Stop Editing Videos Manually! This Open Source AI Pipeline Does It All

Stop Editing Videos Manually! This Open Source AI Pipeline Does It All

What if I told you that one line of Python^{↗ Bright Coding Blog} could replace your entire video production team?

No, seriously. One. Single. Command.

Picture this: It's 2 AM. Your content calendar is screaming. You've got three Shorts due tomorrow, your editor ghosted you, and you're staring at a blank timeline wondering how you'll crank out another 90-second tech explainer before sunrise. Sound familiar?

Here's the brutal truth most creators won't admit: the bottleneck isn't creativity—it's execution. Researching topics, writing hooks that actually land, finding B-roll that doesn't scream "stock footage," recording voiceovers without sounding like a robot, burning in captions that people actually read, and somehow uploading before the trend dies... it's exhausting.

But what if the machine could handle the machine work?

Enter Verticals v3—the open-source AI content engine that transforms a raw topic into a published YouTube Short in roughly three minutes of wall time for about eleven cents. Not eleven dollars. Eleven cents.

Built by Dr. Rushindra Sinha—a Stanford GSB graduate who built a #1 global game server at 17 and co-founded South Asia's only Valorant Champions Tour franchise—this isn't some weekend hackathon project. This is battle-tested automation that understands something critical: every niche speaks its own language.

Ready to see how deep this rabbit hole goes? Let's dismantle the entire pipeline.

What Is Verticals v3?

Verticals v3 (repository name: youtube-shorts-pipeline) is an end-to-end automated pipeline for producing short-form video content. Born from v2's esports news automation roots, v3 evolved into a general-purpose content engine capable of handling any niche, any topic, any creator persona.

The project lives at github.com/rushindrasinha/youtube-shorts-pipeline under the MIT license, meaning you can fork it, modify it, and even commercialize it without legal headaches.

But here's what makes it genuinely disruptive: Niche Intelligence.

Most AI video tools treat content like a commodity—same generic voice, same visual style, same forgettable output regardless of whether you're covering crypto crashes or sourdough starters. Verticals v3 flips this model entirely. Every stage of its pipeline reads from a niche profile—a YAML configuration file that shapes script tone, visual aesthetics, caption styling, music mood, and thumbnail strategy.

Ship a cooking Short? It writes like a cooking creator, generates food photography B-roll, and picks warm, upbeat background music. Ship true crime? The tone shifts to suspenseful, visuals go dark and cinematic, music drops to ambient tension.

Fifteen niches ship out of the box: tech, gaming, finance, fitness, cooking, travel, true crime, science, politics, entertainment, sports, fashion, education, motivation, and comedy. Build your own in five minutes by copying and editing any existing profile.

This isn't template-based automation. This is context-aware content synthesis—and it's why Verticals v3 is quietly becoming the secret weapon among developers who understand that distribution speed wins in the attention economy.

Key Features That Separate It From the Pack

Let's dissect what makes this pipeline genuinely powerful, not just theoretically interesting.

Multi-Provider LLM Flexibility

You're not locked into OpenAI's pricing or Anthropic's rate limits. Verticals v3 supports Claude, Gemini, GPT, and local Ollama models out of the box. Want premium quality? Claude costs roughly $0.02 per script. Want free? Ollama runs entirely locally with zero API spend. The pipeline abstracts provider differences so you can swap backends without rewriting prompts.

Free Tier That Actually Works

Edge TTS—Microsoft's free, cross-platform text-to-speech engine with 300+ voices—is the recommended default. No, really. The creator explicitly recommends it over paid alternatives for most use cases. Combine this with Pexels stock footage (completely free, no API key needed) and Gemini's generous free tier for image generation, and you've got a $0.00 mode that genuinely produces publishable content.

Anti-Hallucination Research Layer

Here's where most AI content tools fail catastrophically: they let the LLM hallucinate facts. Verticals v3 routes all research through DuckDuckGo searches with optional source URL scraping. Every name, number, and claim in the final script traces back to this live research data. The LLM is explicitly instructed: use only facts from research, never training knowledge. This is the difference between content that gets you sued and content that builds authority.

Word-Level Caption Precision

Using OpenAI's Whisper for transcription, the pipeline generates both ASS files (burned-in with word-by-word yellow highlighting) and SRT files (uploaded for YouTube's closed captions). The visual styling follows niche profiles—bold energetic fonts for gaming, clean minimal for tech, warm handwritten feel for lifestyle. This isn't generic subtitle dumping; it's attention-engineered text animation.

Ken Burns B-Roll with Audio Ducking

The assembly stage uses ffmpeg to combine animated B-roll (slow zooms and pans across generated images), voiceover, burned captions, and mood-matched background music with automatic voice ducking—music volume drops when speaking resumes. These are professional post-production techniques that typically require After Effects expertise or expensive editors.

Multi-Platform Export Architecture

YouTube uploads are stable today. TikTok, Instagram Reels, and X (Twitter) exports are architected for v3.1. The pipeline generates platform-specific metadata during scripting—different hooks, descriptions, and hashtag strategies optimized per platform.

Real-World Use Cases Where This Shines

Solo Creator Scaling to Daily Uploads

You're a one-person operation covering tech news. Previously, each Short consumed 3-4 hours: research, scripting, recording, editing, uploading. With Verticals v3, you batch-generate five topical Shorts during your morning coffee, review and tweak during lunch, and maintain a daily publishing schedule that algorithms reward without burning out.

News Aggregation Channels

Build automated channels that track trending topics across Reddit, Hacker News, Google Trends, and RSS feeds. The verticals topics --niche tech --limit 20 command surfaces fresh angles. Combine with --auto-pick for hands-off topic selection. This is how you build thematic channels that never sleep.

Agency White-Label Production

Marketing agencies managing multiple client accounts can maintain distinct niche profiles per client—one brand gets the "finance" persona, another gets "lifestyle." Same infrastructure, completely different output character. The YAML-based configuration makes client-specific customization trivially version-controllable.

Multilingual Content Expansion

With support for English, Hindi, Spanish, Portuguese, German, French, Japanese, and Korean, you can repurpose successful English content for global audiences without re-hiring production teams. The --lang flag propagates through script generation, voice synthesis, and caption creation.

Rapid Trend Response

When breaking news hits, speed beats perfection. The pipeline's ~3 minute turnaround from topic to upload means you can be first to cover emerging stories—critical for algorithmic favor on platforms where recency signals boost distribution.

Step-by-Step Installation & Setup Guide

Prerequisites

• Python 3.10+
• ffmpeg installed system-wide
• Git

Option A: CLI Installation (Recommended for Developers)

# Clone the repository
git clone https://github.com/rushindrasinha/youtube-shorts-pipeline.git

# Enter project directory
cd youtube-shorts-pipeline

# Install dependencies
pip install -r requirements.txt

Option B: Docker^{↗ Bright Coding Blog} (Simplest for Non-Developers)

# Build and launch with web UI
docker compose up --build
# Opens Gradio interface at http://localhost:7860

First-Run Configuration

The setup wizard triggers automatically on first execution:

# This will prompt for API keys interactively
python -m verticals run --topic "your test topic" --niche tech

Or configure manually by creating ~/.verticals/config.json with 0600 permissions (owner read/write only):

{
  "ANTHROPIC_API_KEY": "sk-ant-...",
  "GEMINI_API_KEY": "AIzaSy...",
  "PEXELS_API_KEY": "your-pexels-key"
}

Minimum viable setup for free tier:

# Zero API cost configuration
python -m verticals run \
  --topic "Sam Altman just mass-fired 200 safety researchers" \
  --niche tech \
  --provider ollama \
  --voice edge \
  --visuals pexels

This uses local LLM inference (requires 7B+ parameter model capability), free Microsoft voices, and stock footage. Quality is lower than premium stack but completely free.

YouTube Upload Setup

OAuth authentication requires one-time setup:

python scripts/setup_youtube_oauth.py

Follow the browser flow to authorize upload permissions. Videos publish as private by default—review before manually switching to public.

REAL Code Examples from the Repository

Let's examine actual implementation patterns from the Verticals v3 codebase, with detailed explanations of what each section accomplishes.

Example 1: The One-Command Full Pipeline

python -m verticals run --topic "Sam Altman just mass-fired 200 safety researchers" --niche tech

This is the headline feature—a single CLI invocation that triggers the entire six-stage pipeline. Behind this simplicity lies sophisticated orchestration: the --topic string seeds the research stage, while --niche tech loads niches/tech.yaml to configure every downstream decision. The pipeline automatically handles dependency ordering (research must complete before scripting, scripting before visual generation, etc.) with resume capability if interrupted.

Example 2: Niche Profile Definition (Tech Niche)

# niches/tech.yaml — excerpt showing script intelligence
name: tech
display_name: "Tech & AI News"

script:
  tone: "informed, slightly opinionated, conversational"
  pacing: "fast, dense with facts, no filler"
  hooks:
    - pattern: "contrarian_take"
      template: "Everyone is celebrating {topic}. Here's why that's a problem."
    - pattern: "breaking_news"
      template: "This just happened and nobody is talking about it."
    - pattern: "prediction"
      template: "{topic} changes everything. Here's what happens next."
    - pattern: "explainer"
      template: "Let me explain {topic} in 60 seconds because most people are getting this wrong."
    - pattern: "comparison"
      template: "{thing_a} vs {thing_b}. One of these wins and it's not even close."
  cta_variants:
    - "Follow for daily tech breakdowns."
    - "Subscribe. I cover AI news nobody else is talking about."
    - "Drop a comment: do you agree?"
  word_count: "150 to 170"
  forbidden: ["like and subscribe", "smash that bell", "what's up guys"]

This YAML is the secret sauce. Notice how it doesn't just define tone—it constrains behavior. The forbidden list prevents cringe-inducing generic calls-to-action that scream "amateur." The hooks section provides templated patterns with variable substitution ({topic}, {thing_a}), giving the LLM structured creativity rather than blank-page anxiety. The word_count constraint ensures 150-170 words translates to roughly 90 seconds at 160 WPM—optimal for Shorts retention algorithms.

Example 3: Provider-Agnostic CLI Invocation

# Override defaults per execution without config changes
python -m verticals run \
  --topic "headline" \
  --niche cooking \
  --provider ollama        # Use local LLM instead of Claude

The --provider flag demonstrates the abstraction layer's power. Without touching configuration files, you can downgrade to free local inference for experimental topics, then switch back to Claude for premium productions. This flexibility is critical for cost management—test cheap, publish premium.

Example 4: Granular Stage Control

# Generate script draft only, skip production
python -m verticals draft --topic "headline" --niche tech

# Produce video from existing draft
python -m verticals produce --draft <path> --lang en

# Upload previously produced video
python -m verticals upload --draft <path> --platform youtube

# Discover trending topics for your niche
python -m verticals topics --niche tech --limit 20

These atomic stage commands enable sophisticated workflows. Draft multiple scripts during low-cost hours, human-review for brand safety, then batch-produce during off-peak electricity rates. The --discover --auto-pick combination enables fully autonomous operation—find trending topics, select highest-potential, generate and upload without human intervention.

Example 5: Zero-Cost Configuration Explicit

python -m verticals run \
  --topic "your topic" \
  --niche tech \
  --provider ollama \
  --voice edge \
  --visuals pexels

This explicit stack declaration is the democratization play. Ollama (local LLM, zero API cost), Edge TTS (Microsoft's free voice synthesis), and Pexels (stock footage, no generation cost). The creator documents this as "$0.00 Mode" with honest quality caveats—lower than premium stack, but functional. This transparency builds trust and enables global creators with limited USD access.

