LiaGPT

This LoRA was trained to reproduce a specific female character identified by the trigger word “liagpt”. The model focuses primarily on facial identity and subtle natural features while maintaining compatibility with a wide range of base models and lighting conditions.

The character typically presents:

warm amber eyes

soft natural freckles across the nose and cheeks

full natural lips

thick expressive eyebrows

dark brown slightly messy hair

fair skin with natural texture

The LoRA was trained using caption-based descriptions to encourage natural photographic results rather than heavily stylized outputs. It performs best in portrait and medium-shot compositions but can generalize to wider shots depending on prompt and base model.

The recommended training checkpoint is epoch 8, which provides the best balance between identity retention and natural skin texture without overfitting.

Recommended Settings

Trigger word: liagpt

Recommended weight: 0.7 – 0.9

Best for: portraits, medium shots, cinematic lighting

Works with: realistic photography prompts and natural lighting setups