- Where can I find vvv?
- actionow-ai/vvv is on GitHub at https://github.com/actionow-ai/vvv.