ComfyUI-LG_SamplingUtils

ComfyUI-LG_SamplingUtils
★ 186

采样工具Flow MatchingZImage/Lumina2优化ComfyUI扩展
为ComfyUI提供一组实用采样节点，专注Flow Matching（如ZImage、Lumina2）采样优化，提升操作直观性与效率。
💡 在ComfyUI中优化Flow Matching模型的采样流程与参数调试。
🍴 4 Forks💻 Python🔄 2025-12-24
📦
网盘下载
复制链接后前往夸克网盘下载
https://pan.quark.cn/s/af9fbf81e746
Image
Example 1
Example 2
Example 3
Example 4
Example 5
Example 6
Example 7
Example 8
📄 README

ComfyUI-LG_SamplingUtils

中文文档 | English

Overview

ComfyUI-LG_SamplingUtils is a comprehensive toolset designed for ComfyUI by LAOGOU-666, providing a series of practical sampling nodes that make operations more intuitive and convenient. This extension focuses on advanced sampling techniques, particularly optimized for Flow Matching models like ZImage and Lumina2.

Features

This extension includes four powerful nodes:

1. 🎈 ZImage Timestep Noise

Adds noise perturbation to timesteps during the sampling process to break model homogenization and produce more diverse outputs with different seeds.

Key Features:

  • Two modes: sigma (traditional diffusion models) and flow (Flow Matching models)
  • Adjustable noise strength and application range
  • Optional mask support for localized effects
  • Seed-based reproducibility

    • Parameters:

  • mode: Choose between sigma (multiplicative noise) or flow (additive noise)
  • noise_strength: Control the intensity of noise (0.0-2.0)
  • seed: Random seed for reproducibility
  • start_percent / end_percent: Define the sampling range where noise is applied
  • mask (optional): Limit the effect to specific regions

    • 2. 🎈 LG Noise Injection

    Injects features from a reference image into the generation process through the CFG mechanism, allowing the model to “learn” specific qualities from the reference.

    Key Features:

  • Inject surface details like water droplets, sweat, textures
  • Add material properties and reflections
  • Mask support for targeted feature injection
  • Strength decay over sampling steps

    • Parameters:

  • reference_image: Image containing desired features
  • strength: Injection strength (0.1-0.2 subtle, 0.2-0.4 noticeable)
  • start_percent / end_percent: Define when feature injection occurs
  • mask (optional): White areas receive feature injection

    • Variant: 🎈 LG Noise Injection (Latent) – Works directly with latent representations and automatically uses noise_mask if present in the latent.

    3. Model Sampling ZImage

    Adjusts sampling parameters for ZImage/Lumina2 models.

    Key Features:

  • Proper timestep scaling for Flow Matching models
  • Adjustable noise schedule shift
  • Compatible with ZImage, Lumina2, and AuraFlow

    • Parameters:

  • shift: Noise schedule shift (default 3.0 for ZImage)
  • shift=1.0: Linear schedule
  • shift>1.0: Shifts toward high noise, more aggressive early steps
  • multiplier: Timestep multiplier (1.0 for ZImage/AuraFlow, 1000 for SD3/Flux)

    • 4. Sigmas Editor 🎚️

    Interactive visual editor for adjusting the sigmas curve in real-time.

    Key Features:

  • Drag-and-drop curve editing
  • Real-time visualization
  • Fine-tune noise schedules for optimal results

    • Examples

    Installation

    Method 1: ComfyUI Manager (Recommended)

  • Open ComfyUI Manager
  • Search for “ComfyUI-LG_SamplingUtils”
  • Click Install

    • Method 2: Manual Installation

    cd ComfyUI/custom_nodes
git clone https://github.com/LAOGOU-666/ComfyUI-LG_SamplingUtils.git

    Restart ComfyUI after installation.

    Contact

  • WeChat: wenrulaogou2033
  • Bilibili: 老狗_学习笔记

    • Repository

    https://github.com/LAOGOU-666/ComfyUI-LG_SamplingUtils

    Support

    If you encounter any issues or have suggestions, please open an issue on GitHub.

    Changelog

    v1.0.0

  • Initial release
  • Added ZImage Timestep Noise node
  • Added LG Noise Injection nodes (Image and Latent variants)
  • Added Model Sampling ZImage node
  • Added Sigmas Editor node

    • Enjoy creating with ComfyUI-LG_SamplingUtils! 🎈