Efficiency Nodes for ComfyUI

A collection of ComfyUI custom nodes to help streamline workflows and reduce total node count.

Direct Download Link

Nodes:

Efficient Loader & Eff. Loader SDXL

• Nodes that can load & cache Checkpoint, VAE, & LoRA type models. (cache settings found in config file 'node_settings.json')
• Able to apply LoRA & Control Net stacks via their lora_stack and cnet_stack inputs.
• Come with positive and negative prompt text boxes. You can also set the way you want the prompt to be encoded via the token_normalization and weight_interpretation widgets.
• These node's also feature a variety of custom menu options as shown below.
  

  note: "🔍 View model info..." requires ComfyUI-Custom-Scripts to be installed to function.

• These loaders are used by the XY Plot node for many of its plot type dependencies.

KSampler (Efficient), KSampler Adv. (Efficient), KSampler SDXL (Eff.)

• Modded KSamplers with the ability to live preview generations and/or vae decode images.
• Feature a special seed box that allows for a clearer management of seeds. (-1 seed to apply the selected seed behavior)
• Can execute a variety of scripts, such as the XY Plot script. To activate the script, simply connect the input connection.

           

Script Nodes

• A group of node's that are used in conjuction with the Efficient KSamplers to execute a variety of 'pre-wired' set of actions.

• Script nodes can be chained if their input/outputs allow it. Multiple instances of the same Script Node in a chain does nothing.

  

  XY Plot

  • Node that allows users to specify parameters for the Efficiency KSamplers to plot on a grid.

  

  HighRes-Fix

  • Node that the gives user the ability to upscale KSampler results through variety of different methods.
  • Comes out of the box with popular Neural Network Latent Upscalers such as Ttl's ComfyUi_NNLatentUpscale and City96's SD-Latent-Upscaler.
  • Supports ControlNet guided latent upscaling. (You must have Fannovel's comfyui_controlnet_aux installed to unlock this feature)

  

  Noise Control

  • This node gives the user the ability to manipulate noise sources in a variety of ways, such as the sampling's RNG source.
  • The CFG Denoiser noise hijack was developed by smZ, it allows you to get closer recreating Automatic1111 results.
  Note: The CFG Denoiser does not work with a variety of conditioning types such as ControlNet & GLIGEN
  • This node also allows you to add noise Seed Variations to your generations.
  • For trying to replicate Automatic1111 images, this node will help you achieve it. Encode your prompt using "length+mean" token_normalization with "A1111" weight_interpretation, set the Noise Control Script node's rng_source to "gpu", and turn the cfg_denoiser to true.

  

  Tiled Upscaler

  • The Tiled Upscaler script attempts to encompas BlenderNeko's ComfyUI_TiledKSampler workflow into 1 node.
  • Script supports Tiled ControlNet help via the options.
  • Strongly recommend the preview_method be "vae_decoded_only" when running the script.

  

  AnimateDiff

  • To unlock the AnimateDiff script it is required you have installed Kosinkadink's ComfyUI-AnimateDiff-Evolved.
  • The latent batch_size when running this script becomes your frame count.

  

Image Overlay

• Node that allows for flexible image overlaying. Works also with image batches.

SimpleEval Nodes

• A collection of nodes that allows users to write simple Python expressions for a variety of data types using the simpleeval library.
• To activate you must have installed the simpleeval library in Python

pip install simpleeval

       

Workflow Examples:

1. HiRes-Fixing
   
   

2. SDXL Refining & Noise Control Script
   
   

3. XY Plot: LoRA model_strength vs clip_strength
   
   

4. Stacking Scripts: XY Plot + Noise Control + HiRes-Fix
   
   

5. Stacking Scripts: AnimateDiff + HiRes-Fix
   
   

Dependencies

Simpleeval library is required to use the simpleeval nodes.

Install:

To install, drop the "efficiency-nodes-comfyui" folder into the "...\ComfyUI\ComfyUI\custom_nodes" directory and restart UI.