I wanted one thing: a fully automated, zero-cost pipeline that could turn an idea into a YouTube Short, start to finish, using nothing but free tools and an AI coding assistant. No stock footage subscriptions, no editing software, no manual screen recording. Just a script in, a video out.
I got closer than I expected. Then I hit a wall that had nothing to do with code.
What actually got built
Over one long session, the pipeline came together piece by piece:
- Text-to-speech via Google Cloud's free tier (WaveNet/Neural2 voices, a few million free characters a month — far more than a couple of Shorts a week would ever use)
- Visuals generated entirely in code (Python + Pillow): chat-bubble mockups, file-browser mockups, "results" grids, all procedurally drawn, no stock assets
- Assembly via ffmpeg, stitching narration, captions, and generated slides into a finished vertical video
- One orchestration script tying it all together, so a new video was just: drop in a script file, run one command
Along the way there were real, unglamorous obstacles — a newly-split Korean administrative district that Google's billing system didn't recognize yet, a service-account-key policy that blocked the "obvious" authentication path (fixed by switching to OAuth-based application-default credentials instead). Nothing fatal, just the normal texture of actually shipping something.
Where it got interesting
The slideshow-style visuals worked, but they looked exactly like what they were: slides. Meanwhile, a very different kind of AI content was clearly having a moment — short, photorealistic AI-generated video clips (animals doing improbable things, mostly) racking up millions of views. The tools behind them — Google's Veo (via Flow), Kling AI, Hailuo/MiniMax — all advertise usable free tiers.
So the obvious next move was: swap the static slides for actual AI-generated video clips.
Where it stalled
Checking the free tiers properly turned up real limits, not just marketing copy:
- Free-tier clips capped at around 8 seconds
- A visible "Made with Veo" watermark on free exports
- Kling and Hailuo's free tiers explicitly exclude commercial use in their terms
The commercial-use question mattered most, since the whole point was a monetized channel. I went looking for a clear answer in the actual terms of service rather than trusting summary blog posts (several of which flatly contradicted each other on the watermark question alone). I couldn't find an unambiguous yes. Google's own policy pages didn't spell out whether free-tier output could be used in monetized YouTube content; third-party writeups disagreed with each other.
That's not a legal opinion, and I'm not a lawyer — it's just an observation: when the terms are that unclear, building a monetization plan on top of them felt like the wrong foundation, however good the free credits looked on paper.
So, this instead
Rather than keep gambling on shifting terms of service, I pivoted to writing about the build itself — which sidesteps the whole problem. It's fully original, the "content" is just documenting what actually happened, and there's no ambiguity about whether I'm allowed to publish it.
The pipeline still exists and still works, free-tier TTS and all. It's just B-roll for a different kind of project now.
If you're building something similar: check the actual terms before you build a business plan on a "free" AI tool, not just the blog posts telling you it's free. The credits are real. The fine print sometimes isn't what you'd hope.
Top comments (0)