A style LoRA for Krea 2 that turns the base model's semi-real look into a glossy, oil-painting-like anime illustration: distinctly anime faces with large expressive eyes, luminous glossy skin, strong directional light (sunbeams, dappled light through a straw hat, sunset backlight) and richly detailed, realistic backgrounds. Everyday slice-of-life scenes are its home ground.

Recommended settings

- Base model: Krea 2 Turbo (trained on Krea 2 RAW, works on both)

- Strength: 0.6 (usable range 0.5–0.7; use 0.7 for extra gloss, avoid 0.85+)

- Trigger: gloss_anime_style

- Sampler: 8 steps, CFG 1.0, euler / beta (standard Turbo settings)

- Prompts: natural-language sentences work best. If you want a specific eye color, name the color — eye color is intentionally left open.

- Optional hires: latent upscale 1.5x + denoise 0.35 redraw (all showcase images use it, 1248×1872)

Traits & quirks

- White tights / pantyhose tend to appear on their own — they are baked in as part of the style. If you don't want them, say so in the prompt.

- The finish is painterly, closer to oil painting than to flat cel anime. That is the point of this LoRA.

- Stacks well with character LoRAs (tested: character LoRA 0.9 + this LoRA 0.6).

Training

39 images, Krea 2 RAW (bf16), musubi-tuner, rank / alpha 32, lr 1e-4, 3 epochs. The v2 dataset replaced narrow / downcast-eye images with wide-open-eye ones, so no special eye wording is needed in prompts.

License

This model is a derivative of Krea 2 and is distributed under the Krea 2 Community License.