Trigger Word: south-indian-beauty

Base Architecture: FLUX.2 [klein] 9B

Training Depth: 15 Epochs | 3,188 Steps

This model leverages the optimized FLUX.2 [klein] architecture, striking the perfect balance between high-parameter intelligence and generation speed. By training over 15 full epochs with a refined Unet LR of 0.0001, this LoRA captures an extraordinary level of anatomical stability and textural detail. It is designed to render the authentic, vibrant essence of South Indian aesthetics with hyper-realistic clarity.

• Epochs: 15 (Extended duration for superior feature retention)

• Total Steps: 3,188

• Num Repeats: 10 (Focused reinforcement per image)

• Learning Rate: 0.0001 (Fine-tuned for precision skin grain and fabric weaves)

• Resolution: 1024 x 1024 (Native high-res bucketing enabled)

• Clip Skip: 1 (Maximum compatibility with FLUX text conditioning

Recommended Inference Settings

For the Klein 9B architecture, use these settings for the best balance of quality and speed:

• LoRA Weight: 0.8 - 1.2

• Sampling Steps: 9 - 12

• CFG Scale: 1

• Sampler: Euler

• Size: Optimized for 1024x1024 or higher.

Flux.2 Klein 9B - V1: