I've been tired of lacking good short nylon socks and decided to train a lora by myself!

The Krea2 model is trained based on Krea-2-Raw, it can be applied to Krea-2-Turbo directly for inference.

The ZIT model is trained based on Z-Image-Turbo.

The IL model is trained based on Diving-Illustrious Real-Asian.

This is my first time training a lora, only tested on the base model for training.

2026/07/06

Krea2-V1 released, this version outperforms the previous models, the result is fantastic. Try different color, and different textures.

Trigger word: nyl0ns0cks

2026/03/05

ZIT-V1 released, updated high quality datasets for training, now it only supports skin tone , as I realized any other color may affect the result.

Tips: you can add prompts to emphasize the texture to get a "Thick nylon" effect, but it will affect transparency a little bit.

Enjoy!

2025/12/11

ZIT supported, however sometimes the texture of the socks is weird, somehow it's still usable. Enjoy.

Use beige or grey to switch color of the socks, I didn't labeled the colors during training.

[Archived: Now the ZIT model only supports skin tone color]

Supported colors: skin tone, grey, let me know if you are keen to other colors.

Black nylon socks will never show in this lora, black is EVERYWHERE ELSE!