TikTok・YouTube Shorts・Instagram Reels で副業を始めたいが、Premiere を覚えたくない、顔出しも避けたい——MoneyPrinterTurbo(MIT、2026年5月時点 GitHub 6.7万+ スター)は トピック1つから縦横ショートを自動生成:脚本、ストック、ナレーション、字幕、BGM、レンダーまでほぼ無人です。
完全初心者向けの顔出しなし動画工場です。Streamlit Web UI はポート 8501、REST API は 8080。macOS では常時起動 Mac で夜間バッチが有効。調査→CSV→原稿の多エージェント運用は Hermes サブエージェント委任ガイド を参照——本記事は動画出力に集中します。
MoneyPrinterTurbo の内部パイプライン
MoneyPrinterTurbo uses a clear MVC layout (per upstream README). Data flows in one direction:
Topic / keyword
→ LLM script (config.toml llm_provider + API keys)
→ Stock footage (Pexels API) or local clips
→ TTS voice (edge / Azure)
→ Subtitles (edge or whisper ~3GB)
→ Background music (resource/songs/)
→ ffmpeg + ImageMagick → MP4 (9:16 or 16:9)
| Stage | Default path / config key | Output |
|---|---|---|
| Script | llm_provider in config.toml | Chinese or English copy |
| Footage | pexels_api_keys | HD royalty-free clips |
| Voice | Azure / edge voices | docs/voice-list.txt catalog |
| Subtitles | subtitle_provider = edge | SRT burned into video |
| Render | ffmpeg_path, ImageMagick | 1080x1920 (9:16) or 1920x1080 |
Quotable: Upstream recommends 8 GB RAM minimum, 16 GB for smoother batch runs, and 4 GB+ GPU VRAM only for heavier local whisper—not required with cloud LLMs and edge subtitles.
Source: MoneyPrinterTurbo README.
プラットフォーム別マトリクス
| Platform | Aspect ratio | Typical length | MoneyPrinterTurbo setting |
|---|---|---|---|
| TikTok | 9:16 vertical | 15–60 s | 1080x1920, short segments |
| Instagram Reels | 9:16 vertical | 30–90 s | Same; review text overlay manually |
| YouTube Shorts | 9:16 vertical | ≤60 s | Trim in upload or shorten segment duration |
| YouTube standard | 16:9 | 3–8 min | 1920x1080, longer script in Web UI |
If you target Shorts only, do enable batch generation and pick the best of 3–5 renders. If you need talking-head credibility, do not expect this tool to replace on-camera builds—it excels at stock B-roll + voice-over faceless formats.
macOS 8ステップ・ランブック
Eight steps from clone to optional API automation—tested on Apple Silicon and Intel Macs you control.
- Clone the repo — Avoid Chinese characters or spaces in the install path (upstream warning).
- Install dependencies with uv — Python 3.11 is the upstream default.
- Install ImageMagick — Required for subtitle rendering on macOS (ffmpeg via Homebrew).
- Configure API keys — Copy config.example.toml → config.toml. Set pexels_api_keys (free tier at pexels.com/api).
- Launch the Web UI — Open http://127.0.0.1:8501 in Chrome or Edge if the auto tab is blank.
- Generate your first vertical short — Enter a topic (e.g. "3 habits that improve sleep"), choose 9:16, pick voice + subtitle style, click generate.
- Batch mode for matrix publishing — Generate 3–5 MP4s with varied hooks; schedule on a Mac that does not sleep (pmset -c sleep 0 on plugged-in laptops).
- Optional API automation — Start uv run python main.py and hit http://127.0.0.1:8080/docs—wrap inside a Hermes delegate_task worker that drops topics from a spreadsheet.
git clone https://github.com/harry0703/MoneyPrinterTurbo.git
cd MoneyPrinterTurbouv python install 3.11
uv sync --frozenbrew install imagemagick ffmpeguv run streamlit run ./webui/Main.py --browser.gatherUsageStats=FalseHardware reference: Apple Mac mini specifications list 16 GB on base M4—enough for Web UI + one render; choose 24 GB with whisper locally plus browser tabs.
トラブルシューティング
RuntimeError: No ffmpeg exe could be found
症状: Render fails immediately; stack trace mentions IMAGEIO_FFMPEG_EXE.
対処: Install ffmpeg (brew install ffmpeg) or download a static build from gyan.dev ffmpeg builds:
[app]
ffmpeg_path = "/opt/homebrew/bin/ffmpeg
ImageMagick policy blocks temporary text files
症状: Error referencing pattern="@" and policy.xml.
対処: Edit ImageMagick policy.xml (Homebrew: /opt/homebrew/etc/ImageMagick-7/policy.xml)—change the @ pattern from rights="none" to rights="read|write".
OSError: Too many open files during batch render
症状: Batch job dies mid-queue on macOS.
対処: Raise the limit before launching Streamlit:
ulimit -n 10240
uv run streamlit run ./webui/Main.py --browser.gatherUsageStats=False
Whisper model download fails (China network)
症状: HuggingFace timeout when subtitle_provider = whisper.
対処: Stay on edge subtitles first. If you need whisper, manually download whisper-large-v3 (~3 GB) per upstream mirrors into MoneyPrinterTurbo/models/whisper-large-v3/.
副業ループ:トピック→バッチ→公開
A repeatable solopreneur loop for faceless channels:
- Monday — Brainstorm 10 keywords in a spreadsheet (Notion or CSV).
- Tuesday — Batch-generate 5 vertical MP4s; pick top 3 by manual preview.
- Wednesday–Friday — 水〜金 — TikTok / YouTube Shorts / Instagram Reels に1本ずつ投稿。フックはカルーセルやスレッドのキャプションに再利用。
- Weekend — Review analytics; feed winning topics back into next week's batch.
Pair with automation: Hermes subagent Worker A researches trending hooks, Worker B fills the CSV, you paste topics into MoneyPrinterTurbo—or call the REST API from a cron script on an always-on Mac.
関連記事
関連:Hermes サブエージェント委任ガイド(調査→CSV→原稿)が顔出しなし動画と補完関係です。
FAQ
MoneyPrinterTurbo は本当に無料?
MIT でセルフホスト無料。LLM API、Azure TTS、Pexels は従量課金。カジュアル運用で月 $5–20 程度の目安です。
顔出しなしで使える?
はい。ストック + 合成音声 + 焼き込み字幕が標準です。
TikTok と YouTube Shorts の設定は?
どちらも 9:16・1080x1920。Shorts は 60 秒未満が無難です。
Mac mini M4 に GPU は必要?
クラウド LLM + edge 字幕なら不要。ローカル whisper 時のみ GPU が有利です。
オンライン生成ツールとの違いは?
自ホストはバッチ制御と API。ターミナル不要なら https://reccloud.com 等を比較してください。