# YouTube Shorts Pipeline 🎬 > Turn a one-line topic into a published YouTube Short in minutes. > Fully automated: **research → script → AI visuals → voiceover → captions → upload.** --- ## What it does The pipeline has three stages, each runnable independently or as one shot: | Stage | What happens | |-------|-------------| | **Draft** | DuckDuckGo research → Claude writes a 60–90 sec script → Gemini generates b-roll image prompts, YouTube title/description/tags, Instagram caption, thumbnail prompt | | **Produce** | Gemini Imagen 3 generates 3 AI visuals (9:16 portrait) → Ken Burns animation → ElevenLabs voiceover → ffmpeg assembles video → Whisper generates SRT captions | | **Upload** | YouTube Data API v3 uploads the video with full metadata + SRT caption track | **Anti-hallucination gate:** Claude is shown live DuckDuckGo research and instructed to use *only* names, scores, and facts found there. No fabricated content. --- ## Demo A typical run looks like this: ``` $ python3 scripts/pipeline.py run --news "India wins VCT Pacific 2026" 👋 First run detected. Running setup... [wizard prompts for API keys...] 🎬 Drafting: India wins VCT Pacific 2026 Researching topic via DuckDuckGo... Found 8 snippets. ✅ Draft saved: ~/.youtube-shorts-pipeline/drafts/1708512345.json 📝 Script: "India just made history at VCT Pacific 2026..." 🎬 Producing EN video... Generating b-roll frame 1/3 via Gemini Imagen... Generating b-roll frame 2/3 via Gemini Imagen... Generating b-roll frame 3/3 via Gemini Imagen... Generating en voiceover via ElevenLabs... Generating SRT captions via Whisper... Assembling video... Video assembled: ~/.youtube-shorts-pipeline/media/pipeline_1708512345_en.mp4 Uploading pipeline_1708512345_en.mp4... Upload progress: 100% Uploaded: https://youtu.be/xXxXxXxXxXx Captions uploaded. 🎉 Done! https://youtu.be/xXxXxXxXxXx ``` Total time: ~3–5 minutes per video on a typical internet connection. --- ## Prerequisites - **Python 3.10+** - **ffmpeg** — video assembly ```bash brew install ffmpeg # macOS apt install ffmpeg # Ubuntu/Debian ``` - **Whisper** — caption generation ```bash pip install openai-whisper ``` - **API accounts:** - [Anthropic](https://console.anthropic.com) — Claude (script generation) - [ElevenLabs](https://elevenlabs.io) — voiceover *(optional; macOS `say` fallback)* - [Google AI Studio](https://aistudio.google.com) — Gemini Imagen (b-roll) - [Google Cloud Console](https://console.cloud.google.com) — YouTube Data API v3 (upload) --- ## Quick Start **1. Install Python dependencies** ```bash pip install anthropic google-api-python-client google-auth google-auth-oauthlib pillow requests openai-whisper ``` **2. Clone or extract this skill** ```bash unzip youtube-shorts-pipeline.skill -d youtube-shorts-pipeline cd youtube-shorts-pipeline ``` **3. Run the pipeline — setup wizard launches automatically** ```bash python3 scripts/pipeline.py run --news "your topic here" --dry-run ``` The `--dry-run` flag skips produce/upload so you can preview the script first. **4. Set up YouTube OAuth** *(when ready to upload)* ```bash python3 scripts/setup_youtube_oauth.py ``` Follow the prompts — you'll need a `client_secret.json` from Google Cloud Console (see [Full Setup](#full-setup)). **5. Produce and upload** ```bash python3 scripts/pipeline.py run --news "your topic here" ``` --- ## Full Setup See [`references/setup.md`](references/setup.md) for: - Step-by-step API key acquisition - YouTube OAuth setup with Google Cloud Console walkthrough - Config file reference --- ## Usage ### Draft — generate script and metadata only ```bash python3 scripts/pipeline.py draft --news "your topic" [--context "esports news channel"] ``` Output: `~/.youtube-shorts-pipeline/drafts/<timestamp>.json` ### Produce — generate video from a saved draft ```bash python3 scripts/pipeline.py produce --draft ~/.youtube-shorts-pipeline/drafts/<id>.json [--lang en|hi] ``` Override the script inline (after editing the draft): ```bash python3 