I built a text-to-video AI in 30 days.

Here’s the stack, the bills, and why my initial launch flopped.
Long-time lurker, first-time poster. 👋

I’ve spent the last month building Textideo, a tool that turns text scripts into ready-to-post videos.

In the spirit of Indie Hackers, I’m not here to pitch you. Instead, I want to share the unpolished reality of building an AI product in 2024—specifically the tech choices, the surprisingly high server costs, and how I completely messed up my first launch.

😫 The Itch: The "Content Treadmill"

I’m a developer, not a video editor. But to market my previous side projects, I needed video content.

I found myself spending 4 hours editing a 30-second clip just to post it on Twitter. The friction was insane. I had folders full of blog posts and notes that were dying on my hard drive because I didn't have the energy to visualize them.

I wanted a tool where I could dump text and get a decent base video out. Since I couldn't find one that fit my budget/workflow, I decided to build it.

🛠 The Stack (Keeping it lean)

I time-boxed the MVP build to 30 days to avoid feature creep.

• Frontend: Next.js + Tailwind (Hosted on Vercel)
• Backend: Python/FastAPI (Needed for better library support with media processing)
• The "Brain": OpenAI GPT-4o API (For script analysis and scene splitting)
• The Visuals: [Stable Diffusion / Pexels API / Stock Footage API]
• Video Assembly: FFmpeg (The true hero of this project, running on a separate worker)
• Database: Supabase
• Payments: Stripe

The Technical Nightmare:
Handling video rendering in the cloud is painful. My first few renders took 10+ minutes for a 30-second clip because I was processing everything synchronously. I had to rewrite the backend to use a job queue (Redis + Celery/BullMQ), which cut the user's perceived wait time significantly.

💸 The Bills (The "AI Tax")

This is something people don't talk about enough. AI wrappers are expensive to run.

• OpenAI API costs: ~$0.03 per video script.
• Image Generation/Stock API: ~$0.10 - $0.20 per video.
• GPU/Worker Server: ~$40/month (fixed cost).

My margins are thinner than a traditional SaaS. This forced me to rethink my pricing model immediately—I couldn't offer a generous "Unlimited Free Tier" without going bankrupt.

📉 The Launch & The Flop

I launched on [Twitter/Product Hunt] thinking, "This solves a real problem, people will flock to it."

Result:

• Visitors: 400
• Signups: 15
• Paying users: 0

Why? I positioned it as a generic "Text to Video" tool.
It turns out, nobody wakes up looking for a "general video tool." They look for "How to make TikToks without showing my face" or "How to turn blog posts into YouTube Shorts."

🔄 The Pivot

I stopped coding and started DMing. I reached out to content writers and newsletter authors specifically. I offered to manually convert one of their articles into a video using my tool.

This "Do things that don't scale" approach worked.

• I got direct feedback on the video pacing.
• I found my first paying customer after the 5th manual demo.

🧠 Lessons for other Indie Hackers

1. UX > AI Models: Users don't care if you use Llama 3 or GPT-4. They care if the video export fails. Stability is your best feature.
2. Niche Down Hard: The "AI Video" space is crowded by giants (Runway, Sora, HeyGen). As an indie, you can't compete on quality; you have to compete on workflow specific to a niche.
3. Watch your margins: If you are building GenAI tools, calculate your unit economics on Day 1.

🤔 I need your help

I’m currently struggling with the pricing model. I'm debating between:

1. Credit System: Pay-as-you-go (Good for margins, bad for MRR).
2. Monthly Sub: Predictable revenue, but power users might kill my API budget.

If you’ve built an API-heavy wrapper, how do you handle this?

I’ll be hanging out in the comments all day. Feel free to roast the landing page (link in the first comment) or ask about the FFmpeg implementation!