Advanced Usage & Best Practices

Batch Production with Topic Discovery

# Generate 10 videos from trending tech topics
for topic in $(python -m verticals topics --niche tech --limit 10 | jq -r '.[].title'); do
  python -m verticals run --topic "$topic" --niche tech --dry-run
done

Use --dry-run to validate scripts before committing production costs. Review drafts in ~/.verticals/drafts/, approve winners, then re-run without --dry-run.

Custom Niche Development

Copy niches/tech.yaml to niches/your_niche.yaml. Modify visual subjects, voice characteristics, and hook patterns. Reference with --niche your_niche. The YAML structure is self-documenting—no code changes required.

Cost Optimization Strategy

Resume and State Management

The pipeline maintains stage completion state. If visual generation fails (rate limit, GPU OOM), fix the issue and re-run—the pipeline skips completed stages automatically. Use --force to override and re-execute specific stages.

Comparison with Alternatives

The verdict: Verticals v3 wins on cost control, customization depth, and data sovereignty. SaaS alternatives optimize for ease-of-use at the expense of recurring costs and generic output. If you're technical enough to run Python, Verticals offers 10x cost efficiency with infinite customization potential.

FAQ

Is this against YouTube's terms of service?

YouTube prohibits spam, deceptive practices, and low-quality mass uploads. Verticals v3 produces substantive, researched content with human-reviewable drafts. The default private upload setting ensures you review before publishing. As with any automation tool, responsible use—quality over quantity—keeps you compliant.

Can I use this for commercial client work?

Absolutely. MIT license permits commercial use without attribution requirements. Many agencies white-label the output. The niche profile system makes maintaining distinct brand voices across multiple clients straightforward.

How good is the "free" quality actually?

Ollama's quality depends heavily on model choice—Llama 3.1 70B approaches Claude 3.5 Sonnet for scripting, while 7B models show noticeable degradation. Edge TTS voices are surprisingly natural for free synthesis. Pexels stock footage limits visual uniqueness but maintains professional baseline. For channels where content velocity matters more than premium polish, free mode competes effectively.

What about copyright on AI-generated visuals?

Current legal landscape is evolving. Gemini Imagen and Replicate Flux outputs have unclear copyright status in some jurisdictions. Pexels stock footage carries explicit CC0 or attribution licenses. For risk-averse commercial use, Pexels route or original ComfyUI generations with documented prompts provide strongest legal footing.

Can I modify the pipeline for long-form content?

The architecture supports extension. Adjust word_count in niche profiles, modify ffmpeg assembly templates for 16:9 or horizontal formats, and extend upload targets. Community forks have adapted for podcast clip generation and educational course modules.

How do I contribute or report bugs?

The repository accepts pull requests. Key contribution areas: additional niche profiles, new provider integrations (especially emerging TTS models), and platform upload extensions. Issues are tracked on GitHub with responsive maintainer engagement.

Is there a hosted version if I don't want to self-host?

Yes—verticals.gg offers the same pipeline as a managed service with web interface, eliminating setup complexity for non-technical creators.

Conclusion

The creator economy has a dirty secret: the tools that promise democratization often re-create gatekeeping through subscription pricing and opaque algorithms. Verticals v3 genuinely inverts this model. Open source. Self-hostable. Provider-agnostic. Cost-transparent down to the penny per video.

What Rushindra Sinha has built isn't merely a video generator—it's a content operating system that encodes the tribal knowledge of successful creators into reproducible, customizable automation. The niche intelligence concept alone represents a generational leap beyond template-based tools.

Is it perfect? No. The free tier requires local GPU capability. The Gradio UI is functional, not beautiful. Multi-platform upload beyond YouTube awaits v3.1. But the foundation is architecturally sound, the cost structure is unbeatable, and the customization ceiling is essentially infinite.

If you're a developer-creator who believes that automation should amplify judgment rather than replace it, this tool belongs in your arsenal. Start with the free tier. Iterate on niche profiles. Scale what works.

Grab the code, run your first pipeline, and join the small but growing community of builders who've stopped editing manually.

The future of content isn't more editors. It's better systems.

Built by Dr. Rushindra Sinha. Follow development on X @irushi or explore the hosted version at verticals.gg.