LEA Character LoRA

LEA is a character LoRA created with the goal of producing a naturally beautiful, believable and distinctive female character with a delicate, petite figure.

Rather than following the typical highly stylized AI beauty aesthetic, LEA is designed around a more natural and authentic appearance. Her facial features are intentionally kept realistic, with subtle asymmetry, natural skin texture, delicate facial proportions and a believable everyday look.

Her character design focuses on a slender, petite physique with delicate proportions, rather than exaggerated curves or an overly athletic or artificially sculpted body. The result is a character that works particularly well for lifestyle photography, candid snapshots, fashion, portraits and everyday scenes.

LEA is designed to retain her recognizable identity across different environments, poses, clothing styles and photographic situations while still giving the base model enough freedom to create varied images.

The LoRA was trained across multiple checkpoints and extensively tested with different camera perspectives, environments, poses and levels of LoRA strength. The goal was not simply maximum likeness, but the best balance between character consistency and natural image generation.

Note: This Lora has been created without the use of a QWEN Dataset generator, instead every Image from the Dataset was created with KREA2.

In my opinion: This preserves much more details of LEA, her skin, her figure. Tooks way longer but the result is perfectly (for me).

Let me know what you think!

⭐ Recommended LoRA Strength

Sweet Spot: 0.60 – 0.70

• 0.60 → More natural and subtle, with greater freedom for the base model. Recommended for creative scenes, varied outfits and lifestyle photography.

• 0.70 → Recommended default. Strong character identity while maintaining a natural and believable appearance.

• 0.80 and above → Stronger character definition, but can begin to make the facial features more pronounced.

For most use cases, I recommend starting at LoRA strength 0.70 and adjusting from there depending on how much creative freedom you want from the base model.

⭐ Recommended Workflow Tools

--> Upscaler with 4xClearRealityV1

--> Eye Detailer: EyefulV2

All images have been created with my workflow:

https://civarchive.com/models/2895137/krea2-t2i-character-consistence-upscale-image-composition-detailer-pm-qol-improvements

My upscaler settings (also found in the workflow), they are just slightly with a denoise of 0.10 to 0.15, i dont want to mess it up just add some clarity what works pretty well:

Character Concept

LEA is intentionally designed to feel natural rather than manufactured. She does not rely on exaggerated proportions, extreme makeup, hyper-symmetry or the stereotypical "Instagram model" aesthetic.

Her defining characteristics are:

• Natural, distinctive facial features

• Soft brown, slightly wavy hair

• Light eyes

• Subtle freckles and realistic skin texture

• Delicate facial structure

• Petite, slender body type

• Natural feminine proportions

• Subtle imperfections and asymmetry

• Minimal, understated beauty styling

The intention is to make LEA feel like a real person who happens to be photographed, rather than an AI-generated fashion archetype.

Trigger word: kan_lea

(it also works without the trigger word, dont ask me why)