SDXL-V2:

For the model, I used same dataset as 100 images with 3,000 steps and 10 Epochs and 2 repeats

In training, the "resolution" was 1024. Hints have been compiled for all images.

All images used in the model are created in civit ai using my flux mode

SDXL-V1:

For the model, I used 10 test to images with 1500 steps and 10 Epochs and 2 repeats

In training, the "resolution" was 512. Hints have been compiled for all images.

This is the SDXL compatible version of the South Indian LoRA. Creates modest looking Saree outfit for whatever your subject is and with little tweaking can make it little bit risque.

Designed to be as flexible as possible:

• Employs cinematic or natural beauty techniques

• Consistent identity in full-body, mid-body, and close-up images

A flexible female character that may be tailored to your artistic vision.

This project started from a single Text-to-Image generation. I used that original spark to create 10 distinct styles, which I then trained using CivitAI’s default settings. To be honest, I was learning as I went and didn't know much about advanced settings at the time—but the results were so unique using onsite generator.