Face Consistency: SFW is much better in that point.
I tested this model with many settings and directly compared some NSFW and SFW images. Why is the facial consistency so much better with SFW? What options are available to achieve the same result with NSFW? Could it be that I have a mistake in my workflow? It's actually very simple.
NSFW has far more LORAs, steering the model a bit more towards faces that were in NSFW datasets.
I'm waiting for the next Qwen release to make another set of Qwen AIOs, as I hear character consistency is suppose to be better.
I tried to include a LORA for consistency. But I admit, I'm still quite new to this field.
I've noticed that it's easy to get it wrong with the prompt. As soon as you describe hair, eyes, or anything else near the head, the consistency is gone. It doesn't matter whether it's SFW or NSFW. Unfortunately, that's exactly what I'm looking for. I would like to portray a person in different styles. This also applies to hair and hairstyle, etc. But the face should remain the same.
Are there perhaps better workflows for such a case?