AI Content Factory: Build an End-to-End Automation Pipeline — From Real Work to 14+ Platforms in TH + EN

Read time: 12 min | Updated: February 28, 2026

I started this project to answer one question: what does a solo founder's content output look like if you treat it like a software system — not a creative process? The answer surprised me. Not just the volume, but how little of it I actually have to touch.

Why does a solo founder need an AI Content Pipeline?

Content is distribution. And distribution, for a solo founder without a marketing team, is the difference between building in public and building in obscurity.

The math is brutal. A mid-size startup content team — 3 writers, 1 video editor, 1 social manager, 1 ads manager — runs you $8,500+ per month. Easy. That's before tools, stock assets, or the endless Slack threads about "brand voice."

I chose a different path. Not by working more hours — by building a pipeline that runs while I sleep. The AI Content Factory is a 9-stage end-to-end automation system. Feed it 2-3 raw inputs per day — a voice note, a screenshot, a code snippet — and it generates 14-21+ finished content pieces across platforms. Blog posts, short-form video, carousels, ads, translated variants. All of it. At ~$70/month.

How does the 9-Stage Pipeline work?

Each stage feeds the next. Raw work becomes research. Research becomes scripts. Scripts become video. Video becomes clips, carousels, ads, and blog posts — each repurposed for the platform it lives on. Every stage is automated via n8n (self-hosted workflow automation, free).

Let's break down each stage.

Stages 1-3: How does raw work become slides and scripts?

Stage 1: CAPTURE — This is the "Human Signal." I don't sit down to create content from nothing. I work. Code in Cursor AI, trade in MT5, build websites with Next.js. Everything I do becomes raw content. I just note what I did, what problems I hit, how I solved them — then n8n's webhook picks it up automatically.

Stage 2: RESEARCH — NotebookLM does deep research from raw content + relevant docs. OpenCrawl scrapes competitor data and trending discussions. Claude via OpenRouter synthesizes everything into a structured Content Brief. This stage takes under 5 minutes. Doing it manually? At least an hour.

Stage 3: CREATE — Claude generates the slide structure from the Content Brief. python-pptx converts JSON to .pptx using a Brand Template (red-black-gray-white). Then Claude writes the narration script for each slide — in both Thai and English. The English version isn't translated — it's re-written. Different hooks, global analogies, USD instead of THB.

Stage 4: Voice Clone + VDO Production — does it actually work?

Yes. And the quality gap between a cloned voice and a professional recording is now smaller than the gap between a $20/hour and $200/hour editor.

MiniMax handles voice cloning. I trained it on 15 minutes of clean audio — once. Every new script is read aloud by a voice model that sounds like me. FFmpeg does video assembly: slides + voice → timed video with transitions. Whisper (self-hosted) generates subtitles automatically, then FFmpeg burns them in.

Processing time for one finished video with subtitles and 3 aspect ratio variants (16:9, 9:16, 1:1): ~18 minutes. Unattended. A human video editor would take 2-4 hours and charge $50-100 per piece.

Manual vs AI Pipeline — what's the real cost difference?

I tested both approaches. Outsourcing to a team of freelancers: blog writer $500/piece, video editor $400/piece, translator $200/piece. Cost per content piece distributed across all channels = ~$1,700. Monthly for 20 pieces = $8,500+.

AI Pipeline? OpenRouter API ~$12/month + MiniMax $8/month + VPS $25/month + misc $25/month = $70/month total. That's a 99% cost reduction. Same volume. Better consistency.

Auto-posting to 14+ platforms — how does distribution work?

n8n is the backbone. After content is produced — video, blog, images — a distribution workflow fires. Claude generates platform-specific captions (LinkedIn = professional ≠ TikTok = fun ≠ X = concise) → n8n parallel-posts to every API → logs all post IDs to the database for performance tracking.

Every content goes out in 2 languages as separate posts. Not translated — re-written with native DNA. English captions get global references and USD. Thai captions get local analogies and THB. All automated by Claude.

What's the full Tech Stack?

The stack is organized into 6 layers. Each has a primary tool and a fallback. Nothing requires vendor lock-in — every component can be swapped. OpenRouter is the gateway to all AI models through a single API — switch models instantly without code changes. 80% of the stack is self-hosted for full control and low cost.

What does the daily workflow look like?

I tracked my daily workflow for 3 weeks. Here's the pattern that works: morning = real work, noon = pipeline runs, afternoon = auto-post, evening = ads optimize, midnight = analytics. Everything scheduled via n8n cron — I don't touch the trigger.

Want to build this — where do you start?

Don't build all 9 stages at once. The indie hacker graveyard is full of over-engineered systems that never shipped.

Week 1-2: MVP — Set up n8n, connect OpenRouter API, create Brand Template, build auto-post pipeline to FB + YouTube + LinkedIn. Just these three platforms will show results.

