Making a batch vary

Left alone, a batch clusters. Every variant drifts toward the most likely interpretation of your brief, which means the most average one.

Steps

1. Set your batch size to something you will actually review. Ten is plenty.
2. Set at least two axes to vary rather than fixing them.
3. Generate, then sort the results by how different they are from each other, not by how good they look.
4. Kill the duplicates before you kill the ugly ones.

Why sorting by difference first

A batch of ten where four are near-identical is really a batch of seven. Spotting that before you judge quality tells you to widen the brief, not to regenerate.

What to watch for

Very specific briefs produce very similar batches, which is correct behaviour and not a bug. If you have described the image precisely, there is only one image that matches. Precision and variation are a trade, and you pick which one you need per run.