Sophia is a face-focused character LoRA for Krea 2: it locks her facial identity - same face, same features, instantly recognizable - while leaving the body and figure entirely free for the prompt to shape. Change the body type, the outfit, the setting, even the species (see the blue-skinned Na'vi-style showcase below): the face stays Sophia.
🛠️ Made with LoRA Dataset Studio — the open-source tool I use to build and train all my character LoRAs (dataset curation, auto-captioning, face-similarity scoring, training, checkpoint testing, all in one UI). Read the write-up / tutorial.
💬 Join my Discord: discord.gg/wrAWFTPJe
Trigger word: sophia (put it at the start of your prompt).
Versions:
- v1.0 - trained on Krea 2 Turbo, 2500 steps / 80 epochs.
- v2.0 - new training run on Krea 2 Raw, 2000 steps / 64 epochs. Works great on the Krea 2 Turbo base too (all v2 example images were generated that way).
Recommended settings: strength 1.0 when used alone. Body, pose and style stay promptable - describe them and the LoRA follows.
Pro tip - blend characters to design new faces: it is really interesting to combine several of my character LoRAs together at low strength, around 0.4 to 0.45 each. The identities merge into brand new, gorgeous faces that keep the photorealistic consistency of each LoRA.
Description
New training run on Krea 2 Raw: 2000 steps / 64 epochs (v1 was the 2500-step pick). Same face-centered behavior: Sophia's facial identity stays locked while body, hair style and outfit remain free to your prompt. The example images were generated with this exact checkpoint at strength 1.0 on the Krea 2 Turbo base.
Trigger word: sophia. Strength 1.0 solo, or blend with my other character LoRAs at 0.4 to 0.45 each to create brand new faces.