clawvard-short-video

A Node CLI that turns one topic into one 1080×1920 vertical short MP4 — AI-written script, AI voiceover, Pexels stock B-roll, burned subtitles, ducked BGM — in one command.

Built on top of @clawvard/sdk so script generation and voiceover stay on Clawvard's typed service surface; no OpenAI-compatible base URL leaks into your wrapper.

Install

npm install -g clawvard-short-video
# or run ad-hoc:
npx clawvard-short-video --help

Prerequisites on your machine:

• Node ≥ 18 (node -v)
• ffmpeg on PATH (macOS: brew install ffmpeg; Ubuntu/Debian: sudo apt install ffmpeg)
• (optional fallback) Python 3 + pip install --user edge-tts for local voiceover

Environment:

• CLAW_API_KEY — mint at https://clawvard.school/token-relay. Required for cv.llm.chat (script) and cv.media.voiceover (narration).
• PEXELS_API_KEY — sign up free at https://www.pexels.com/api/. Required for stock-footage search.

Usage

npx clawvard-short-video \
  --topic "为什么坚持早起的人，更容易抓到生活里隐藏的小确幸" \
  --duration 60 \
  --lang zh \
  --voice cv:zh-female-warm \
  --out ./output/short-earlybird.mp4

You get back a JSON summary on stdout with the file path, voice source, captions, and Pexels source URLs for every clip used (for attribution).

What it does

1. Script — cv.llm.chat writes a ~60-second spoken script: 3-second hook + 2–3 concrete visual moments + one short CTA. Stage directions / BGM cues are auto-stripped.
2. Voiceover — cv.media.voiceover renders the script as mp3. If the platform TTS upstream returns service_unavailable (or 404/503), the wrapper falls back to local Edge-TTS (installed on demand).
3. Stock — searches the Pexels Videos API for B-roll matching keywords drawn from each script segment, prefers orientation=portrait, and downloads the closest-to-1080×1920 file.
4. Subtitles — splits the script into ≤22-CJK-char (≤60 Latin) captions, proportionally allocates each caption a slice of the voice duration, and writes a standard .srt.
5. Assemble — ffmpeg crops every clip to 1080×1920, concatenates to the voice length, burns subtitles into the bottom safe area, mixes in a soft synth pad BGM (or your --bgm path.mp3), and sidechain-ducks the BGM under the voice.

Library use

The pipeline is also importable, so agent frameworks can drive it without spawning a subprocess:

import { runPipeline } from "clawvard-short-video";  // npm name (the @clawvard scope is reserved for our SDK packages)

const result = await runPipeline({
  topic: "为什么坚持早起的人，更容易抓到生活里隐藏的小确幸",
  out: "./output/short-earlybird.mp4",
  lang: "zh",
  voice: "cv:zh-female-warm",
  duration: 60,
});

Source

This package is the engineered wrapper for the Clawvard course topic-to-short. See it at https://clawvard.school/courses/topic-to-short.

Course design draws on the open-source MoneyPrinterTurbo workflow (MIT). The implementation is original — there is no upstream Python dependency to install.

License

Apache-2.0