Odhanan said:
Crothian's got it right: you begin with whatever feels right to you. That's a bit like singing: you've got to find your voice, and after a while, you'll have your own.
As a matter of example, I can speak of the way I do it. Usually, it starts with an idea. Like... Atlantis just before its fall, for instance. I have a start of background in my head.
With all due respect, I really
would not start any way you feel like. I think your second paragraph is a much better way to start.
For instance, if you start with the detail (as most budding worldbuilders do), you are apt to wander off in directions that may be fascinating to you but of little use to your campaign. Moreover you may miss a chance to embue your world with a cool theme or concept.
If you start off with the idea as you suggest then you have a core concept to build on, sort of a guiding principle to use. This guiding principle not only is likely to make the setting more memorable but when you are sputtering for ideas, and we all do at times, you can come back to it and mine it for more starting points.
I've made many settings in my time (I tend to do one for each new campaign every couple years or so). The ones that I and the players remember are the ones that worked from a key concept.
I'd do a true brainstorm: true in the sense that you really write down any idea that pops into your head, including ones that you probably would never use as you never know what concept would lead to the really useful ones. Kick it around until you have something that feels really good, maybe weeks if that is what it takes because your core concept is going to define the whole world and you are going to spend many, many hours creating your world. I would also recommend not being afraid about tying your world to a campaign plot, the two together can build top-class gaming experiences. You can brain storm the campaign plot together with the world concept or separately.
Once you have the basic concept from your brainstorming, you can then start fleshing out the world. usually, with foundation in hand, knowing what to work on next will be a lot easier.
On the brainstorming, if you have a creative player or two you could brainstorm the concept with them, brainstorming is a lot more effective in groups because what one person tosses into the mix may get you thinking on a whole new, profitable track. Of course, I wouldn't brainstorm plot ideas with your players for obvious reasons
As an abbreviated example it might go like:
Lord of the rings
underworld
Thirteenth Warrior [got a movie theme going in your head. That's okay, mine it until something else comes up]
Mask of zorro [Antonio bandarras link no doubt, may be useless but put it out there,could go somewhere]
Desparado
Fighting against the corrupt powerful?
mage corruption?
undead good guys [underworld echo?]
World where corrupt wizard kings rule opposed by good-guy vampires [getting a bit more specific but still brainstorming]
World where corrupt wizard kings rule opposed by good-guy lycanthropes? [minor refinement but avoid undead PCs which many players will balk at]
And so on. this is much truncated and you probably want to do this on a big sheet of paper so you can go 2D. Big whiteboard can be nice for this.
But let's say you settle on the last concept. Now the next steps in world creation are obvious:
* need to create a history that leads to the wizard-kings and the underground lycans
* need to flesh out your lycans. Myabe alter the mechanics, add new types
* need to figure out how big the world is and how much time you are going to spend on maps, kingdoms etc. (make it too big and you might feel compelled to spend too much time on making lots of kingdoms. is that necessary? Do you see the game moving into many kingdoms? This could be a rich setting with just one kingdom or a handful)
* need to determine what races are in the world, what monsters. May not need much other than humans and lycans in this sort of world
and so on...
One thing I would watch out for is getting bogged down in minor detail. Yeah, it might be kind of cool writing up a three page history of the halfings. If your campaign relies on halfings great. If none of your players will ever play a halfing and you don't really intend to do much with them in the game, is that really a good use of your time?
This type of deep-detail dive is a symptom of a worldbuilder avoiding the hard problems. It is easy to take refuge in a detail like this but it isn't getting your campaign closer to a start. It is a lot like the budding novelist writing out 50 page character backgrounds and not actually working on the harder part of the problem: his plot.

Having a firm world concept (and hopefully some idea of what type of game you will run, a plot is even better) will help you stay focused on relevant details.