The photo that was almost complete
kevinbaiThere's a group photo from a trip a few months ago that I really liked. Good light, everyone relaxed, the kind of shot that actually captures the feeling of the day. Except one person was missing — she'd had to leave early, and by the time we took the photo she was already on her way home.
I tried to fix it the obvious way. Opened Photoshop, found a photo of her from earlier that day, and spent about twenty minutes trying to cut her out and place her in. The edges were terrible. The lighting didn't match. I closed the file without saving.
What stayed with me wasn't the frustration — it was the thought that this should be a solved problem by now.
---
A few days later I started looking around to see how many people had the same experience. Not looking for anything specific, just curious. What I found was that this kind of problem shows up everywhere: someone's grandmother missed the family reunion, a kid blinked in the only good shot, two photos from the same trip that would look great as one but don't quite fit together.
People have been asking these questions online for years. The answers are almost always "use Photoshop" or "hire someone." Neither of which is a real answer for a casual photo you just want to look right.
That felt like enough reason to try building something.
---
The result is AI Image Combiner. You upload a few photos, describe what you're going for in plain text, and the AI handles the blending — light, shadows, proportions. No layers, no masking, no manual color correction.
The hardest part to get right was making the result look like it was always one photo, not two images sitting next to each other. That gap between "technically merged" and "actually convincing" is where most of the work went.
---
I don't know yet how far this will go. But it started with one photo that was almost complete, and the feeling that making it whole shouldn't require twenty minutes of fighting with software you barely know.
Maybe that's enough of a reason to build something.