⚡ Trigger Word: Sonja

Recommended settings (for WAN workflows): Use this same lora for both High and Low noise. High Noise Strength: 0.6 Low Noise Strength: 0.9

USE WITH: Performs flawlessly in WAN 2.1 and WAN 2.2, for both images and videos.

Who is Sonja? Sonja is a wholly fictional AI-crafted character, embodying the spirited blend of Irish and American heritage. Her features reflect her Irish heritage: fair skin with a sprinkle of freckles, striking green eyes that shimmer with mischief, and vibrant fiery red hair. While her default look boasts this bold red hue, you’re free to prompt alternate shades for variety.

Her body shape defaults to relatively slender but is easily changed and customized by your prompt wording.

How she was created

Sonja was meticulously built from scratch as an original AI persona:

Base dataset: Generated using Google’s Nano Banana, guided by a detailed vision of her Irish-American traits and features.

Extended dataset: Additional variations were crafted in Flux Kontext MAX, using Nano Banana outputs as references to explore diverse angles, expressions, outfits, and settings.

Image cleanup: Each dataset image was refined in Photoshop to eliminate aberrations and artifacts.

Captioning: Dataset was captioned in BDTM using Qwen2.5-VL-7B-Instruct with a custom QUERY template for structured, vivid captions.

Training: LoRA trained in AI-Toolkit on RunPod for 3000 steps.

Default hair: fiery red, but promptable to other shades.

Eyes: green usually subdued with a bit of brown near the center. Prompt for vibrant or brighter green eyes if you want that.

Flexibility

This LoRA is incredibly versatile—Sonja adapts to a wide array of aesthetics, outfits, and genres. She layers beautifully with other LoRAs and style models, making her ideal for portraits, editorial fashion, glamour, lifestyle, lingerie, swimwear, or even cinematic video projects. Her Celtic roots and American boldness shine through, offering endless creative possibilities.

Please share your generations!