Split testing is the only way to know. There's no way to tell beforehand. Different things work for different traffic sources, and different demographics, in different niches. Make a batch of a dozen, run them for a week or so. Clip out the ones that are obviously performing weak. Look at the 3 that are performing the best, try and replicate the features that you think are generating the clicks with so variety, toss them into the mix and see how they perform. Wash, rinse, and repeat. Keep narrowing it down until you have a small batch that perform well.
Don't just watch CTR. Your banner can affect the resulting conversion ratio too. You may have a lower CTR, but perform better for the entire conversion path.
