I got 1,200 YouTube creators to use my free AI thumbnail tool — here's every mistake I made building it

Eight months ago I had a simple theory: thumbnail CTR is the most underrated lever on YouTube, and nobody has built a good tool to move it.

I was right about the problem. Almost everything else I got wrong.

Here's the honest story of building Miniagen.

— Mistake 1: I built for quality, not for testing —

My first version generated one "perfect" thumbnail per prompt. Users hated it. Not because the output was bad — because one output is useless. You can't learn anything from a single data point. Creators need variations to A/B test. I rebuilt the entire generation pipeline around producing 6 genuinely different compositions per prompt. Signups doubled the week I shipped it.

— Mistake 2: I gated the product behind a signup —

Classic mistake. I was scared of abuse. Turns out, the people who generate a thumbnail before signing up convert at 3x the rate of people who sign up first. The product needs to earn trust before asking for an email. Removed the gate. Activation rate jumped.

— Mistake 3: I ignored visual consistency —

Users kept coming back with the same complaint: "My second thumbnail looks nothing like my first." They weren't asking for better quality — they were asking for a coherent channel identity. That's why we built the avatar system: create your look once, every generation stays on-brand.

— Where we are —

1,200+ creators. Free to start, no credit card. Still improving the generation quality and the avatar consistency across sessions.

The one thing I'd do differently: talk to 20 YouTubers before writing a single line of code. The problem I thought I was solving and the problem they actually had were related but not identical.

Happy to answer questions — especially if you're building for creators.

Free to try → miniagen.co