>>8668008
It mostly remain static, maybe with some tiny changes.

Here's a test showing Euler a (an Ancestral sampler) versus DPM++ 3M (a non-Ancestral sampler). You can see as the step count increases the Euler a gen keeps changing the character, while the DPM++ one after it gets to a convergence point the character is mostly the same and it's just background details or such changing. If you kept going step 200 for DPM would look very similar to step 100 while for Euler a it would be totally different.

This also shows how Euler takes fewer steps to denoise. That's one reason for it's major popularity, it's fast and requires fewer steps. That makes a big difference on lower-spec machines especially which are what most people have.

Euler a is my standard but in the past when I've done try-hard effortgens I've used DPM++ 3M because it's more consistent and I think it results in better/sharper detail. You have to be willing to wait more steps for it though.

With the new Reforge update the new Sampler SEEDS 3 DP was included which is even SLOWER apparently, but does seem to have really good detail as well. I'm experimenting with it and it might be my new try-hard sampler.