💙 Read for best results 💙

Hello everyone, how are you? This is my second LoRA, and I hope you make good use of it. My goal in this training was in Cyberpunk or Android style! I used Animagine XL V3.1 V2 for the training, but feel free to test it on other Checkpoints SDXL.

~ If you use Seaart feel free to help me on the site 🙂

PARAMETERS:

Trigger Word: cyborgirl

Lora Weight: 0.6 - 1

Sampling Steps: 26

CFG: 6
~ I noticed that the lower, the better; do not exceed 28 and 7

Sampling Method: DPM++ 2M KARRAS or Euler A (Both are great)

Clip Skip: 2

RECOMMENDATION:

I recommend using LoRAs for image and character quality to emphasize and improve the quality of your creations even further. I mainly tested with these below (feel free to use others):

• Eye detail XL anime (Eyes)

• Xl Real / Enhancer (Resolution)

• Hands XL + SD 1.5 (Hands)

These gave me perfect results!

📌 For better results in SD, use Hires.fix

Upscaler: ERSGAN_4X / Hires steps: 15-20 / Strength: 0.4-0.6

Final thoughts:

I trained using mostly female images, so train with this goal, but you can use it for other genders, styles, and concepts:

• I didn't train with NSFW images, but feel free to test; I'd love to know if it works too;

• Be careful! When generating male characters, they may have a feminine body, but if you like femboys, it will work fine. (I showed an example below 💀);

• I noticed that Clip Skip: 2 is much smoother and produces better images than Clip Skip: 1, but you can generate both to see which is the best option.

• Leave suggestions and feedback below; I want to know what I can improve in the future of this LoRA.

You can get my DataSet on Mega or at Seaart.