This is my second Krea-2 LoRA. With my first one I trained on the Base model and the results weren’t great… but I was honestly shocked when I tested the Base LoRA on the Krea-2 Turbo models — the quality jumped massively.

So this is a Base LoRA specifically tuned to shine on Turbo models (you can still try it on the Base model, I just haven’t tested that yet). examples images were Generated with WANGP, you can also try comfy but i haven’t tested that yet.

Dataset: Images from Gemini
Trigger word-Honestly you dont need it but if you want you can also try "blonde anime woman" "gemma"

About the character

This character was originally planned for a Z-Image LoRA, but I decided to train it on Krea-2 instead.
a Z-Image Turbo version is also available

What this LoRA does well

• Extremely consistent face

• Handles complicated poses (including yoga poses)

• Strong expressions (angry faces and more)

• Lots of room to experiment — try different styles and see what else it can do
  

Training setup: RTX 3080 10GB + 64GB DDR4 + Ryzen 2700X

A higher-quality version is available on my Patreon.