Character LoRA for Krea 2
WHO IS ELENI?
Eleni is a synthetic character: a 19-year-old Greek woman with evenly sun-tanned golden-olive skin. All training images are AI-generated using https://civarchive.com/models/2676616/sick-ollie.
The LoRA keeps one identity consistent – face, hair and body – so you only need the trigger word instead of long appearance descriptions.
Trigger word: nnEl3ni
WHAT THE LORA LEARNED
Face
Round face with full cheeks, a soft rounded jawline and a smooth, evenly rounded chin
Large, almond-shaped hazelnut-brown eyes with long dark lashes, thin black eyeliner and light warm-brown eyeshadow
Thick, dense dark-brown eyebrows with a soft arch and a clean, well-groomed outline
Straight nose with a softly rounded tip
Full, natural rose-pink lips, lower lip slightly fuller than the upper lip
Light freckles across nose and cheeks, warm pink sun flush on cheeks and nose bridge
Hair
Long, jet-black, straight, sleek and smooth hair falling to the middle of the back
Center part, tucked behind the ears
Rear view: outer sections of the hair gathered at the back of the head with a thin black hair tie, forming a small smooth tail over the loose hair
Body
Slim, athletic build, average height, narrow shoulders, lean toned arms
Small, firm, softly rounded bust with light-brown areolas and nipples
Flat stomach with lightly defined abs, narrow waist, slim hips, firm toned buttocks
Slim thighs, toned calves, slim ankles, slender feet
SFW and NSFW
WHAT YOU GET
A very consistent face: eyes, bone structure, makeup and freckles stay locked
Consistent body and hair
Works with detailed natural-language prompts and with photo-style prompts such as "amateur smartphone flash photo"
Extra details like tattoos, piercings, jewelry and nail polish can be added in the prompt [only keep after test]
Covers nude and clothed content, full body, waist-up and close-up portraits
HOW TO USE
Trigger word: nnEl3ni
The trigger word can sit inside a normal sentence ("...photo of nnEl3ni standing on a whitewashed village street...")
LoRA weight: start at 1.0 and adjust to taste
Describe only what should change: pose, outfit, accessories, location, lighting, camera angle. You do not need to describe face, hair or body shape.
Trained on Krea 2 Raw, tested on these Krea 2 Turbo checkpoints: Cyber Realistic Krea 2 (v4), Gonza Lomo v4, Sick Ollie, Sick Realsim v4 => It sticks to the character but keeps the model bias
Example: Amateur smartphone photo of nnEl3ni walking along a whitewashed village street in the evening. She is wearing a white linen dress and leather sandals. The camera captures her from the front, from head to feet, at eye level, in this full body well lit shot. Golden-hour backlight, warm summer atmosphere.
TRAINING DATA
56 images
8 viewing angles: front, front three-quarter (left/right), side (left/right), rear three-quarter (left/right), rear
24 studio shots on plain light-gray, dark-gray, beige or white backdrops (16 full body, 8 to mid-thigh)
12 scene shots (rooms, beach, mountain stream, olive grove, terrace and more)
8 clothed everyday full-body shots
12 close-up portraits with different outfits, expressions and lighting
Image sizes: 1088×1920 (9:16) for full body, 1152×1536 (3:4) for mid-thigh shots and close-ups
TRAINING SETUP
Base: Krea 2 Raw
2,500 steps, 10 checkpoints (one every 250 steps), released checkpoint: E10, final
Rank/alpha 32/32, learning rate 1e-4, AdamW8bit, constant scheduler, batch size 1
GOOD TO KNOW
The hair is always long, black and straight. The small tail with the hair tie appears only in rear views. Other hairstyles may not keep the identity as reliably.
One fixed look: one age, one body type, one hair color. The LoRA is not meant for age or style changes of the character.
No hats, hoods or sunglasses appear in the training images. Face and hair were always visible.
LICENSE
You are allowed to:
Use the LoRA without crediting me
Use the LoRA on CivitAI or any other service
Sell generated images
You are not allowed to:
Sell the LoRA
Merge the LoRA
Description
Initial revision
See Storyline





