Yes indeed!

I will be making another article very soon explaining in more details. Let's call it part two to the "Turn it to 11" article where I recommended joining multiple lycoris together at lowered weights.

My current meta is to use Embedding at default (1.0) value and join it with lycoris at 0.4 and lora at 0.2

This gives me consistency (pretty much every image gets good face) and likeness.

As @pexero mentioned, if you want a younger version, put more weight on lora and less on lycoris.

In general: embedding guides us in the proper direction, lycoris adds texture (skin texture, moles etc, but also it includes wrinkles) and lora is more of a general concept (so if you do not care for hyper realism but are more into stylism - put more emphasis on lora at the cost of lycoris)

@pexero in a way yes, the SD unfortunately exaggerates the trained features so if someone had some wrinkles - in the lycoris there will be more (this is why i now suggest to use a mixture of models to achieve greatness)

Thanks :)

And nope, I use kohya, here is the guide i've made about my training process: https://civitai.com/articles/3114/textual-inversion-embedding-training-guide