ComfyUI-XJNodes

ComfyUI-XJNodes

A modular collection of custom nodes for ComfyUI, organized by data type for easy navigation and maintenance.

Note: This repository is primarily for personal use and experimentation.

Available Nodes

This package provides custom nodes organized into four main categories:

Image Nodes

• Image Batch Resize (Pad/Crop) (XJNodes/image)
  • Combine up to 8 image inputs into one batch, in input order. Each input may itself be a batch.
  • image_1 is required; image_2 through image_8 may be disconnected or None and are skipped.
  • Set width and height for the output resolution (default: 1024 × 1024).
  • pad: preserve the full image with centered black padding. crop: fill the target and crop excess from the center. Both preserve aspect ratio up to pixel rounding and use bilinear resizing with antialiasing.
  • Inputs must have matching channel counts. For more than 8 inputs, chain nodes using the same resolution.

Loaders:

• Random Image from Dir (XJNodes/image)

  • Returns one image, its transparency mask, and its absolute file path on each execution.
  • random: visits every image once in shuffled order before reshuffling; avoids an immediate repeat between cycles when there is more than one image.
  • iteration: starts at a random position in the sorted file list, advances one file per execution, and wraps around.
  • Choose input or output and enter a subdirectory, or use Browse Subdirs, matching Load Image With Metadata. Enable recursive to include subdirectories. The selected image appears in the node preview after execution. Supports PNG, JPEG, WebP, BMP, TIFF, and GIF (first frame only), with EXIF orientation applied.
  • Memory is local to the node instance, is not saved in the workflow, and resets when the instance is recreated or the directory, mode, recursive setting, or file list changes. Empty/missing folders and unreadable images raise errors.

• Load Images From Dir Batch / Load Images From Dir List

  • Load multiple images from a directory as a batch/list, with options for image cap, start index.
  • Allow Recursive loading
  • Inspired by (Dr.Lt.Data's Inspire-Pack)

  

• Load Image With Metadata

  • Load image from input/output directories with full metadata extraction

    • dimensions
    • format
    • PNG text chunks
    • workflow data
    • extra metadata fields

    

  • Browser UI for selecting images

    

• Load Image By Path

  • Simple loader that takes directory and filename as string inputs
  • Useful for dynamic loading based on workflow logic (E.g. Loading image path from metadata)

  

Savers:

• Save Image With Metadata

  • Save images with embedded metadata, workflow, and custom metadata fields
  • Options for output directory, filename patterns, and format (PNG/JPEG)
  • Inspired by (WAS's WAS Node Suite)

  

Transforms:

• Image Transform
  • Apply transformations to images with specified parameters
• Image Random Transform
  • Apply random transformations to images for data augmentation

Scaling:

• Image Scale Calc
  • Calculate scaling dimensions for images
• Image Scale Megapixel
  • Scale images to target megapixel resolution

Batch:

• Image Grid
  • Arrange multiple images into a grid layout
• One Image From Batch
  • Extract a single image from a batch by index
• Image List Filter
  • Filter images from a list based on criteria
• Random Images From Batch
  • Randomly select images from a batch

Compare:

• Image Pair Compare
  • Compare two image sets with paired selection
  • Numeric tabs (1, 2, 3...) to select corresponding A/B image pairs
  • Slide and Click comparison modes
  • Sliding right reveals A; sliding left reveals B; "Open Image" opens the image with the larger visible area
  • Image size display under each image
  • Inspired by rgthree's Image Comparar

Lora Nodes

• Conditional Lora Loader
  • Load multiple loras dynamically with a single boolean condition
  • Enable/disable all loras with one toggle
  • Supports multiple lora inputs (lora_1, lora_2, etc.) with individual strength controls
  • Inspired by rgthree's Power Lora Loader

Text Nodes

Loaders:

• Random Text From List

  • Select text from a numbered list widget (fixed or random selection)
  • Fixed or random selection options
  • Simple YAML-like list syntax
  • Use full-line comments such as # Landscapes as section headings. Click T (Table of contents) in the editor toolbar to list headings with source line numbers, then select one to scroll to and highlight its line. Tab/Enter navigate the buttons; Escape closes the contents. Headings remain excluded from text selection.
  • Useful for prompts, captions, tags, selection.
  • Optional memory (off by default) reduces repeats in random and list modes. Enable it to draw from a shared-across-instances, per-node bag of distinct candidate texts. Selected texts are removed; before the next draw, the bag refills if strictly less than 25% remains. With 60 distinct candidates, the first 46 draws are distinct, then the bag refills on draw 47. Refills avoid immediately repeating the last result when alternatives exist; earlier results can repeat after a refill.
  • Set cache_time to the selection memory idle timeout in seconds (default 7200 = 2 hours, minimum 1). Each draw restarts that bag’s timer. Bags idle for more than their configured timeout are removed on the next memory-enabled draw; restarting ComfyUI clears all bags. This controls selection memory, not ComfyUI’s output cache. Changing candidate texts or their order resets the bag; changing seed does not.
  • Memory is separated by workflow ID (when supplied), node ID, and selection mode. Set a different memory_id per workflow when workflow metadata is unavailable (for example, API submissions); otherwise workflows using the same node ID and mode can share memory. A nonempty memory_id overrides the workflow ID.
  • With memory enabled, duplicate indices/texts count as one candidate, and results depend on both seed and previous draws. Only actual node execution consumes a draw: keep changing seed between runs, since cached outputs do not advance the bag. fixed mode and memory-off selection retain their previous behavior.

  

• Random Text From File

  • Select text from .txt or .md files in input directory (supports comments with #, list items with -)

• Text List From File

  • Load all valid lines from a text file as a list

Info:

• Text File Info
  • Get information about text files (line count, etc.)

JSON:

• JSON Extractor

  • Extract value from JSON data structures
  • Supports x.y.z notation for nested keys
  • Useful for parsing metadata from files.

  

Primitives

Adapters:

• Scheduler Adapter
  • Adapter for ComfyUI scheduler compatibility
• Sampler Adapter
  • Adapter for ComfyUI sampler compatibility
• StringPass
  • Simple string passthrough utility

Integers:

• Int Offset

  • Add offset to integer values
  • Add or subtract offset based on parameter
  • Useful for seed offset from main seed.

  

• Integer Increment

  • Increment integer by a value

• Integer Decrement

  • Decrement integer by a value

Floats:

• Float Range List
  • Generate a list of float values within a range

Lists:

• Empty Image List - Create an empty list for accumulation
• Append to Image List - Add images to a list (handles batch tensors)
• Get Image from List - Extract single image by index
• Image List Slice - Extract range of images
• Image List Length - Get list size
• Wrap as List / Unwrap from List - Convert between single values and lists

Flow Control

Loop System:

• Loop Start / Loop End
  • Index-based iteration with up to 20 value slots
  • Featured: Preserves lists as-is - No auto-expansion, lists stay intact through iterations
  • Supports image list accumulation patterns
  • See docs/loop-system.md for implementation details

Specialized

SUPIR:

• SUPIR Parameters
  • Helper node for SUPIR upscaling workflow parameters

Installation

1. Navigate to your ComfyUI custom nodes directory:

cd ComfyUI/custom_nodes/

2. Clone this repository:

git clone https://github.com/yourusername/ComfyUI-XJNodes.git

3. Restart ComfyUI

Requirements

See requirements.txt for any additional dependencies.

License

This project is licensed under the GNU General Public License v3.0 (GPL-3.0).