When Every “AI Headshot” Looked Fake, I Spent 2 Weeks Hacking Together My Own

I built the first MVP of this AI photo app in about two weeks of late nights and weekends, because after trying everything on the market, nothing looked “real” enough.

I’m not a photographer. I’m a solo indie maker who’s just obsessed with those “magazine feel” photos — golden hour light, candid street shots, images that look casual but somehow still look great.

Last year I wanted a few decent photos of myself. Not selfies — real photos. The kind you’d actually feel okay putting on a dating profile or LinkedIn. Hiring a photographer for a 2‑hour shoot would easily cost $300–500, and I’m pretty awkward in front of a camera anyway.

So I tried every AI photo tool I could find. Midjourney, all the headshot generators, everything. And honestly, they all had that obvious “AI look”: over-smoothed skin, flat lighting, backgrounds that feel like a computer drew them from a text prompt. You can tell instantly it’s not a real photo. You’d never fool anyone.

That started to really bother me — more than it probably should have.

I kept thinking: why can’t AI just make photos that actually look like they were shot with a real camera? Not paintings. Not illustrations. Not “AI art.” Just… photos. With real lens blur, natural skin texture, believable shadows. The kind of image where nobody’s first reaction is, “Wait, is this AI?”

I looked around and realized almost nobody was really focused on that specific problem. There are tons of AI image generators, but most are chasing “creative art” and stylized outputs. Very few seem to genuinely care about making something that passes as a real photograph taken in a real place.

So I decided to build it myself. I had no idea how hard it would be, but I started anyway.

Over two pretty intense weeks of late nights and weekends, I hacked together the first MVP of CozAIPhoto — a tiny AI photo “studio” focused on camera‑realistic portraits instead of generic AI images. You upload a few clear reference photos of yourself, pick a style (street, studio, travel, dating profile, event scenes, etc.), and it generates a batch of photos that look like someone actually took them with a camera.

The hardest part wasn’t the model architecture — it was getting the “photography feel” right. Real photos have a lot of subtle details: how light wraps around a face, how the background falls off into bokeh, the color and contrast you get from different cameras and lenses. I spent a lot of time just going through photographers’ portfolios, trying to understand what makes a photo feel “shot” instead of “generated.”

Of course, this MVP is far from perfect. Sometimes the hands are still weird (classic AI problem), sometimes the lighting doesn’t fully match the scene. But in many cases, when I show people the results, their first reaction isn’t “this looks AI” — it’s “who took these for you?” That’s basically the outcome I was aiming for.

Right now, CozAIPhoto supports styles like travel shots, street portraits, dating profile photos, studio looks, and event scenes. It can generate up to 4K resolution, and the image quality at that size has been better than I expected for an early version.

I’m sharing this project because I know a lot of people here care about both product and aesthetics. I started it for a very simple reason: I just wanted a few personal photos that didn’t scream “AI-generated.” If that resonates with you — especially if you’re a photographer or someone who really cares about images — I’d genuinely love your feedback: do these photos pass your eye test, and what would you want me to improve next?