Week 3-4: Full Pipeline — Add IG, X, TikTok, Medium, Dev.to. Add VDO production (FFmpeg + Whisper). Add EN re-write. Add Ads automation.

Month 2: Optimize — Add Grafana dashboard, AI topic recommendations, A/B test content formats, Substack newsletter.

Month 3-6: SaaS — Package the entire system as a multi-tenant SaaS for other founders.

Try it: Prompt for Content Brief Generation

Here are the actual prompts I use in production — copy and use them directly.

Prompt 1: Content Brief Generator (Stage 2)

Use with: Claude / ChatGPT | Level: Intermediate

You are a senior Content Strategist for a technical founder's personal brand.

From this raw input:
"""
{{raw_input}}
"""

Generate a Content Brief in JSON format:
{
  "core_insight": "Single most interesting thing in these notes (1 sentence)",
  "title_options": ["3 title variants with numbers"],
  "hook": "Opening sentence that stops the scroll (under 20 words)",
  "key_arguments": ["3-5 main points, in logical order"],
  "counterargument": "Strongest objection a skeptic would raise + how to address it",
  "target_audience": "Specific audience (e.g., 'developers building SaaS' not 'tech people')",
  "keyword_clusters": ["5-8 search terms"],
  "platform_priority": {"youtube": "long-form", "linkedin": "carousel", "x": "thread"},
  "stats": ["3-5 specific numbers/statistics"]
}

Rules:
- Titles must contain numbers
- Hook must be under 20 words
- Stats must be specific, not generic
- key_arguments must be actionable

--- Variables ---
{{raw_input}} = Your daily work notes, voice transcripts, or screenshots

Why it works: JSON output format lets n8n parse it instantly and feed it into the next stage. The "titles must have numbers" + "stats must be specific" rules eliminate 80% of generic AI output.

Prompt 2: n8n Workflow Architecture (Stage 3)

Use with: Claude | Level: Advanced

You are an n8n Workflow Architect specializing in automation pipelines.

Design an n8n workflow for {{task}} that:
1. Receives input from {{trigger}} (webhook/cron/manual)
2. Processes through {{stages}} stages
3. Outputs to {{destinations}}

For each node specify:
- Node type (HTTP Request, Function, IF, Switch, etc.)
- Key configuration
- Error handling strategy
- Estimated processing time

Rules:
- Minimize node count
- Error notification on every critical path
- Retry logic for API calls that may fail
- Log every step to Supabase

--- Variables ---
{{task}} = Content Brief → auto-post all platforms
{{trigger}} = Webhook from Cursor AI / Voice Note app
{{stages}} = Research → Create → Produce → Distribute
{{destinations}} = FB, YT, LinkedIn, IG, X, idea2logic.com

End Goal: Why productize this as SaaS?

This pipeline isn't designed for me alone — I built it to be packaged as SaaS from day one. Every component is modular, every config is a variable, every workflow is exportable.

Gartner predicts that by 2027, 80% of organizations will use generative AI in content production. But most solutions are enterprise-priced ($500+/month to start). This pipeline targets solo founders at an accessible price — that's the gap no one is filling yet.

FAQ — Frequently Asked Questions

Q: Do I need programming skills to build this pipeline?

A: n8n is a visual workflow builder — drag and drop, no coding required. But basic JavaScript/Python knowledge helps you customize deeper (e.g., custom MiniMax API nodes or FFmpeg commands). Phase 1 (MVP) can be built with zero code.

Q: What's the actual monthly cost breakdown?

A: Self-hosted: OpenRouter API ~$12/mo (for 60+ pieces/month), MiniMax Voice $8/mo, VPS (4 vCPU, 8GB RAM) $25/mo, other APIs ~$25/mo = total ~$70/month. Cloud services instead of self-hosted bumps it to ~$150-250/month.

Q: How realistic is the voice clone quality?

A: MiniMax needs 30 sec - 2 min of your real voice. Results: 90-95% similarity in the source language, 85-90% in the second language. Weakness: intonation sometimes sounds "flat" — fixable with SSML markup that Claude generates automatically.

Q: What business types does this work for?

A: Best for: freelancers/consultants building personal brands, SaaS startups doing content-led growth, SMEs with limited budget but omnichannel ambitions. Not ideal for: luxury brands requiring 100% human-crafted premium content.

Q: What about platform policies on AI-generated content?

A: As of 2026, major platforms (FB, YT, LinkedIn, X) don't ban AI-generated content — but some require labeling. This pipeline supports a "Human-in-the-Loop" mode: review before posting, or run 100% auto. Your call.

The final thought: this pipeline didn't make me "produce more content." It freed me to think about what actually matters. The 4-6 hours per day I used to spend on content production? Now it's 30 minutes of raw input — then AI handles the rest.

Time reclaimed? Building new products, talking to customers, learning new things — all of which become raw input for the pipeline again.

The best loop is the one that runs itself.