AntiMatter Nodes

AntiMatter Nodes for ComfyUI

AntiMatter is a ComfyUI custom node collection for practical image workflows.

Nodes

• Anti_aspect_ratio_master
• AntiMatter Style Preset Mixer
• Antimatter Text File Appender
• Batch Loader from folder
• LinePrompt_MasterLoad
• LinePrompt_MasterLoad_JSON

Node tree:

AntiMatter/Image/Anti_aspect_ratio_master
AntiMatter/Text/AntiMatter Style Preset Mixer
AntiMatter/Text/Antimatter Text File Appender
AntiMatter/Image/Batch Loader from folder
AntiMatter/Text/LinePrompt_MasterLoad
AntiMatter/Text/LinePrompt_MasterLoad_JSON

Anti_aspect_ratio_master

Creates an empty latent using Flux, Z-Image, and ERNIE-oriented presets, manual dimensions, or an input image size. It can also resize a connected image with its aspect ratio preserved, and returns the resized image for the next image node.

Inputs:

• source: from_preset, from_manual, or from_image
• preset: Flux, Z-Image, and ERNIE-oriented width x height presets
• manual_width, manual_height: manual dimensions when source is from_manual
• round_to: rounds dimensions to a selected multiple
• orientation: auto, portrait, landscape, or swap
• batch_size: latent batch size
• latent_channels: latent channel count, default 16 for Flux, Z-Image, and ERNIE-style workflows
• resize_image: enables proportional resizing of the connected image
• resize_scale: signed resize factor. Negative values reduce and positive values enlarge. For example, -0.7 keeps 70% of the source dimensions, -2 makes the image twice smaller (50%), and +2 enlarges it to 200%. 0 is not valid.
• upscale_method: nearest-exact, bilinear, area, bicubic, or lanczos; this uses the same resize backend as ComfyUI's built-in Image Scale By node
• image: optional image source for dimensions
• downsample_factor: latent downsample factor, usually 8 for VAE latent workflows

When source is from_image and resizing is enabled, width, height, and final_preset describe the actual resized image dimensions. round_to is deliberately not applied in that case, so it cannot distort the original aspect ratio.

Outputs:

• latent
• width
• height
• final_preset
• image_name
• resized_image: original image when resizing is disabled; proportionally resized image when enabled

AntiMatter Style Preset Mixer

This node replaces the eight-slot style subgraph in KREA2_BATCH_WILD. Each slot has its own enabled, style_file, and read_mode control, then the enabled selections are combined with optional input_text into one prompt string.

The bundled presets are deliberately independent from Style_evo; they live in styles/krea2_batch_wild inside this node package. Edit a .txt file or add folders and .txt files there freely. Restart ComfyUI after adding or renaming files so they appear in the dropdowns.

Read modes are random, random_no_repeat, sequential, and reverse. The selected_style_files output reports the preset files that supplied text during the current execution.

Antimatter Text File Appender

Saves incoming text to .txt or .json files. It can append normally, or replace a specific scene when a scene name/number is connected.

Inputs:

• text: manual text/string to save.
• text_input: optional connected string input. If connected, this is saved instead of text.
• scene_name: optional scene number/name for replacement workflows. Accepts values like 35, scene-035, or scene_35.
• directory: folder path on any disk.
• filename: output file name. If no extension is provided, the node adds .txt or .json.
• file_format: txt or json.
• prefix: optional prefix added before every saved text.
• blank_lines_between_entries: spacing between appended .txt entries.
• index_start, index_padding: controls {index} and {counter} prefix placeholders.
• json_indent: pretty-print indent for .json; set 0 for compact JSON.

Outputs:

• saved_text
• file_path

Scene replacement:

Connect Scene Name from LinePrompt_MasterLoad_JSON to scene_name when manually regenerating a scene. For .json, the node replaces a record with the same scene_name or index. For .txt, it replaces the block containing that scene marker; if no marker is found, it replaces the N-th text block.

Batch Loader from folder

Loads images directly from a local folder and outputs them as a ComfyUI IMAGE batch. It is built for repeated processing, dataset testing, prompt/model comparisons, and long workflows where manually uploading files one by one becomes slow.

