牧濑红莉栖角色 LoRA，基于 Krea-2-Turbo 训练，265 张图像涵盖敞开白大褂、仅衬衫、驼色开衫三类服装，触发词 Makise Kurisu，推荐强度 0.8、18 步、CFG 1.0、er_sde + simple、1024×1376，仅配 Krea-2-Turbo 底模有效，仅供学习研究禁止商用。

白大褂

Makise Kurisu from Steins;Gate, a slim young woman with long reddish-brown hair and large violet-blue eyes whose irises are deep purple at the core fading to blue at the edge, wearing a knee-length pure-white lab coat open at the front, the two front panels falling apart at the sides, plain white notched lapels, long white sleeves lying close to her arms, a straight narrow cut, a smooth plain white front. Under the coat a white collared shirt with thin blue piping on the collar and chest pocket, a red necktie, black shorts with a gold square belt buckle, black pantyhose, black ankle boots

只穿衬衫

Makise Kurisu from Steins;Gate, a slim young woman with long reddish-brown hair and large violet-blue eyes whose irises are deep purple at the core fading to blue at the edge, wearing a long-sleeved white collared shirt with thin blue piping on the collar and chest pocket, buttoned up to the throat, a red necktie knotted at the collar, black shorts with a gold square belt buckle, black pantyhose, black ankle boots

驼色外套

Makise Kurisu from Steins;Gate, a slim young woman with long reddish-brown hair and large violet-blue eyes whose irises are deep purple at the core fading to blue at the edge, wearing a loose camel-brown jacket slipped off her shoulders and hanging on her upper arms, the long sleeves bunched around her elbows, a black leather strap with a metal buckle fastened around her upper arm. Under the jacket a white collared shirt with thin blue piping on the collar and chest pocket, a red necktie, black shorts with a gold square belt buckle, black pantyhose, black ankle boots

Makise Kurisu 角色 LoRA · Krea-2-Turbo

【模型说明】本 LoRA 基于 Krea-2-Turbo 架构训练，用于生成《命运石之门》角色牧濑红莉栖的插画形象。训练集共 265 张图像，按服装分为敞开白大褂、无外套（仅衬衫）、驼色开衫三类，覆盖全身、半身、胸像三种景别。训练参数为 rank 32、8000 步。

【适用底模】仅适用于 Krea-2-Turbo。底模需配合 Qwen3-VL 4B 文本编码器与 Qwen-Image VAE 使用。

【使用方式】1. 将文件放入 ComfyUI/models/loras 目录，子目录不限。2. 触发词为 Makise Kurisu，写在提示词开头，本模型未设其他触发方式。3. 提示词建议写成自然语言长句，先写角色名与外观，再写服装，最后写场景与构图；不建议使用逗号堆叠的标签串。

触发词与描述示例：Makise Kurisu from Steins;Gate, a slim young woman with long reddish-brown hair and large violet-blue eyes whose irises are deep purple at the core fading to blue at the edge, wearing a knee-length pure-white lab coat open at the front, ...

【推荐参数】LoRA 强度 0.8（可用区间 0.6–0.9）；采样步数 18；CFG 1.0；采样器 er_sde；调度器 simple；分辨率 1024×1376（短边 1024）。

【使用提示】· 眼睛颜色需写成渐变关系（虹膜核心深紫、边缘过渡到蓝），仅写 blue eyes 会得到饱和钴蓝。· 白大褂的默认先验为敞开状态，描述改不动，这是训练集决定的。· 本链路 CFG 固定为 1.0，负面提示不起作用，不要依赖负面词做修正。· 画风由训练集固定，追加画风词几乎没有效果；如需要不同画风，应改用其他底模或在推理时切换模型。· Krea-2-Turbo 底模本身已能识别该角色，LoRA 的主要作用是把瞳色、服装细节和画风收紧到训练集的样子。

【许可声明】本模型仅供学习与研究使用，不得用于商业用途。角色版权归原权利人所有。

Makise Kurisu Character LoRA · Krea-2-Turbo

Model Description: This LoRA is trained on the Krea-2-Turbo architecture to generate illustrations of Makise Kurisu from Steins;Gate. The training set contains 265 images across three outfit categories — open lab coat, shirt only, and camel-brown cardigan — covering full-body, half-body and bust framings. Trained at rank 32 for 8000 steps.

Base Model: Krea-2-Turbo only. Requires the Qwen3-VL 4B text encoder and the Qwen-Image VAE.

Usage: 1. Place the file in ComfyUI/models/loras; subfolders are fine. 2. The trigger word is Makise Kurisu, placed at the start of the prompt; no other trigger form is supported. 3. Write the prompt as natural-language sentences — character name and appearance first, then outfit, then scene and framing. Comma-stacked tag lists are not recommended.

Prompt example: Makise Kurisu from Steins;Gate, a slim young woman with long reddish-brown hair and large violet-blue eyes whose irises are deep purple at the core fading to blue at the edge, wearing a knee-length pure-white lab coat open at the front, ...

Recommended Settings: LoRA strength 0.8 (usable range 0.6–0.9); steps 18; CFG 1.0; sampler er_sde; scheduler simple; resolution 1024×1376 (short side 1024).

Notes: Describe the eye colour as a gradient (deep purple at the iris core, fading to blue at the edge); writing "blue eyes" alone yields a saturated cobalt blue. The lab coat defaults to an open state and cannot be changed by description alone — this comes from the training set. CFG is fixed at 1.0 on this pipeline, so negative prompts have no effect; do not rely on them for corrections. The rendering style is fixed by the training set and additional style words have almost no effect; switch base models or use another model if a different style is needed. The Krea-2-Turbo base model already recognises the character on its own; the LoRA mainly tightens eye colour, outfit details and rendering style towards the training set.

License: For research and study purposes only. Commercial use is not permitted. All character rights belong to their respective owners.