I've only had bad experiences with ORMs. You get implicit behaviours and performance nightmares.

SQL isn't hard to write, unless the model is poorly conceived, in which case throwing an ORM at it will only makes things worse.

It really depends on how many tables you have, and how you generate their definition.

Personally I once had to start a fresh project with a large amount of tables, I generated the few CRUD requests I needed on all the tables, and commit them in my VCS. Then when iterating I would update and tweak them manually, either for performance tuning or new features / joins / schema changes etc...

This is not to say ORMs are to throw in the bin, I'm sure they solve some problems, but I've never fund myself in a situation where they would help me more than slow me down.