ComfyUI Extension: ComfyUI-TripleKSampler

Authored by VraethrDalkr

Created

Updated

32 stars

Triple-stage KSampler for Wan2.2 split models with Lightning LoRA

Custom Nodes (0)

    README

    ComfyUI-TripleKSampler

    Professional triple-stage sampling nodes for Wan2.2 split models with Lightning LoRA integration. Simple to use, powerful results.

    Features

    • Triple-Stage Workflow - Base denoising → Lightning high → Lightning low
    • Two Node Variants - Simple (smart defaults) and Advanced (full control)
    • Intelligent Auto-Calculation - Optimal parameter computation
    • Model-Safe Cloning - No mutation of original models
    • Sigma Shift Integration - Built-in ModelSamplingSD3 support

    Quick Start

    1. Install

      cd ComfyUI/custom_nodes/
git clone https://github.com/VraethrDalkr/ComfyUI-TripleKSampler.git
cd ComfyUI-TripleKSampler && pip install -r requirements.txt

    2. Use - Find nodes under TripleKSampler/sampling category after ComfyUI restart

    3. Configure - Connect your Wan2.2 models and set basic parameters

    Node Types

    | Node | Best For | Key Features | |------|----------|--------------| | TripleKSampler | Most users | Smart defaults, auto-calculation, streamlined interface | | TripleKSampler Advanced | Power users | Full control, 5 switching strategies, dry-run testing |

    Essential Parameters

    • sigma_shift - Sigma shift value (default: 5.0)
    • base_cfg - CFG for base denoising (default: 3.5)
    • lightning_steps - Total lightning steps (default: 8)
    • lightning_start - Starting step in lightning schedule (default: 1)

    Documentation

    Example Workflows

    Example workflows are included in the example_workflows/ directory.

    Custom LoRAs: The t2v_custom_lora_workflow.json demonstrates layering custom LoRAs with Lightning LoRAs. For I2V workflows, apply the same pattern to your I2V models.

    Support

    License

    Apache 2.0 License - see LICENSE file for details.

    Author: VraethrDalkr