FavioVazquez/showtime

What it is

showtime is a tool that lets you describe a video in plain language to your AI coding assistant, and the assistant produces a finished video entirely on your own machine — no cloud AI, no API keys, no uploads. Think of it as a "video studio" that your agent can direct.

What it does

The project provides a command-line tool and a set of 'skills' that plug into coding agents such as Claude Code, Codex, Cursor, Devin, OpenCode, and others that support the Agent Skills standard. Once installed, you simply ask, for example:

"Make a 20-second launch video for this repo, with a voice-over and upbeat music."

Your agent will:

• State any assumptions it made (duration, format, tone)
• Show you a preview (stills or a fast draft)
• Render the final video on your hardware — including motion graphics, voice-over, music, sound design, captions, and footage editing
• Export it to common formats and platforms

Every request gets its own folder with final.mp4, a poster image, a shareable text file, exports, and working files. Nothing is overwritten.

Key features

• Local-first: All processing runs on your computer. No cloud AI services, no API keys, no uploads.
• X-ray: Film a real object (e.g., a typewriter) and showtime will overlay its internal mechanism (drawn from a library) perfectly synced to the audio, even producing music notation from the sound.
• Living diagrams: Create animated, interactive diagrams that explain systems over time — users can hover, scrub, and interact.
• Recap & repro: Automatically turn a coding-agent session or a failed playright test into a short narrated video.
• check command: Reads text drawn on a phone screen, catching issues like too-small or overlapping text.

How it works

showtime is delivered in three ways:

1. Agent Skill – a set of instructions and utilities that any Agent-Skills–compatible agent can load.
2. Command line tool – the showtime binary that orchestrates rendering, QA checks, and more.
3. MCP server – optional, for agents that support it, to expose typed tools.

The setup downloads a small runtime (about 0.5–0.75 GB on first install, 1.4–1.8 GB in `/.showtime`). It works on Intel Macs, Apple Silicon, Linux (x64 and ARM64), and Windows (x64 and ARM).

Compatibility

• Tested with: Claude Code, OpenAI Codex, Cursor, Devin, OpenCode.
• Should work with (not yet fully verified): GitHub Copilot, Gemini CLI, Antigravity, Cline, Kilo Code, Kiro, Zed, Goose, Amp, Factory Droid, Qwen Code.

Installation is done via a plugin, a skill-add command, or a manual setup script depending on the agent. The README includes exact commands for each.

Requirements

• Python package manager uv
• Node.js 24 or 22 LTS (20+ works)
• The agent you use must support Agent Skills and be able to run shell commands.

Current state

• Version 0.4.6 (early, MIT-licensed).
• The project includes a thorough README with quick-start guides, troubleshooting, FAQ, and a demo reel (55 seconds, rendered on a 64-core CPU in 2 minutes).

Notable extras

• The 'crew' of ten sub-agents can speed up some hosts (subject to the agent's capabilities).
• Every finished video passes a built-in QA check before delivery.
• A behavior scoreboard shows that agents using showtime follow instructions more reliably than those without it (35/42 vs 25/42 correct actions in a test).

This is a tool for anyone who wants to generate polished, narrated videos from simple text prompts using their existing coding agent, with full privacy and no per‑use costs.