Questions

What question was I exploring?

After working with some per-town open datasets published by the State of MA, I thought it'd be neat to see how I might use similar AI-assisted scraping to do something interesting with per-town data. This coincided with an exploration of custom-built image gen capabilities via models like Google's gemini-flash-2.5-image.

What did I learn?

Like Artifact 001, I realized that fine tuning a default image gen model is quite easy programmatically. It is the prompting, however, that is tricky. I don't feel that there is an established prompting language-to-image gen standard for any image gen models as of now.

What feels unresolved?

Per a quirky Gemini suggestion, I added some specific flag-gen modes. The prompts for those are not quite there yet. Beyond this, making this a public repository of crowd-sourced flag gens could be a fun social project for people.