CloudMade have been offering this based on top of OpenStreetMap for quite a while (over a year I think) - I imagine the competition there was one of the principle drivers behind adding this feature to Google Maps. There was a panel on mapping at SxSW 2009 with representatives from OpenStreetMap and Google Maps and the OSM presentations were much, much more impressive thanks to this kind of feature. I'm not surprised Google felt the need to catch up.<p>If you want complete control over your maps, check out Tile Drawer - <a href="http://tiledrawer.com/" rel="nofollow">http://tiledrawer.com/</a> - it lets you define your own map styles using a Cascadenik stylesheet, then run your own EC2 instance that will render and serve custom tiles for a specific area using those custom styles (and OpenStreetMap data).
It's been interesting to watch the Google Maps API as it rode from "The Coolest thing EVER" in 2004 to the non-issue status it holds today.<p>Back in the day, companies were getting tens of millions in funding to build apps where users could place pins on a map. That says a lot about the hype surrounding it, since that's essentially the app you got when you pasted their sample code into your IDE and hooked it up to a database.<p>Today, it's just expected that any application you build will have a GMaps implementation somewhere in it, but it's just old-hat. I actually run one of those GMaps-based startups from back in the day, and I haven't kept up on the latest API changes for years.
I've been looking for a no-impact Javascript-API black-and-white map ever since I started putting heat maps on them, and this is (typically for the Maps API team) way more powerful. My favorite announcement of the week for my startup.
Kind of cool, but I think it's actually a mistake to customize an important map much - showing a person a google map lets them know exactly what it is immediately, and they can jump right in. Showing them a Google map that doesn't look like a Google map seems like you could make minor gains in usability in one way, but you're paying a large learnability penalty in the other. Overall, if it's a business project, you'd better have a damn good reason for it.