Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Automated YouTube Shorts pipeline: news → script → AI visuals → voiceover → captions → upload
| Date | Stars |
|---|---|
| 2026-07-24 | 2130 |
| 2026-07-25 | 2131 |
| 2026-07-28 | 2136 |
| 2026-07-30 | 2137 |
| 2026-08-06 | 2137 |
Today
— stars today
This week
— stars this week
This month
— stars this month
Momentum
0.0
growth rate 0.00%/day
# Verticals v3
**The open source AI content engine with built-in niche intelligence.**
> Topic in. Published Short out. Any niche. ~$0.11 per video.
>
> **[Quickstart](#quickstart) · [Hosted Version](https://verticals.gg)**
> Repo note: the product is called **Verticals v3**. The GitHub repository is `youtube-shorts-pipeline`.
```
python -m verticals run --topic "Sam Altman just mass-fired 200 safety researchers" --niche tech
```
That one command researches the topic, writes a hook driven script tuned to tech YouTube, generates cinematic b roll, records a natural voiceover, burns in animated captions, adds mood matched background music, generates a thumbnail, and uploads it to YouTube. ~90 seconds of video, ~3 minutes of wall time, ~$0.11 in API costs.
## What Changed in v3
v2 was an esports news pipeline. v3 is a **general purpose content engine** that works for any niche, any topic, any creator.
The biggest change: **Niche Intelligence**. Every stage of the pipeline now reads from a niche profile that shapes script tone, visual style, caption aesthetics, music mood, and thumbnail strategy. Ship a cooking Short and it writes like a cooking creator, generates food photography b roll, and picks warm upbeat background music. Ship a true crime Short and the tone shifts to suspenseful, the visuals go dark and cinematic, and the music drops to ambient tension.
15 niches ship out of the box. Build your own in 5 minutes.
Other highlights: multi provider LLM support (Claude, Gemini, GPT, Ollama local), free TTS via Edge TTS, YouTube upload, topic discovery, resumable stages, and a local-first config model.
## Current Release: v3.1.0
v3.1.0 brings community-contributed providers and reliability fixes: MiniMax (LLM + TTS) and 60db (TTS) as optional providers, niche-aware caption fonts so CJK and other non-Latin scripts render correctly, a working `edge-tts` pin (6.x is rejected by Microsoft with 403s), and clearer errors for the most-reported setup problems.
Implemented today:
- research with DuckDuckGo plus optional source scraping
- script and metadata generation through Claude, Gemini, GPT, Ollama, MiniMax, LiteLLM, or Claude CLI
- b roll and thumbnail image generation through Gemini Imagen, with fallback frames
- voiceover through Edge TTS, ElevenLabs, MiniMax, 60db, or macOS `say`
- Whisper captions with ASS burn-in plus SRT export, niche-configurable fonts
- ffmpeg assembly with Ken Burns motion, background music, and voice ducking
- private-by-default YouTube upload
Not shipped yet: Gradio UI, Docker, Colab, TikTok/Reels/X upload, Pexels, Replicate, ComfyUI, and Kokoro TTS. Those are roadmap items, not current features.
## How It Works
```
┌─────────────────────────────────────────────────────────────────┐
│ NICHE PROFILE │
│ Loaded once. Shapes every stage. 15 built in or bring your own │
└─────────────┬───────────────────────────────────────────────────┘
│
▼
┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐
│ RESEARCH │→ │ SCRIPT │→ │ VISUALS │→ │ VOICE │→ │ CAPTIONS │→ │ ASSEMBLE │→ UPLOAD
│ │ │ │ │ │ │ │ │ │ │ │
│ DuckDuck │ │ LLM with │ │ Gemini │ │ ElevenLabs│ │ Whisper │ │ ffmpeg │
│ Go + web │ │ niche │ │ fallback │ │ Edge TTS │ │ word │ │ Ken Burns│
│ scraping │ │ persona │ │ frames │ │ say │ │ level │ │ + music │
│ │ │ + hooks │ │ │ │ │ │ ASS+SRT │ │ ducking │
└──────────┘ └──────────┘ └──────────┘ └──────────┘ └──────────┘ └──────────┘
```
**Stage by stage:**
**Research** — Searches DuckDuckGo (and optionally scrapes source URLs) for live facts. Every name, number, and claim in the final script traces back to this research. This is the anti hallucination gate: the LLM is instructed to use only facts from research data, never its training knowledge.
Excerpt of 20,531 characters
Read on GitHub
Claude · @anthropics
5
4
Octopus · Independent Developer · United States
1
Aarish Alam · ScaledFocus · India
1
1
1
Would you bet a product on this? Bounded 0–100 and slow moving.
matched fp:5c81767b8668d450, topic:whisper
matched fp:5c81767b8668d450, topic:automation