Supported image formats:

• PNG
• JPG / JPEG
• WEBP
• BMP
• TIFF / TIF

Advantages

• Loads full batches from any local folder without manual upload steps.
• Supports random, sequential, and batch_freeze modes.
• Uses deterministic seed control for repeatable random selections.
• Supports no-repeat random selection, so images are not repeated until the pool is exhausted.
• Can repeat one chosen image across a batch with batch_freeze, useful for seed, prompt, model, LoRA, or sampler testing.
• Can scan nested folders with recursive mode, useful for organized datasets and project folders.
• Has reload and reset controls, so folder changes can be picked up without restarting ComfyUI.
• Can skip images by exact width, height, or both, which is useful for filtering already resized or already processed files.
• Can move processed images into a subfolder after loading, helping separate completed and pending files during long runs.
• Returns filename text without extension, useful for output naming, captions, logging, or metadata chains.
• Returns source format, useful when branching workflow logic or tracking dataset composition.
• Returns width and height, useful for routing images by dimensions.
• Shows optional previews in the ComfyUI UI.
• Handles EXIF orientation, so camera and phone images load in the expected direction.
• Converts images to RGB automatically for stable ComfyUI image tensors.
• Can resize later batch images to the first image size, allowing mixed-size folders to become a valid ComfyUI batch.
• Works with relative paths such as .uploading and full absolute paths.

Inputs:

• folder: source folder path. Relative paths are resolved from the current ComfyUI working directory.
• mode: random, sequential, or batch_freeze.
• batch_size: number of images to output.
• freeze_count: number of queue executions to reuse the current selection.
• seed: controls deterministic random selection.
• no_repeat_random: prevents repeated random picks until all eligible files are used.
• recursive: scans nested folders when enabled.
• reload: rescans the folder.
• reset: resets sequence, freeze, and random pool state.
• skip_exact_mode: enables dimension-based skipping.
• skip_exact_width: width value for skip filters.
• skip_exact_height: height value for skip filters.
• move_processed_to_subfolder: moves selected images after processing.
• processed_subfolder_name: destination folder for processed images.
• resize_to_first: resizes later batch images to the first image size.
• show_preview: shows selected images in the node UI.
• preview_max: maximum preview images to display.

Outputs:

• images: ComfyUI image batch.
• filename: selected filename without extension, or a comma-separated list for mixed batches.
• format: selected file extension, or a comma-separated list for mixed batches.
• width: output image width.
• height: output image height.

LinePrompt_MasterLoad

Loads prompt lines from .txt files. It supports manual text prefixing, selectable style files, sequential/random/reverse reading, freeze count, delimiter handling, and preview updates.

Inputs:

• enabled: when disabled, returns an empty string.
• input_text: optional prefix added before the selected line as input_text, selected_line.
• text_file: manual path to a .txt file.
• style_file: dropdown populated from the configured styles folder.
• lines_to_take: selects 1, 2, or 3 consecutive lines.
• read_mode: sequential, random, or from_end.
• freeze_iterations: keeps the selected line for N queue runs.
• delimiter: appends ,, /, ., or ; to the selected line if missing.

Outputs:

• text

The ComfyUI setting LinePrompt_MasterLoad styles txt folder controls the parent folder scanned recursively for style_file.

LinePrompt_MasterLoad_JSON

Loads scene prompts from JSON. It is designed for scene-based generation where manual rework should use the current scene number/name.

Inputs:

• json_input: optional connected JSON/string input.
• enabled: when disabled, returns empty strings.
• json_text: manual JSON content.
• json_file: path to a JSON file.
• scene_mode: sequential, manual, or row.
• manual_scene: scene number used in manual mode.
• row: direct 1-based row selection used in row mode.
• repeat_each_scene: number of queue runs per scene.
• after_last_scene: stop_empty or loop.

Outputs:

• prompt
• Scene Name

Manual Install

Clone this repository into your ComfyUI custom nodes directory:

cd ComfyUI/custom_nodes
git clone https://github.com/AntiMatterComfy/antimatter-nodes.git

Restart ComfyUI after installing.

ComfyUI Manager

For automatic installation through ComfyUI Manager's "Install Missing Custom Nodes", this repository must be discoverable by ComfyUI Manager. Use one of these routes:

• Publish the package to the ComfyUI Registry.
• Or submit a pull request to ComfyUI Manager that adds this repository to custom-node-list.json.

This repository includes pyproject.toml and node_list.json to make the node metadata easy for ComfyUI tools to scan.