openshorts-video

# OpenShorts — make a short video from one sentence The user gives a topic, an article URL, or a script. You do everything else: install, health-check, script, footage, voice, captions, render, publish pack. Deliver file paths at the end. Do not stop after explaining the steps. ## Required behavior 1. Ask the user for at most one thing, once: an API key for a text model (DeepSeek / Kimi / GLM / OpenAI-compatible…) **only if** `openshorts new` reports a missing key. Everything else (footage, voice, rendering) is free and needs no key. 2. Do not ask for confirmation before installing, running `doctor`, installing the bundled ffmpeg, or rendering. Send one short progress line, then execute. 3. Run long commands in the foreground with a timeout of at least 15 minutes. Do not poll with `sleep` / `ls` loops. 4. On failure read only the last 30 lines of output; OpenShorts prints the real cause (missing libass, missing key, quality-check items), not just an exit code. 5. Never print API keys or the contents of `~/.ao` / `~/.openshorts/config.json`. ## Defaults Unless the user says otherwise: Chinese talking-head explainer, 60 s, 1080×1920, free Edge TTS voice, Wikimedia Commons / Pexels footage, captions burned in, AI-content label on. Output goes to `~/OpenShorts/<project>/`. For English use `--lang en` (script, voice and caption line-breaking switch together). ## Execution ### 1. Install (once) ```bash node -v # must be >= 20 npx openshorts version # installs on first use; no clone needed ``` ### Alternative: use the MCP server instead of the CLI If your host supports MCP, register `npx openshorts mcp` and call `create_video` → poll `job_status` every 20–30 s → `doctor` on failure. It runs the same pipeline as the CLI steps below and returns the same file paths; prefer it when the host would otherwise time out on a 10-minute shell command. ### 2. Health check — do not skip ```bash node bin/openshorts.mjs doctor ``` If it says ffmpeg lacks **libass** (captions cannot be burned in — Homebrew's ffmpeg has this problem), run: ```bash node bin/openshorts.mjs install-ffmpeg # ~40 MB, pinned build with sha256 check, installed to ~/.openshorts/bin only ``` ### 3. Write the script ```bash node bin/openshorts.mjs new koubo-kepu --topic "<topic or full script>" # Chinese node bin/openshorts.mjs new --lang en --topic "<topic or full script>" # English ``` Add `--local-dir <folder>` to prefer the user's own footage, `--voice <id>` to pick a voice (`npm run voices` lists them). If the command reports a missing text-model key, ask the user for one key, then configure it once: ```bash DEEPSEEK_API_KEY=<key> node bin/openshorts.mjs new koubo-kepu --topic "<topic>" # any provider the engine knows: DEEPSEEK_API_KEY / MOONSHOT_API_KEY / OPENAI_API_KEY … ``` The command prints the project path: `~/OpenShorts/<project>/project.json`. ### 4. Estimate, render, pack ```bash node bin/openshorts.mjs estimate ~/OpenShorts/<project>/project.json # cost is ¥0 on the default path; tells the expected wait node bin/openshorts.mjs run ~/OpenShorts/<project>/project.json # footage → voice → captions → mp4 + quality check node bin/openshorts.mjs export ~/OpenShorts/<project>/project.json --platform douyin # or shorts / bilibili / channels ``` `run` reuses finished shots on re-runs; `run … --only s2` redoes one shot. `openshorts rm <project.json> --yes` deletes a project the user no longer wants (ask first — it removes the whole folder). If `run` ends with quality-check ⛔ items, report them verbatim to the user instead of calling it done. ## Exit handling On success `run` prints: ```text ✓ 成片：/Users/…/OpenShorts/<project>/<project>.mp4（58.3s，出片用时 47 秒） 字幕：…/<project>.srt 封面：…/<project>-cover.jpg 发布文案：…/<project>-发布文案.txt ``` Return those four paths plus the publish pack directory from `export`. If the output contains `⛔`, return the message as-is and stop; do not retry blindly. ## Notes for the agent - Everything runs on the user's machine; no video is uploaded anywhere. Only the script step calls the user's own text-model API. - `openshorts drama --plan -i story="…"` previews the cost of an AI mini-drama before spending anything; only run `drama` without `--plan` after the user has seen the estimate. - Web UI alternative: `npm run openshorts` opens http://127.0.0.1:4174 (the same pipeline, four steps).