Canonical just released a Python-based ORM called Storm.
Mostly it was developed by Gustavo Niemeyer, but since we're in the same team at Canonical and it's developed primarily for our project, I have contributed bits and pieces.
Communal ties will of course require me to mention Storm in relation to Axiom. While I'm still a big fan of Axiom, I do think Storm is definitely the best ORM (now) available for any project which needs to talk to a database other than SQLite, especially remotely.
The design is simple and straightforward, and the code is well-tested. As a cranky, bitter young man, I must maintain my image; frankly, I'm really shocked by the low quality of the Leading ORMs available for Python or any other language. Storm, fortunately, does many of the things that are obvious to most people who are familiar with how to write software which uses an SQL database, and leaves out the useless crap. Of course, it's not perfect, but now that it's open source, we can accept your patches :-)
Figuring out the best way to use Storm from an asynchronous codebase will be a challenge, probably, but there are techniques for it that are usable right now—one of them being "Just block because your indexes are defined to be fast"—but that isn't always appropriate. There are some other simple options, and maybe some complex ones too. I'll probably post (or get someone else to post) about those in the future.