TRIGGER: Hailey
No other triggers are required but training captions included two key features if you want to reinforce or make them stronger: long brown hair with a gradient fade to blonde, striking amber eyes

LoRA Strength: I use .8

Who is Hailey?
She is based on an Erovenus Max Series sex doll by the same name. If you're curious, you can see that here: LINK
I thought she was beautiful, but she is a torso only doll and she needs arms and legs lol. I wanted to "imagine" her in real life so I used her product photos as image references with Flux Kontext (MAX) to generate a lot of varied training images. Then I pulled those training images into my WAN 2.2 I2I workflow to add extra detail +sharpening/grain to end up with a dataset of 40 1024x1024 images trained in 3000 steps on an RTX 6000 Pro.

Image generation parameters I use:

• Sampler: res_2s

• Scheduler: bong_tangent

• High noise: 5 steps, cfg 1, start at 0, end at 2

• Low noise: 9 steps, cfg 1, start at 1, end at 10000