Sign In

Zelda (TotK / BotW) - Realistic | Krea 2

0

Not recently loaded

Download

1 variant available

bf16 SafeTensor

zelda_realistic_krea2.safetensors

BF16, good balance • 218 MB

Verified:

Type
LoRA
Stats

199

0

Reviews
Published

Oct 30, 2026

Base Model
Training
Steps: 1,250Epochs: 10
Usage Tips
Strength: 1
Hash
AutoV27B13AFF756
Trigger Words
zld_perso

Krea 2 is licensed under the Krea 2 Community License Agreement. For more information, visit https://krea.ai/krea-2-licensing.

03_totk_feu.png

Photorealistic adult Zelda (Tears of the Kingdom / Breath of the Wild) for Krea 2. The LoRA carries the face only: outfits and hairstyles stay fully promptable, so you can use the game outfits or dress and style her any way you like.

Trigger word: zld_perso

Recommended prompt start: "A photo of zld_perso, a grown woman in her mid-twenties with adult facial features, with ... wearing ..."

Hairstyles learned by name:

TotK: "her short Tears of the Kingdom hairstyle" (chin-length bob with the thin braid headband and blue clips).

BotW: "her Breath of the Wild hairstyle, very long golden blonde hair falling straight down to her waist" (long hair with the braided crown; describe the length, otherwise it tends to come out short).

Anything else works too: ponytail, bun, braid, other colors...

Game outfits: describe them as real clothes (e.g. "a short brown hooded capelet over a royal blue tunic with gold embroidery" for TotK, "a royal blue long-sleeved wool tunic with gold trim over a white ribbed bodice" for the BotW champion outfit). The white prayer gown, the royal blue ceremonial gown and the white fur snow coat were also in the dataset.

Settings: CyberRealistic Krea 2 v4.0 (trained on Krea 2 Raw, tested on 6 Krea 2 checkpoints), 8 steps, CFG 1, euler / simple, LoRA weight 0.8-1.0. Showcase images also use Detailer-KREA2 at 1.0 for skin texture. Tips: add "real skin texture with pores, film grain, not a 3d render, not a video game" for a true photo look; for full body shots add "tall grown woman, adult body proportions".

Trained with ai-toolkit, 64 images, rank 32, 1250 steps. Dataset made from a single original reference (no real person LoRA involved). She is always an adult woman: please keep her that way.