DEV Community

tamay erdogdu
tamay erdogdu

Posted on

Why I Stopped Paying for Video Editors and Built an AI Factory Instead

#ai #automation #python #tutorial

Why I Stopped Paying for Video Editors and Built an AI Factory Instead

I’ve always been that person who drafts a script at 2 a.m., scrolls through royalty‑free footage, and then spends a small fortune on a freelance video editor to stitch it all together. It worked … until the invoices started adding up and my side‑project budget turned into a monthly headache. That’s when I decided to test the hype around AI video automation and see if I could replace the whole “pay‑someone‑else” pipeline with a single n8n workflow I could control.

Below is a candid, week‑by‑week diary of how I turned a $150‑a‑month expense into a $20 one‑time investment, the bumps I hit along the way, and why I now recommend building your own AI Shorts factory for anyone chasing passive income AI streams.

Week 1 – The “I’m Skeptical” Moment

I started with a simple question: could an AI write a concise script, find relevant images, generate a voiceover, and splice everything into a short video without any human touch? My first test was a 60‑second tutorial on “how to clean a coffee maker.” I wrote the script in Notion, copied it into a free text‑to‑speech tool, and manually downloaded a few stock clips. It took me 3 hours to produce a 30‑second clip, and the result looked… robotic.

The content automation promise sounded great on paper, but the reality was a mess of mismatched audio timing and low‑resolution stock footage. I was convinced that I’d need a professional editor to clean it up. That’s when I started digging into n8n, an open‑source workflow automation platform that lets you stitch together APIs without writing a full‑blown app.

Week 2 – First Steps with an n8n Workflow

I set up a local n8n instance on my laptop (thanks, Docker) and began building a tiny pipeline:

  1. Trigger – a new markdown file in a GitHub repo.
  2. OpenAI – generate a 150‑word script from a headline.
  3. Pexels API – fetch 3‑5 relevant images.
  4. Google Text‑to‑Speech – create a voiceover.
  5. FFmpeg node – merge audio and images into a video.

The whole thing ran in about 45 seconds. I felt like a kid who just discovered a new LEGO set. My first output was a 10‑second “AI Shorts” clip about “why cats love cardboard.” It wasn’t perfect—some image transitions were abrupt, and the voice sounded a bit flat—but it proved the concept.

Week 3 – The First Setback: API Limits and Timing Issues

Excited, I tried scaling up to a 30‑second tutorial series. That’s when I hit two roadblocks:

  • API rate limits – Pexels throttled my image requests after the 5th call, and the Google TTS quota ran out after a few minutes of testing.
  • Audio‑image sync – My FFmpeg node was using a static frame duration, causing the voiceover to finish before the last image faded out.

I spent a full day digging through the n8n docs and community forums. The solution? Add a rate‑limit node to throttle requests and use a dynamic duration calculation based on the length of each text segment. It added a few extra nodes, but the workflow became more robust.

Week 4 – Adding Automated Video Production and Scheduling

With the technical kinks ironed out, I introduced two more pieces:

  • Automated video production – I swapped FFmpeg for a cloud‑based video rendering service (Shotstack) that handles transitions, subtitles, and branding overlays automatically.
  • Auto‑posting – Using the YouTube, TikTok, and Instagram APIs, I set the workflow to upload the final video to each platform as soon as it’s rendered.

The whole pipeline now looked like this:

GitHub → OpenAI (script) → Pexels (images) → Google TTS (voice) → Shotstack (render) → YouTube/TikTok/IG (post)
Enter fullscreen mode Exit fullscreen mode

I scheduled three videos per week, and the first batch went live without any manual steps. The engagement numbers were modest (a few hundred views each), but the passive income AI potential was evident—no more paying $30‑$50 per video edit.

Week 5 – Fine‑Tuning and Real‑World Results

I spent the week tweaking two main aspects:

  1. Script quality – Prompt engineering for OpenAI made the scripts sound more human. Adding a “tone: friendly, conversational” parameter boosted watch time by ~15 %.
  2. Thumbnail generation – I added an additional node that pulls the most eye‑catching frame and overlays bold text using the Canva API. Click‑through rates rose from 1.2 % to 2.8 %.

After 30 days, my AI Shorts channel had produced 12 videos, each averaging 250 views. The total cost? $20 one‑time for the n8n workflow (I packaged it as a Gumroad product). No recurring freelancer fees, no licensing headaches, and I’ve reclaimed roughly 12 hours of my month.

Week 6 – The “What I Wish I Knew” Moment

If I could go back, I’d start with a template n8n workflow instead of building from scratch. Also, I’d set up proper monitoring (via n8n’s built‑in error handling) earlier; the first time a TikTok upload failed, I didn’t notice until a day later, missing the optimal posting window.

Nevertheless, the journey taught me that content automation isn’t a magic button—there’s still a learning curve and occasional manual oversight. But the payoff, especially for side‑hustlers seeking passive income AI, is huge.

Final Thoughts – Build Your Own AI Video Factory

I’m still experimenting with longer formats and adding a quick transcription node for subtitles, but the core workflow is solid. If you’re tired of paying for video editors and want a reproducible, low‑cost solution, I highly recommend giving an n8n workflow a try.

The tool I'm using is called AI Shorts Factory (https://8622430312019.gumroad.com/l/gujqfy) — it's an n8n workflow that costs $20 one-time and handles everything: AI script generation, image search, voiceover, video production, and auto‑posting to YouTube, TikTok, and Instagram.

Happy automating! 🚀

Top comments (0)