CivArchive
    Flux.2 Klein - Face ID & Art Style Dual Lock Workflow (脸部与画风双重锁定工作流) - v1.0
    NSFW
    Preview 141457125
    Preview 141457137
    Preview 141457174

    ### 🌟 Overview & Features

    This ComfyUI workflow is built on Flux.2 Klein, specifically optimized for Dual Locking: Character Face ID + Illustration Art Style.

    - Simultaneous Consistency: Keeps both facial identity and rendering art style consistent when modifying outfits, poses, and backgrounds.

    - LoRA Style Bleed Suppression: Effectively tames and suppresses strong LoRA style contamination (recommended LoRA weight: 0.8), preserving desired concept features without distorting the intended art style.

    - Beginner-Friendly Layout: Streamlined structure with clearly separated reference inputs, avoiding unnecessary tangled connections.

    ### ⚙️ Core Requirements

    - UNet / Base Model: flux-2-klein-9b

    - CLIP: qwen3-8b

    - Face ID Node: PuLID-Flux2

    - VAE: full_encoder_small_decoder

    ### 💡 Usage Tips

    1. Reference Selection: For best results, ensure your face reference and art style reference share the same artistic lineage.

    2. Weight Tuning: If an external LoRA causes slight style deviation, adjust its model strength between 0.5 ~ 0.8.

    ### ☕ Creator's Note

    I searched across the community and couldn't find a reliable workflow capable of cleanly locking both face identity and art style simultaneously. Even when complex setups exist, they are often far too convoluted and frustrating for beginners to navigate.

    I spent about a week dialing in this setup. My goal is to save fellow creators from constantly having to reinvent the wheel, so you can skip the friction and jump straight into creating. Enjoy!

    ---

    ### 🌟 核心特点与概述

    这是一个基于 Flux.2 Klein 构建的 ComfyUI 工作流,专为**角色脸型(ID)与画风双重锁定**深度优化:

    - 双重锁定:在更换服装、动作与背景场景时,能同时稳住角色面部特征与固有插画风格。

    - 抑制 LoRA 风格污染:实测能有效压制 LoRA 自带的画风污染(推荐 LoRA 权重 0.8),在借用 LoRA 元素的同时不破坏原画风。

    - 初学者友好:节点逻辑与输入区域划分清晰直观,开箱即用,无需复杂的连线调试。

    ### ⚙️ 核心依赖模型

    - UNet / 底模: flux-2-klein-9b

    - CLIP: qwen3-8b

    - 锁脸节点 (Face ID): PuLID-Flux2

    - VAE: full_encoder_small_decoder

    ### 💡 使用建议

    1. 参考图选择:建议脸型参考图与画风参考图属于同一种画风体系,成图效果与一致性最佳。

    2. 权重微调:若外挂的 LoRA 画风过强产生干扰,可将其模型权重在 0.5 ~ 0.8 之间进行微调。

    ### ☕ 作者杂谈

    在社区里找了一圈,我基本上没发现能够同时稳稳锁住画风和脸型的工作流。即使有些复杂工作流能实现部分功能,对初学者来说上手门槛也实在太高、用起来非常繁琐。

    我花了差不多一周时间把这套节点和权重逻辑调通做出来,就是希望大家不用再反复“重新造轮子”,能帮大家省去折腾节点的时间,直接专注在出图创作上。希望对大家有所帮助!

    Description

    Initial release (v1.0): ComfyUI workflow for simultaneous Face ID and Art Style consistency on Flux.2 Klein.

    初始版本发布 (v1.0):基于 Flux.2 Klein 的脸型与画风双重锁定 ComfyUI 工作流。

    Workflows
    Flux.2 Klein 9B-base

    Details

    Downloads
    77
    Platform
    CivitAI
    Platform Status
    Available
    Created
    9/1/2026
    Updated
    9/9/2026
    Deleted
    -

    Files

    flux2KleinFaceIDArtStyleDualLock_v10.json