The notes-doc that became a website

It started, like a lot of personal projects do, with a text file.

Sometime in early 2025 I was using Midjourney for a small client illustration job, and a Suno track for a friend's birthday video, and a quick Flux generation for a presentation slide, and I noticed something inconvenient: every time I sat down with a new model, I had to relearn the same thing. Not the underlying tech — the prompt grammar. The little rituals each tool wanted you to perform to get a result that didn't look like a stock-image fever dream.

So I started keeping a notes file. Just a flat list of prompts that had actually worked, with a sentence each about *why*. After a few weeks I noticed I was opening that file more often than I was opening the tools' own galleries. After a few months I noticed I was sending it to friends. After a few more months a friend asked, "Why isn't this a website?"

Honest answer at the time: because I'm a software engineer, and software engineers know that the gap between "useful private notes" and "useful public website" is mostly several hundred unglamorous hours of plumbing — auth, search, mobile layout, caching, copy-editing, accessibility, error handling, the long tail of edge cases you didn't think about until someone with a different browser hits one. I had a day job. I didn't want to do hundreds of hours of plumbing for free.

But the question stuck. Because the more I looked, the more I noticed: the tools that *did* exist for prompt curation were either thin (a list of 30 prompts, never updated) or sprawling (a marketplace with no curation at all, which is the same problem in a fancier suit). The thing I actually wanted — a working library, where every entry had been run through the model it claimed to support, and the variables were exposed so you could adapt them to your own situation — wasn't there.

So I built it.

The first version of PromptCraft was, generously, not good. The prompt count was small. The categories were arbitrary. The English mirror was machine-translated. The mobile layout broke if you held your phone slightly wrong. But the spine was right: every entry had a sample, a context note, and a list of variables. You weren't getting "100,000 prompts" of which 30 were actually any good — you were getting the smaller library where I'd already done the filtering for you.

That was nine months ago. The site is bigger now. The English mirror is hand-edited. There are seven free generators that compose prompts on the fly for the major models. The library has 180+ entries across image, video, music, and text models, every one of which I've personally run.

It is still, fundamentally, the same notes file. The thing I built for myself. Just with the lights on, and the door open, and a coat of paint that doesn't fall off when you touch it.

That's the only reason a one-person studio gets to ship anything. You build the thing you actually use, you keep using it, you fix what breaks, and at some point you notice it's good enough that other people can use it too. There's no shortcut around the using-it-yourself part. If you don't, you build the thing you imagined people would want, which is almost always wrong in ways you can't see until it's too late.

Anyway. That's how PromptCraft happened. There's no big origin story. Just a notes file that wouldn't shut up.