← all repositories
linyqh/NarratoAI

The AI video explainer that scammers love to resell

An open-source pipeline that writes commentary, edits clips, dubs, and subtitles videos—so popular that knockoff sellers are peddling it on TikTok.

9.7k stars Python Domain AppsAgents
NarratoAI
Velocity · 7d
+15
★ / day
Trend
steady
star history

What it does NarratoAI is a Python-based pipeline that ingests a video and, via LLM calls, spits out a finished “movie commentary” clip: scripted narration, automated edits, synthesized voiceover, and burned-in subtitles. It targets the short-drama and film-explanation content niche that dominates Chinese social video platforms. You bring API keys; it brings the glue between vision models (Qwen2-VL), LLMs (DeepSeek, Gemini, etc.), TTS engines (Tencent, IndexTTS2 for voice cloning), and MoviePy.

The interesting bit The project is a direct descendant of MoneyPrinter/MoneyPrinterTurbo, but pivoted from generic viral video assembly to culturally specific “影视解说”—a genre where a fast-talking narrator summarizes a film’s plot over heavily edited clips. The README’s most prominent section is a fraud warning: scammers on Douyin and Bilibili are rebranding the free tool and selling it. That is one strange metric of product-market fit.

Key highlights

  • One-click Docker or Windows bundled install; Streamlit UI on localhost:8501
  • Supports “short-drama remix” mode with auto-transcription, cache clearing, and CapCut draft export
  • Recently added Fun-ASR for subtitle transcription and removed LiteLLM for a unified OpenAI-compatible request path
  • Cloud-hosted version exists at narratoai.cn; local version requires Python 3.12+, 4-core CPU, 8GB RAM, no GPU required
  • License explicitly prohibits commercial use without author contact

Caveats

  • Documentation and UI are Chinese-first; English README exists but the project is clearly optimized for a Chinese-speaking workflow
  • Roadmap items (face matching, more TTS engines) are unchecked, so the current feature set is what you get
  • The “register for free tokens” affiliate pitch for Silicon Flow is plastered in the README, which feels more marketing than engineering

Verdict Worth a spin if you produce short-form film commentary or need a reference architecture for LLM-driven video pipelines. Skip it if you need polished English output, commercial rights, or a project with a stable long-term roadmap—the author is already promoting a successor product (Speclip) as the “true video editing agent.”

heatdrop uses Google Analytics to see which pages get read — nothing else. Your call. How we handle data.