Model Description
A character LoRA trained on Krea 2, capturing a consistent original female character — early-to-mid 20s, pink hair, green/hazel eyes, light freckling, slim build with fuller bust and hip curves.
Trained on a curated dataset of portrait and waist-up photos across varied lighting (golden hour, studio, overcast, neon, winter daylight), backgrounds (urban streets, cafes, parks, office interiors, home settings, formal events), hairstyles (loose waves, updos, braids, ponytails, buns), clothing (casual, business, athletic, evening wear), and camera angles (front-facing, profile, low-angle, high-angle) — all captioned to keep the identity consistent while letting the model generalize scene, pose, and outfit.
Trigger word: mia
Recommended settings:
Base model: Krea 2 Raw (for training) / works on Krea 2 Turbo for fast inference
LoRA strength: 0.8–1.0
Steps: 20–30 (Raw) or 8-step (Turbo)
CFG: ~3.5 (Raw) / no CFG needed (Turbo)
Resolution: 1024×1024 or higher
Tips:
Combine the trigger word with natural, descriptive prompt language (Krea 2 responds well to mood, lighting, and material descriptions).
For best facial consistency, keep prompts focused on close-up or waist-up framing, matching the training data.
Works well across casual, editorial, and studio-style portrait prompts.
License note: Distributed under the Krea 2 Community License Agreement (derivative model of Krea 2). Commercial use permitted under the license's terms — check current revenue thresholds and content moderation requirements before deploying.