NanoPhotoStudio: a postmortem

December 2, 2025 to June 2, 2026. Autopsy written September 22, 2026.

What it was

An AI portrait studio. You picked a style, uploaded one selfie, and got a studio-quality portrait in a few minutes: cinematic, vintage, monochrome, editorial. Built solo, as a practice project wearing production clothes.

The numbers

Time invested: 6 months of nights and weekends
Commits: 1,648 · merged pull requests: 282, all through CI
Subscription revenue over 2 months of a $4.99 plan: $0 (0 conversions)
Visitors in the last measured week: 27
Total revenue: $0
Total spent on advertising: $0
358
562
400
140
116
70
2
Dec
Jan
Feb
Mar
Apr
May
Jun

Commits per month. The curve of a project dying is exactly as smooth as you would expect.

How it died

I built for 20 weeks before I tried to sell anything. Edge rendering on Cloudflare Workers, async generation workflows, a payment abstraction, an admin dashboard, a style catalog, 4K upscaling, magic-link auth. Every piece of it worked. None of it mattered, because nobody came.

The market never had a gap. HeadshotPro had 196,000 customers. Aragon claimed about 2 million users. BetterPic raised $2.5M. Google's Nano Banana generates a decent portrait from one selfie, free. My $1-per-portrait price read as a trial price, because that is what this market taught people one dollar means.

In April I finally built the thing I should have built first: a weekly scoreboard. It said 27 visitors a week, 0 clicks on the main call to action, 0 paid orders. I had redesigned pricing three times for that audience. The data was a verdict, and I kept shipping CSS.

The sharpest embarrassment: in March I designed a $10/day Meta ads experiment. Three angles, six creatives, a complete production checklist, an ROAS spreadsheet. I launched none of it. Spending money felt like risk. Redesigning a navigation drawer felt like progress. That sentence is the whole disease.

What I learned

What survived

A production-grade SaaS template (Next.js on Vercel, Cloudflare Workers, D1, R2, Workflows, full test suites, CI). The weekly scoreboard SQL, which now runs in another project. And a standing rule for my next venture: the first 10 hours go to distribution, not to architecture.