HerRAW-krea2-lora

HerRAW - v0.2 | Krea 2 LoRA

HerRAW 是一个面向 Krea 2 的通用女性真实摄影 LoRA，目标是在尽量保持原始 Prompt、人物特征、构图和场景控制能力的前提下，提升女性人像的自然摄影质感。

本模型无需触发词。加载 LoRA 后，可以直接使用正常的自然语言 Prompt 进行生成。

v0.2 是 HerRAW 的第二版迭代，在 v0.1 基础上继续优化通用女性摄影的真实质感表现。

---

HerRAW is a general-purpose female photography LoRA for Krea 2, designed to improve natural photographic realism while preserving as much of the original prompt, subject characteristics, composition, and scene control as possible.

The model requires no trigger word. Simply load the LoRA and use normal natural-language prompts.

v0.2 is the second iteration of HerRAW, further refining the natural photographic realism of general female photography on top of v0.1.

---

What's new in v0.2 / 0.2 更新内容

Better face diversity — reduces the issue of different people gradually converging to similar-looking faces, better preserving each subject's original face shape and temperament. 人脸多样性更好 — 减少不同人物逐渐长得相似的问题，更好地保留原本的脸型和气质。

More natural skin — less airbrushed / plastic look, with more realistic skin-tone variation, skin texture and highlight rendering. 肤质更加自然 — 降低磨皮感和塑料感，增加更真实的肤色变化、皮肤纹理和高光表现。

More harmonious East Asian faces — improved overall aesthetics while avoiding a fixed "AI beauty face". 东亚女性面部更协调 — 提升整体审美，同时尽量避免固定的「AI 美女脸」。

More complete photographic feel — further improvements in hair strands, fabric, lighting, and half-body / full-body / environmental portraits. 整体摄影感更完整 — 发丝、衣料、光线以及半身、全身、环境人像都有进一步改善。

---

Samples

Recommended settings

Usage (ComfyUI)

1. Load the Krea 2 Turbo checkpoint (krea/Krea-2-Turbo).
2. Add a Load LoRA node and select HerRAW-v0.2.safetensors (v0.1 also available below).
3. Set LoRA strength around 0.3 - 0.5.
4. Generate with normal natural-language prompts — no trigger word needed.

Files

Version history

• v0.2 — Second iteration: better face diversity, more natural skin, more harmonious East Asian faces, and a more complete photographic feel.
• v0.1 — First experimental baseline, primarily to evaluate the feasibility of a general female photography LoRA from training data.

License

other — LoRA weights trained on top of the Krea 2 Community License base model. See krea/Krea-2-Turbo for the base model terms.