ComfyUI Extension: ComfyUI-TripleKSampler

Authored by VraethrDalkr

Created

Updated

71 stars

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

Custom Nodes (0)

    README

    ComfyUI-TripleKSampler

    Triple-stage sampling nodes for Wan2.2 split models with Lightning LoRA integration.

    Features

    • Triple-Stage Workflow - Base denoising → Lightning high → Lightning low
    • Six Node Variants - Simple/Advanced/Advanced Alt for both native KSampler and WanVideoWrapper workflows
    • Intelligent Auto-Calculation - Optimal parameter computation
    • Model-Safe Cloning - No mutation of original models
    • Sigma Shift Integration - Built-in ModelSamplingSD3 support
    • Automatic Sigma Refinement - Theoretical optimization for perfect boundary alignment (refined strategies)

    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. Optional: WanVideoWrapper Integration - Install ComfyUI-WanVideoWrapper to enable TripleWVSampler nodes

    3. Use - Find nodes under TripleKSampler category after ComfyUI restart

      • TripleKSampler/sampling - Native KSampler workflow nodes
      • TripleKSampler/wanvideo - WanVideoWrapper integration nodes
      • TripleKSampler/utilities - Switch Strategy utility nodes

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

    Why Use TripleKSampler?

    The TripleKSampler node streamlines complex multi-model workflows while respecting base model step resolution. The diagram below compares four different approaches:

    Workflow Comparison

    Workflow Comparison:

    1. Base Models Only - Maximum quality, slowest generation (full base model processing)
    2. Lightning Models Only - Minimum quality, fastest generation (full lightning processing)
    3. Typical 3 KSamplers - Manual setup with decent quality and improved motion, but doesn't respect base model step resolution
    4. TripleKSampler Node - Automated approach with decent quality, improved motion, and proper base model step resolution

    The example shown uses lightning_start=2, lightning_steps=8 with the default Base Quality Threshold and the 50% switch strategy. This demonstrates how TripleKSampler automates the complex model switching that would otherwise require manual KSampler coordination.

    Node Types

    | Node | Category | Best For | Key Features | |------|----------|----------|--------------| | TripleKSampler (Simple) | sampling | Most users | Smart defaults, auto-calculation, streamlined interface | | TripleKSampler (Advanced) | sampling | Power users | Full control, 8 switching strategies, dynamic UI, dry-run testing | | TripleKSampler (Advanced Alt) | sampling | Power users | Full control, 8 switching strategies, static UI, dry-run testing - use if dynamic UI causes issues | | TripleWVSampler (Simple) | wanvideo | WanVideoWrapper users | Smart defaults for TripleWVSampler workflows | | TripleWVSampler (Advanced) | wanvideo | WanVideoWrapper power users | Full control for TripleWVSampler workflows, dynamic UI, dry-run testing | | TripleWVSampler (Advanced Alt) | wanvideo | WanVideoWrapper power users | Full control for TripleWVSampler workflows, static UI, dry-run testing | | Switch Strategy (Simple) | utilities | Simple node users | External strategy control, 5 strategies | | Switch Strategy (Advanced) | utilities | Advanced node users | External strategy control, 8 strategies |

    Essential Parameters

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

    Documentation

    Example Workflows

    Example workflows are included in the example_workflows/ directory.

    Text-to-Video (T2V):

    • t2v_simple.json - Simple node with smart defaults
    • t2v_advanced.json - Advanced node with full parameter control
    • t2v_simple_custom_lora.json - Demonstrates layering custom LoRAs with Lightning LoRAs

    Image-to-Video (I2V):

    • i2v_simple.json - Simple node with smart defaults
    • i2v_advanced.json - Advanced node with full parameter control

    WanVideoWrapper Workflows (requires ComfyUI-WanVideoWrapper):

    • t2v_wanvideo_advanced.json - Text-to-Video with TripleWVSampler Advanced
    • i2v_wanvideo_advanced.json - Image-to-Video with TripleWVSampler Advanced

    Hybrid Workflow: hybrid_workflow.json showcases the Switch Strategy utility nodes for external strategy control. Demonstrates using different switching strategies for T2V and I2V branches in a single workflow.

    Math Node Comparison: tripleksampler_vs_math.json demonstrates how to replicate TripleKSampler (Simple) behavior using manual math node calculations. This workflow provides a side-by-side comparison to help understand the internal calculations and validate the node's behavior.

    Known Limitations

    WanVideoWrapper Integration

    ComfyUI-WanVideoWrapper is explicitly a work-in-progress project that receives frequent updates and integrates new features regularly. TripleWVSampler nodes:

    • Cannot be comprehensively tested with all WanVideoWrapper features
    • Some advanced features may not behave correctly with cascaded sampling
    • Some features may conflict with Lightning LoRA workflows
    • Some features may require specific denoising schedules incompatible with triple-stage sampling
    • May break with WanVideoWrapper updates that change the sampler interface

    Before reporting issues with TripleWVSampler nodes: Always test with the original WanVideoSampler node first to confirm the issue is specific to TripleWVSampler and not an upstream WanVideoWrapper issue.

    Support

    License

    Apache 2.0 License - see LICENSE file for details.

    Author: VraethrDalkr