Hi all, this is Iris von Everec from The Witcher 3 expansion Hearts of Stone, now for MOSS Image.
Trained on MOSS-Image Base; I personally use it on MOSS-Image SFT, with or without other LoRAs.
I tested strength from 0.8 to 1.2. I usually use 1.0 — try what fits your prompt.
Example prompt:
<lora:Character_IrisVonEverec_Moss_v1:0.9>
3D CGI video game screenshot, close-up portrait of irisvoneverec, fair skin with light freckles, yellow-green eyes, smudged dark eyeliner, dark lipstick, black hair in a low bun, high-collar black floral damask dress with small buttons, three-quarter view, slight sad smile, moonlight from a window, in-engine render..
The only parts you really need:
Face / likeness:
irisvoneverec, fair skin with light freckles, yellow-green eyes, smudged dark eyeliner, dark lipstick, black hair in a low bun
Clothes (canonical outfit — drop this block for OOD clothing):
high-collar black floral damask mourning dress, grey sash, hood, veil, violet rose
Tip: MOSS follows adjectives very literally. "pale skin" turns her into a corpse and "smudged eye makeup" into a mask; "fair skin with light freckles" and "smudged dark eyeliner" keep her alive.
Hope you enjoy it, and if you use it please share your creations; I'm curious!
Check my profile for other loras (Krea2/Anima): gkhan_
Description
v2: very close to v1, a bit more reliable in some situations.
Same dataset and same 2000 steps as v1. The only change is in the captions: the identity line (freckles, eyes, eyeliner, lipstick, low bun) is now fixed on the images where her face is visible.
Same prompts and seeds in the comparison images, tested on MOSS-Image SFT int8 at strength 1.0. In my tests (72 images, judged by eye) v2 does better here:
- Low bun: it stays more often, especially outside her usual scenes (rainy street, cafe), where v1 tends toward short loose hair.
- Lipstick: it follows the requested dark lipstick more often (pool image: v1 goes red).
- Eyeliner in daylight: a little cleaner around the eyes (cafe image).
Where nothing really changes:
- With a clothing LoRA on top, the two are almost identical (dress image).
- In the canonical outfit the differences are small (photo image).
- v2 comes out slightly darker in dark scenes (CGI image), and dark scenes can still give the heavy "mask" makeup on both.
If v1 already works for you, there is no need to switch.