MOSS-Image Release 💜
Seraphina has been rebuilt for MOSS-Image using a refined dataset and natural-language captions, bringing her familiar purple hair, violet eyes, and distinctive appearance to a new architecture.
Trained with KohakuMossTrainer for 1,500 steps (Rank 16 / Alpha 32).
Trigger:
sera_charRecommended strength:
0.8Tested with: MOSS-Image-Base and StableYogi MOSS Realism
Best results: Portrait, chest-up, and waist-up compositions for stronger facial consistency.
A fresh foundation for Sera, with improved photographic realism and flexibility across different scenes and styles.
SDXL:
V2.0:
Retrained with a more precise dataset, with greater control over the level of detail and quality of the dataset.
Tested with DPM++ 2M SDE, 27 steps, 3.5 CFG
Renders made with Stable Yogi's Realism SDXL 8.0fp32
Trigger word is still sera_char
You can add "purple_eyes" to the prompt to help enforce the purple eyes.
Recommend Adetailer/FaceDetailer
After my initial attempts at trying to design a LoRA for my AI Seraphina's design (which incorporated her nonhuman elements), I realized crafting a stable base would work best, so here she is! I generated a raw "human" version using her stylized self as a base, then crafted a dataset around that. Feel free to share your creations.
Tested with realismSDXL by Stable Yogi v70FP16
Works well with the following settings:
DPM2++ Karras
CFG: 5
Steps: 30
Optional trigger: sera_char

Description
Seraphina | MOSS-Image Character LoRA
A new model, a new foundation, and a familiar face. 💜
After developing Seraphina through several SDXL LoRA iterations, I've brought her over to MOSS-Image, rebuilding her character LoRA with a curated dataset and natural-language captions tailored to the new architecture.
Seraphina (Sera) is one of my original AI characters. Her design originally incorporated several nonhuman elements, but during my earliest attempts at training her, I realized that establishing a consistent, realistic human appearance would make a much stronger foundation.
Using her original stylized design as a reference, I developed a human interpretation of Sera and built a dedicated dataset around her. This MOSS version continues that work, aiming for stronger photographic realism, recognizable facial features, and flexibility across different environments and styles.
She's recognizable by her long, flowing purple hair, violet eyes, and mature, feminine appearance.
Feel free to experiment with her, explore different styles, and share your creations!
Model Information
Base model: MOSS-Image-Base
Training: KohakuMossTrainer
LoRA rank: 16
Alpha: 32
Training steps: 1,500
Trigger word:
sera_charRecommended strength:
0.8
Recommended Usage
Start with the trigger word sera_char and describe your scene naturally. MOSS responds well to descriptive prompts covering clothing, poses, environments, lighting, and camera composition.
If her eye color isn't coming through correctly, explicitly describe her violet or purple eyes in your prompt.
I've tested this LoRA with both MOSS-Image-Base and StableYogi MOSS Realism, with particularly pleasing photographic results from StableYogi.
For consistent facial identity, I recommend portrait, chest-up, and waist-up compositions. Wider full-body shots can lose some of her facial detail, so closer framing generally works best.
Showcase Images
The included examples demonstrate Sera across casual everyday scenes, cinematic portraits, and intimate boudoir photography.
Most showcase images were generated using StableYogi MOSS Realism at a LoRA strength of 0.8.
A Note About Sera
This LoRA represents Seraphina's realistic human appearance, rather than her original stylized, nonhuman design. It's intended as a versatile character foundation that can be used in a wide range of settings without requiring a specific outfit, pose, or environment.
As always, I'd love to see what you create with her. 💜






