prompt2video

Promptable, AI-automated video generation. Give it a prompt describing what you want, and it drives Claude Code to build a Remotion project and render a short narrated video — MP4 + per-scene voice-over + a multi-page slides PDF.

You describe the topic; Claude writes the scenes, the narration, wires up the composition, and renders. You get the artifacts.

How it works

prompt2video build does the orchestration:

1. Scaffolds a blank Remotion project (npx create-video)
2. Installs the Remotion claude-code skill (npx skills add remotion-dev/skills)
3. Copies the bundled prompt2video skill into the project
4. Wraps your stdin as the video topic in an explicit "generate a narrated video" instruction (so the prompt2video skill always triggers) and runs claude -p "<wrapped prompt>" with streaming output piped through claude_stream_viewer
5. Copies the generated video.mp4, slides.pdf, and the raw Claude event log to your output directory

The SKILL.md is what teaches Claude the actual craft: budgeting narration to a 1–2 minute target, generating per-scene voice-over with macOS say + ffmpeg, sizing each Remotion <Sequence> from its measured MP3 duration, and assembling still frames into a PDF with ImageMagick.

Usage

Run it straight from npm — no install needed. Your stdin is read as the video topic — you don't need to spell out "generate a video", build wraps it into a video-generation instruction for you:

echo "Tell me about quantum physics like I'm 10" \
  | npx prompt2video build

# or pipe a prompt file
cat prompt.txt | npx prompt2video build

# with custom directories
echo "my prompt" \
  | npx prompt2video build --tmp-dir /tmp --output-dir ~/Videos

Or install it globally:

npm install -g prompt2video
echo "my prompt" | prompt2video build

A richer prompt with a topic and source material:

USER_PROMPT=$(cat <<'EOF'
topic: why fastbrowser + a11y_parse are great to scrape the web with AI
description: |
  Based on those 2 folders, in a monorepo
  - https://github.com/jeromeetienne/skillmd_collection/tree/main/packages/a11y_parse
  - https://github.com/jeromeetienne/skillmd_collection/tree/main/packages/fastbrowser_cli
EOF
)

echo "$USER_PROMPT" | npx prompt2video build

The build command exits with an error if no prompt is piped on stdin.

Prompt controls

The prompt isn't just a topic — you can steer the render with optional key: value control lines alongside topic:/description:. Every key is optional; omit one and the skill falls back to a sensible default. You can also just write the instruction in plain English ("make it about a minute, vertical, for kids") and the skill infers the equivalent controls.

| Key | Controls | Example | Default | |---|---|---|---| | duration | target total length | 90s, 2m | 1–2 min | | scenes | number of scenes / slides | 5 | ~4–6 | | aspect | frame shape | 16:9, 9:16, 1:1 | 16:9 | | style | visual look — a preset and/or free-form descriptors | dark, neon accents | clean | | voice | macOS say voice | Samantha, Daniel | system default | | rate | speech rate (words per minute) | 160 | 175 | | tone | narration register | casual, formal | neutral | | audience | reading level / who it's for | kids, experts | general | | language | narration + on-screen language | French | English | | captions | burn per-scene subtitles on screen | on, off | off | | music | background-music mood, or off | upbeat, off | off |

Built-in style presets are clean, dark, corporate, playful, and cinematic; each fixes a palette, font, and transition style so the look stays cohesive across scenes. Free-form descriptors (e.g. brand hex colors, hand-drawn) layer on top.

A prompt exercising several controls:

USER_PROMPT=$(cat <<'EOF'
topic: why fastbrowser + a11y_parse are great to scrape the web with AI
description: |
  Based on those 2 folders, in a monorepo
  - https://github.com/jeromeetienne/skillmd_collection/tree/main/packages/a11y_parse
  - https://github.com/jeromeetienne/skillmd_collection/tree/main/packages/fastbrowser_cli
duration: 90s
scenes: 5
aspect: 9:16
style: dark, neon accents
voice: Samantha
tone: casual, like explaining to a friend
EOF
)

echo "$USER_PROMPT" | npx prompt2video build

Controls are interpreted by the bundled skill, so they're best-effort — the deterministic ones (aspect, scenes, voice) are honored exactly, while duration is verified against the measured voice-over length and adjusted to land within ±10% of the target.

Commands

Usage: prompt2video [options] [command]

Scaffold a Remotion project and stream Claude Code to generate a narrated AI
video from a prompt.

Commands:
  install [skill-folder]  Install all bundled skills into <skill-folder>/skills/
                          (default: .)
  build [options]         Scaffold the Remotion project, run Claude, and copy
                          the generated artifacts.
  help [command]          display help for command

build options

Options:
  -t, --tmp-dir <dir>     parent directory for the generated project (default: "/tmp")
  -o, --output-dir <dir>  output directory for the generated video (mp4/pdf/log) (default: "./outputs")
  -h, --help              display help for command

Requirements

This tool orchestrates several external programs. You need them on your PATH:

• Claude Code CLI (claude), authenticated
• Node.js (runs npx create-video, npx skills, npx claude_stream_viewer)
• ffmpeg / ffprobe — voice-over encoding and duration measurement
• ImageMagick (magick) — assembling slide PNGs into a PDF
• macOS say — text-to-speech for the narration (this piece is macOS-specific)

Skill-only install

The bundled skill can be dropped into any Claude Code agent on its own:

npx prompt2video install ~/.claude
# installs skills/prompt2video into ~/.claude/skills/prompt2video

License

MIT © Jerome Etienne