👍 *** If you love it, like it! ***👍

workflow: https://civarchive.com/models/1088678

👑 Test Mix Face 🎨

About this loras
As my other celebrity loras, this one was train using my best dataset images. The result is a new person every seed (99% of the time).

It also produce a way better quality image that using nothing and still follow the prompt very good.

Changing the size of the latent (like 1.25x) or changing the resolution and keeping the same seed and prompt will still produce a new person.

To get the best result (they are different than my other lora)

Guidance: 2.5

Steps (dev): 30-40

Sampler: euler

Scheduler: simple

daemon detailer (lying sigma sampler): factor: -0.01, start 0.06, end 0.75

Resolution: Original SDXL resolution or Upscale the latent by 1.25

Trigger word is (may work better in certain context): ohwx

Enjoy!