What it does

Feed it a topic and a niche—python -m verticals run --topic "..." --niche tech—and it researches via DuckDuckGo, writes a script with hook patterns, generates B-roll images, records a voiceover, burns in animated captions, adds mood-matched music, creates a thumbnail, and uploads to YouTube. Wall clock time is about three minutes; API cost is roughly $0.11 per video at default settings. It can also run entirely free using local Ollama + Edge TTS.

The interesting bit

The “niche intelligence” system is the actual engine. Every stage reads from a YAML profile that shapes tone, visual style, caption fonts, music mood, and thumbnail strategy without you writing a single prompt. The tech niche forbids “what’s up guys” and demands contrarian hooks; the cooking niche generates food photography and warm handwritten captions. Fifteen niches ship built-in, and you can author your own by editing a YAML file.

Key highlights

• Multi-provider LLM support: Claude, Gemini, GPT, Ollama local, or Claude CLI (free with Max subscription)
• Three TTS tiers: free Edge TTS (300+ voices, recommended default), premium ElevenLabs, or macOS say fallback
• Anti-hallucination gate: LLM is instructed to use only facts from live DuckDuckGo research, not training knowledge
• ffmpeg assembly with Ken Burns motion, automatic voice ducking, and word-level Whisper captions in ASS + SRT
• Resumable stages and individual stage CLI commands (draft, produce, upload)
• YouTube upload is private-by-default; TikTok, Reels, and X export are v3.1 roadmap items

Caveats

• Gradio UI, Docker, Colab, and Kokoro TTS are explicitly listed as “not shipped yet”—roadmap only
• Image generation relies on Gemini Imagen with solid-color fallback frames if it fails; no Pexels or Replicate integration yet
• The repo went through a “credibility cleanup” (v3.0.1) after a viral traffic spike, suggesting the docs previously oversold capabilities

Verdict

Worth a look if you run a faceless channel, test content hypotheses, or just want to see how far you can push automated ffmpeg pipelines. Skip it if you care about platform diversity today or find the idea of AI-generated “contrarian takes” existentially depressing.