CivArchive
    Storyflow Expression Card - v1.0
    Preview 141517746
    Preview 141517770
    Preview 141517761
    Preview 141517772
    Preview 141517780
    Preview 141517794

    备注:平时自己做视频会用的,仅供参考,大部分模型本身就可以做到这些,LoRA只是提高了一下效率。

    一、简介

    Storyflow Expression Card 用于一次生成同一人物的 2×2 表情卡,并固定为以下顺序:

    1. 左上:正常;

    2. 右上:开心;

    3. 左下:生气;

    4. 右下:惊讶。

    它适合角色设定、表情参考、故事板角色资产和后续角色一致性工作流。文件采用 LoKr 结构,在 ComfyUI 中仍通过普通 Load LoRALoraLoader 节点加载。

    二、模型文件

    - 文件名storyflow-expression-card-v2-lokr.safetensors

    - SHA-256bdcec9c8cea41853264c9809fd9bd37be80aa36038d7d5a971d26f0c2c1698db

    - 推荐强度:Model 0.7、CLIP 0.7

    - 已测试底模:Krea 2 Turbo BF16、Moody-Krea BF16

    三、推荐参数

    - 分辨率1024 × 1280

    - Steps8

    - CFG1.0

    - SamplerEuler

    - SchedulerBeta

    - Denoise1.0

    - Text encoderqwen3vl_4b_fp8_scaled.safetensors

    - VAEqwen/qwen_image_vae.safetensors

    四、使用方法

    1. 将模型放入 ComfyUI/models/loras/krea2/

    2. 在底模和文本编码器之后连接 Load LoRA,Model/CLIP 强度均设为 0.7

    3. 使用下面的 Prompt 结构,保持四个槽位的名称与顺序不变。

    4. 按推荐参数生成。也可以直接把本目录中的 workflow-*.png 拖进 ComfyUI,读取完整节点和参数。

    五、Prompt 模板

    [CHARACTER]
    
    <人物身份、年龄、面部、肤色、发型等描述>; exactly the same person in every panel
    
    [CARD]
    
    expression
    
    [LAYOUT]
    
    2x2 equal portrait grid, read left to right and top to bottom
    
    [PANELS]
    
    top-left normal: relaxed neutral face, closed mouth, calm eyes
    
    top-right happy: genuine broad smile, visible upper teeth, raised cheeks
    
    bottom-left angry: lowered knitted brows, hard stare, pressed lips, tense jaw
    
    bottom-right surprised: raised eyebrows, wide-open eyes, rounded open mouth
    
    [WARDROBE]
    
    <服装描述> in every panel
    
    [RENDER]
    
    realistic studio photograph; front-facing; neutral light-gray background; identical lighting and consistent chest-up framing in all four panels; no text, no labels, no captions

    建议把人物身份和服装写得具体,但不要改变 [PANELS] 的顺序。若需要非写实风格,可修改 [RENDER],同时保留固定网格和一致性约束。

    六、发布图说明

    - 0102:分别在 Turbo BF16 和 Moody-Krea 下进行同 prompt、同 seed 的 LoRA OFF/ON 0.7 全览对比。

    - 0306:LoRA ON 0.7 原始 PNG,均内含 ComfyUI promptworkflow 元数据。

    Description

    一、简介

    Storyflow Expression Card 用于一次生成同一人物的 2×2 表情卡,并固定为以下顺序:

    1. 左上:正常;

    2. 右上:开心;

    3. 左下:生气;

    4. 右下:惊讶。

    它适合角色设定、表情参考、故事板角色资产和后续角色一致性工作流。文件采用 LoKr 结构,在 ComfyUI 中仍通过普通 Load LoRALoraLoader 节点加载。

    二、模型文件

    - 文件名storyflow-expression-card-v2-lokr.safetensors

    - SHA-256bdcec9c8cea41853264c9809fd9bd37be80aa36038d7d5a971d26f0c2c1698db

    - 推荐强度:Model 0.7、CLIP 0.7

    - 已测试底模:Krea 2 Turbo BF16、Moody-Krea BF16

    三、推荐参数

    - 分辨率1024 × 1280

    - Steps8

    - CFG1.0

    - SamplerEuler

    - SchedulerBeta

    - Denoise1.0

    - Text encoderqwen3vl_4b_fp8_scaled.safetensors

    - VAEqwen/qwen_image_vae.safetensors

    四、使用方法

    1. 将模型放入 ComfyUI/models/loras/krea2/

    2. 在底模和文本编码器之后连接 Load LoRA,Model/CLIP 强度均设为 0.7

    3. 使用下面的 Prompt 结构,保持四个槽位的名称与顺序不变。

    4. 按推荐参数生成。也可以直接把本目录中的 workflow-*.png 拖进 ComfyUI,读取完整节点和参数。

    五、Prompt 模板

    [CHARACTER]
    
    <人物身份、年龄、面部、肤色、发型等描述>; exactly the same person in every panel
    
    [CARD]
    
    expression
    
    [LAYOUT]
    
    2x2 equal portrait grid, read left to right and top to bottom
    
    [PANELS]
    
    top-left normal: relaxed neutral face, closed mouth, calm eyes
    
    top-right happy: genuine broad smile, visible upper teeth, raised cheeks
    
    bottom-left angry: lowered knitted brows, hard stare, pressed lips, tense jaw
    
    bottom-right surprised: raised eyebrows, wide-open eyes, rounded open mouth
    
    [WARDROBE]
    
    <服装描述> in every panel
    
    [RENDER]
    
    realistic studio photograph; front-facing; neutral light-gray background; identical lighting and consistent chest-up framing in all four panels; no text, no labels, no captions

    建议把人物身份和服装写得具体,但不要改变 [PANELS] 的顺序。若需要非写实风格,可修改 [RENDER],同时保留固定网格和一致性约束。

    六、发布图说明

    - 0102:分别在 Turbo BF16 和 Moody-Krea 下进行同 prompt、同 seed 的 LoRA OFF/ON 0.7 全览对比。

    - 0306:LoRA ON 0.7 原始 PNG,均内含 ComfyUI promptworkflow 元数据。

    FAQ

    Comments (4)

    ImpossibleBelt5488885Sep 2, 2026
    CivitAI

    Sorry, new to this. Can you use this lora as an I2I starting with a reference image?

    Coffmeteor
    Author
    Sep 2, 2026

    Krea2 doesn't support image input; a simple LoRA wouldn't be enough to implement it. It is on my to-do list, though I haven't started working on it yet.

    JellaiSep 4, 2026

    @Coffmeteor There are two methods of taking inputs, one of which is Ostris's input node, and training in AI Toolkit. There's also KreaEdit, which is aimed more toward changing the input image in this way, I think.

    Coffmeteor
    Author
    Sep 4, 2026

    @Jellai Yes, I’ve tried both of those, but I haven't achieved the desired result yet.

    LORA
    Krea 2

    Details

    Downloads
    437
    Platform
    CivitAI
    Platform Status
    Available
    Created
    9/1/2026
    Updated
    9/8/2026
    Deleted
    -

    Files

    storyflow-expression-card-v2-lokr.safetensors