This LoRA model was trained using the Kohya_ss fine-tuning framework on Stable Diffusion 1.5 base model. The training dataset consists of 11 high-quality portrait images of a single subject named Janka, capturing various facial expressions and lighting conditions to ensure versatility in generated images. The network architecture uses a network dimension of 16 and alpha of 8, optimized with the AdamW8bit optimizer and cosine learning rate scheduler. Training was performed over 10 epochs with a learning rate of 0.0001 and mixed precision (fp16) to balance performance and efficiency.

This LoRA is designed to enhance portrait generation by injecting the unique facial features and stylistic traits of woman into generated images. It is best used with the trigger word “Janka” in the prompt to activate the model’s influence effectively. The model supports resolutions up to 512x512 and performs optimally when combined with standard Stable Diffusion 1.5 workflows.

Ideal for creators looking to generate consistent, high-quality portraits of woman with natural expressions and subtle lighting variations. This model is suitable for artistic projects, character design, and personalized image generation.