Yeo Yuna
A photoreal adult OC LoRA for Krea 2, trained from a curated synthetic image set. Suited to casual phone portraits, mirror selfies, fashion and everyday scenes.
Quick start
Base: Krea 2 Turbo
LoRA strength: 1.0
Steps: 8
Sampler / scheduler: Euler / simple
CFG: 1.0
Model sampling shift: 1.15
Resolution: 1152 × 1536 for portraits; 1536 × 1152 for landscape scenes
Keeping the face and makeup consistent
Keep the age and makeup description the same across your prompts. This is the wording used in the consistency tests:
She is 24 years old. Her makeup is black upper-lash winged eyeliner, defined lashes, softly blended rose blush and satin rose-pink lipstick. Keep normal skin texture and small facial shadows.
Reuse that line, then describe the outfit, expression, pose, surroundings and lighting. Keep descriptions of facial structure and body proportions light, and keep the hairstyle wording consistent when making a series.
For a natural photographic finish, ask for ordinary phone-photo lighting, restrained highlights and visible skin texture. The makeup line above describes a defined everyday look, not a completely bare face. Avoid combining it with conflicting requests for heavy stage makeup, doll-like skin or an entirely different age.
Start with the character LoRA on its own. The comparison tests used no additional style LoRAs or extra Turbo adapter. Style LoRAs can change makeup and facial appearance, so add them one at a time after finding a look you like.
File details
Rank 32 · BF16 · approximately 235 MB. Standard .safetensors adapter for a compatible Krea 2 workflow.
A note on consistency
Expression, lighting, viewing angle, costume context and seed can still change the face and cosmetic rendering. Keeping the prompt consistent helps control the requested styling; it is not a guaranteed identity lock. Hands, reflections and fine details may also need another seed.
Have fun with her, and feel free to share your favorite results!
Description
Initial Yeo Yuna release. Rank 32, BF16 matrix weights, approximately 235 MB. Selected checkpoint: epoch 15. Tested in ComfyUI with Krea 2 Turbo at strength 1.0, 8 steps, Euler / simple, CFG 1.0 and model sampling shift 1.15. The consistency tests used this character adapter on its own, with no additional style LoRAs or extra Turbo adapter. See the main description for the fixed age and makeup wording used in those tests.
FAQ
Comments (1)
May I know how you generated the same face for all the pics?