scripts/pipeline.py produce --draft <path> --script "Your custom script here." ``` ### Upload — push a produced video to YouTube ```bash python3 scripts/pipeline.py upload --draft ~/.youtube-shorts-pipeline/drafts/<id>.json [--lang en|hi] ``` ### Full pipeline in one command ```bash python3 scripts/pipeline.py run --news "your topic" [--lang en|hi] [--dry-run] [--context "..."] ``` --- ## Configuration On first run, the wizard saves your keys to `~/.youtube-shorts-pipeline/config.json`: ```json { "ANTHROPIC_API_KEY": "YOUR_ANTHROPIC_KEY_HERE", "ELEVENLABS_API_KEY": "YOUR_ELEVENLABS_KEY_HERE", "GEMINI_API_KEY": "YOUR_GEMINI_KEY_HERE" } ``` > **Note:** `config.json` is saved with `0600` permissions (owner read/write only). Never commit this file to version control. **Environment variables take priority** over the config file. You can override any key: ```bash ANTHROPIC_API_KEY=your-key-here python3 scripts/pipeline.py draft --news "..." ``` | Variable | Required | Description | |----------|----------|-------------| | `ANTHROPIC_API_KEY` | ✅ Yes | Claude API key | | `GEMINI_API_KEY` | ✅ Yes | Gemini Imagen API key | | `ELEVENLABS_API_KEY` | Optional | ElevenLabs TTS key (fallback: macOS `say`) | | `VOICE_ID_EN` | Optional | ElevenLabs voice ID for English (default: George) | | `VOICE_ID_HI` | Optional | ElevenLabs voice ID for Hindi | ### Data directories All data is stored in `~/.youtube-shorts-pipeline/`: ``` ~/.youtube-shorts-pipeline/ ├── config.json ← API keys ├── youtube_token.json ← YouTube OAuth token ├── drafts/ │ └── <timestamp>.json ← draft scripts + metadata ├── media/ │ ├── pipeline_<id>_en.mp4 │ └── pipeline_<id>_en.srt └── logs/ ``` --- ## Cost per video Rough estimates at standard API rates: | Service | Cost per video | |---------|---------------| | Anthropic (Claude Sonnet) | ~$0.02 | | Google Gemini Imagen (3 images) | ~$0.03 | | ElevenLabs (60–90 sec audio) | ~$0.05 | | **Total** | **~$0.10** | ElevenLabs free tier is blocked on server IPs. Pro plan ($22/mo) is required for non-local use. Gemini and Anthropic have generous free tiers for low-volume use. --- ## Troubleshooting See [`references/troubleshooting.md`](references/troubleshooting.md) for common errors including: - YouTube quota errors (`uploadLimitExceeded`, `quotaExceeded`) - ElevenLabs 401/403 errors - ffmpeg issues - Whisper Hindi/Urdu output - OAuth token expiry --- ## Extending - **New languages:** Add a `--lang` option and a matching `VOICE_ID_XX` env var. ElevenLabs supports 30+ languages with `eleven_multilingual_v2`. - **Custom voices:** Change `VOICE_ID_EN` env var to any ElevenLabs voice ID. - **B-roll style:** Edit the `broll_prompts` array in the saved draft JSON before running `produce`. - **Script editing:** Use `--script "your edited script"` on the `produce` command. --- ## Security This pipeline handles API keys and OAuth tokens. The following measures are in place: - **Credential storage:** `config.json` and `youtube_token.json` are written with `0600` permissions (owner-only). Never commit these files — they are covered by `.gitignore`. - **API key transmission:** The Gemini API key is sent via the `x-goog-api-key` header, not as a URL query parameter, so it won't leak into logs or error messages. - **Error handling:** API error messages are sanitized to never include credentials. - **Upload privacy:** Videos are uploaded as **private** by default. Change to `public` or `unlisted` manually on YouTube when ready. - **OAuth scope:** YouTube OAuth requests the minimum scopes needed (`youtube.upload` + `youtube.force-ssl`), not full account access. - **Prompt injection mitigation:** Search result snippets injected into the Claude prompt are truncated and wrapped in boundary markers to reduce prompt injection risk. - **LLM output validation:** Fields returned by Claude are type-checked before use in metadata and file operations. --- ## Licence MIT — free to use, modify, and distribute. Attribution appreciated but not required.