Explore my workflow creation space! Feel free to click the link below and experience the future of AIGC generation directly.<a target="_blank" rel="ugc" href="https://www.runninghub.ai?inviteCode=rh-v1559">https://www.runninghub.ai?inviteCode=rh-v1559</a>Fan benefits,register to get 1000 points,daily login 100 points,play 4090!Experience the super power of 48G.Here is the updated, English-only model card tailored for Civitai, optimized specifically for Krea2 Turbo with your exact parameters:<h1 id="krea2_phantomlight_hanfuphoto-or-cinematic-hanfu-and-oriental-aesthetics">🌸 Krea2_PhantomLight_HanfuPhoto | Cinematic Hanfu &amp; Oriental Aesthetics</h1><blockquote>Base Model: Krea2 (Turbo)Type: LoRABest For: Ancient Chinese Costumes, Hanfu Portraits, Dunhuang Feitian, Wuxia Swordswomen, and Oriental Fine-Art Photography.</blockquote><h2 id="overview">🌟 Overview</h2>Krea2_PhantomLight_HanfuPhoto is a specialized LoRA designed to bring authentic 35mm cinematic film aesthetic and high-fashion realism to traditional Chinese clothing photography. Built to overcome the overly smooth "AI plastic" look, this model delivers realistic skin textures, natural lighting depth, and tactile fabric details.<h2 id="key-features">✨ Key Features</h2><ol><li>Photorealistic 35mm Aesthetic:<ul><li>Delivers genuine film grain, volumetric lighting, and true-to-life skin textures and bone structures.</li></ul></li><li>Exquisite Fabric &amp; Jewelry Details:<ul><li>Accurately renders intricate embroidery, glowing silk luster, sheer chiffon ribbons, traditional silver/gold headpieces, and metallic weapon reflections.</li></ul></li><li>Versatile Oriental Styles:<ul><li>Dunhuang Feitian / Mythic Goddesses: Flying ribbons, reverse pipa poses, and celestial aesthetics.</li><li>Wuxia Swordswomen &amp; Assassins: Bamboo forests, rain-slicked rooftops, straw hats (Douli), and atmospheric blade duels.</li><li>Classical Dynastic Styles: Tang Dynasty Qi-Xiong Ruqun, Song Dynasty elegance, Ming Dynasty formal robes, and Wei-Jin wandering scholars.</li><li>Dark Fantasy &amp; Cyber-Gothic: Ink-wash art styles, neon-lit modern Hanfu, and dark mythic spirits.</li></ul></li></ol><h2 id="recommended-settings-(krea2-turbo)">⚙️ Recommended Settings (Krea2 Turbo)</h2><ul><li>LoRA Weight: <code>0.8 - 1.0</code> (Default: <code>0.85</code>)</li><li>Sampling Steps: <code>8 steps</code> (Optimized for Turbo inference)</li><li>CFG Scale: <code>1.0</code></li><li>Sampler: Euler / DPM++ 2M</li><li>Recommended Resolution: <code>832x1216</code> or <code>1024x1536</code> (Portrait orientation)</li><li>Trigger Words: <code>PhantomLight</code>, <code>HanfuPhoto</code>, <code>cinematic 35mm film photography</code></li></ul>

Krea2_PhantomLight_HanfuPhoto

Krea2_PhantomLight_HanfuPhoto.safetensors

v1.0


# Krea2_幻光古风摄影

## 模型介绍

本模型依托魔搭社区（ModelScope）AIGC专区[模型训练](https://modelscope.cn/aigc/modelTraining)环境与算力完成训练。

* 模型类型：LoRA
* 基础模型：[krea/Krea-2-Turbo](https://modelscope.cn/models/krea/Krea-2-Turbo)
* 训练代码：[DiffSynth-Studio](https://github.com/modelscope/DiffSynth-Studio)
* 训练数据量：59
* 总训练步数：3000
* 开源协议：Apache-2.0

## 推理代码

安装 [DiffSynth-Studio](https://github.com/modelscope/DiffSynth-Studio)：

```bash
pip install diffsynth
```

开始推理：

```python
from diffsynth.pipelines.krea2 import Krea2Pipeline, ModelConfig
import torch

pipe = Krea2Pipeline.from_pretrained(
    torch_dtype=torch.bfloat16,
    device="cuda",
    model_configs=[
        ModelConfig(model_id="krea/Krea-2-Turbo", origin_file_pattern="turbo.safetensors"),
        ModelConfig(model_id="Qwen/Qwen3-VL-4B-Instruct", origin_file_pattern="*.safetensors"),
        ModelConfig(model_id="Qwen/Qwen-Image", origin_file_pattern="vae/diffusion_pytorch_model.safetensors"),
    ],
    tokenizer_config=ModelConfig(model_id="Qwen/Qwen3-VL-4B-Instruct", origin_file_pattern=""),
)
pipe.load_lora(pipe.dit, ModelConfig(model_id="zhouwenbin1994/Krea2_PhantomLight_HanfuPhoto", origin_file_pattern="Krea2_PhantomLight_HanfuPhoto_c2-st3000.safetensors"))
prompt = "a cat"
image = pipe(
    prompt, seed=0,
    num_inference_steps=8, cfg_scale=1, mu=1.15,
)
image.save("image.jpg")
```


Files with SHA256: aca3ee757dddd2c71dbb6d462ec527458dcb2f973a14b619e25d76a51796cdfe