Skip to content

tetsuoo-online/Comfyui-TOO-Pack

BranchesTags

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

20 Commits
.github/workflows
.github/workflows
 
 
nodes
nodes
 
 
presets
presets
 
 
web
web
 
 
.gitignore
.gitignore
 
 
LICENSE
LICENSE
 
 
README.md
README.md
 
 
__init__.py
__init__.py
 
 
pyproject.toml
pyproject.toml
 
 
requirements.txt
requirements.txt
 
 

Repository files navigation

ComfyUI-TOO-Pack 🎨

Tetsuoo's custom nodes pack for ComfyUI

A collection of utility nodes designed to enhance your ComfyUI workflow with smart image loading, Krita integration, widget extraction, and more.

Category: TOO-Pack

📦 Installation

cd ComfyUI/custom_nodes
git clone https://github.com/tetsuoo-online/Comfyui-TOO-Pack

Requirements

  • ComfyUI (latest version recommended)
  • Python 3.8+
  • PyTorch
  • Pillow (PIL)
  • NumPy

🎯 Nodes Overview

Node Category Description
Smart Image Loader TOO-Pack/image Flexible image loader with multiple sources
Smart Image Saver TOO-Pack/image Intelligent saver with flexible naming and metadata
Krita Bridge TOO-Pack/image Auto-load images from Krita
Extract Widget From Node TOO-Pack/utils Extract widget values from workflow nodes
Collection Categorizer TOO-Pack/utils Categorize files with local LLM (Ollama)

📚 Node Documentation

Smart Image Loader 🖼

Click to expand full documentation

A flexible image loader that supports multiple input sources with priority order.

Features

  • Multiple sources: txt, direct path, directory, or direct image
  • Smart priority order with configurable inputs
  • Random selection with reproducible seed
  • Multiple formats supported (PNG, JPG, JPEG, BMP, WEBP, TIFF)
  • Returns file path of loaded image

Parameters

Required Parameters

Parameter Type Description Default
seed INT Seed for reproducible random selection 0

Optional Parameters

Parameter Type Description Default
txt_path STRING Path to text file containing image paths -
img_path STRING Direct path to an image file -
img_directory STRING Path to directory containing images -
image IMAGE Direct image input -

Outputs

Parameter Type Description
IMAGE IMAGE The loaded image as tensor
FILE_PATH STRING Path of loaded file ("external_input" or "none")

Priority Order

The node loads images in this order (from highest to lowest priority):

  1. txt_path 📄 : Randomly selects a path from text file
  2. img_path 🖼 : Loads the specific image file
  3. img_directory 📁 : Randomly selects an image from directory
  4. image ⚡ : Uses the provided image input

Usage Examples

Case 1: Text file with image list

# Content of image_list.txt:
# /path/to/image1.png
# /path/to/image2.jpg
# /path/to/image3.webp

txt_path = "/path/to/image_list.txt"
seed = 42  # Reproducible

Case 2: Direct path to image

img_path = "/path/to/specific/image.png"

Case 3: Image directory

img_directory = "/path/to/images/"
seed = 123

Case 4: Combination with priority

txt_path = "/path/to/list.txt"       # Priority 1
img_path = "/path/to/fallback.png"   # Priority 2 (if txt_path fails)
img_directory = "/path/to/backup/"   # Priority 3 (if img_path fails)

Technical Details

Text file format (txt_path)

The text file should contain one image path per line:

/home/user/images/photo1.png
/home/user/images/photo2.jpg
/home/user/images/photo3.webp
  • Empty lines are ignored
  • UTF-8 encoding supported
  • A random path is selected using the seed

Seed handling

  • seed = 0: Different random selection each run
  • seed > 0: Identical selection with same parameters (reproducible)

FILE_PATH values

Value Description
Full path Image loaded from txt_path, img_path or img_directory
"external_input" Image provided via image parameter
"none" No valid source found (error)

Smart Image Saver 💾

Click to expand full documentation

An intelligent image saver that replaces the SAVE_IMG subgraph with flexible filename customization.

Features

  • Flexible naming: Customizable prefix, suffix, and separator
  • Date tokens: YYYY, MM, DD, HH, mm, ss, timestamp support
  • Smart metadata extraction: Auto-extract seed and model name from workflow
  • Node targeting: Target nodes by class name or direct ID (#10)
  • Multiple formats: WEBP (lossy/lossless), PNG, JPG/JPEG
  • Metadata preservation: Saves prompt and workflow in EXIF/PNG info

Parameters

Required Parameters

Parameter Type Description Default
images IMAGE Images to save -
output_folder STRING Output folder (supports date tokens) YYYY-MM-DD
prefix STRING Filename prefix (supports date tokens) ComfyUI_YYYY-MM-DD_HHmmss
seed_node_name STRING Node containing seed (class name or #ID) KSampler
model_node_name STRING Node containing model (class name or #ID) CheckpointLoaderSimple
suffix STRING Filename suffix (supports date tokens) ""
output_format COMBO Image format webp
quality INT Compression quality (1-100) 97

Outputs

Parameter Type Description
images IMAGE Pass-through of input images
filepath STRING Path of the first saved file

Filename Structure

The node builds filenames in this order:

[prefix]_[seed]_[model_name]_[suffix].[format]

Elements are joined by the separator. Empty elements are skipped.

Usage Examples

Case 1: Basic usage with date folder

output_folder = "YYYY-MM-DD"           # → 2024-01-30/
prefix = "render_YYYYMMDD"              # → render_20240130
seed_node_name = "KSampler"
model_node_name = "CheckpointLoaderSimple"

Output: 2024-01-30/render_20240130_123456_mymodel.webp

Case 2: Target node by ID

seed_node_name = "#10"                  # Target node with ID 10
model_node_name = "#5"                  # Target node with ID 5

Case 3: Disable seed or model

seed_node_name = ""                     # Don't include seed
model_node_name = ""                    # Don't include model
prefix = "my_render"
suffix = "HHmmss"                       # Add timestamp

Output: my_render_143025.webp

Date Tokens

All date tokens are replaced with current date/time values:

Token Description Example
YYYY Year (4 digits) 2024
MM Month (2 digits) 01
DD Day (2 digits) 30
HH Hour 24h (2 digits) 14
mm Minute (2 digits) 30
ss Second (2 digits) 25
timestamp Unix timestamp 1706623825

Node Targeting

By Class Name:

seed_node_name = "KSampler"              # Finds any KSampler node

By Node ID:

seed_node_name = "#10"                   # Targets node with ID 10

Krita Bridge 🎨

Click to expand full documentation

Automatically loads the latest image from the input/krita/ folder for seamless Krita integration.

Features

  • Automatic loading of latest image from krita folder
  • Real-time detection of new files
  • Auto-update during generation
  • Alpha support: extracts alpha channel as mask
  • Auto-create folder if doesn't exist
  • No parameters - works directly

Parameters

Required Parameters

No parameters required - The node works automatically!

Outputs

Parameter Type Description
image IMAGE The loaded image (RGB)
mask MASK Mask extracted from alpha channel

Setup

Working Directory

The node automatically loads from:

ComfyUI/input/krita/

If the folder doesn't exist, it's created automatically on first launch.

Export from Krita

  1. Open Krita
  2. Create or modify your image
  3. Export as PNG: File > Export
  4. Destination: ComfyUI/input/krita/filename.png
  5. The node automatically detects the new file

Recommended Workflow

Krita → Export PNG → ComfyUI/input/krita/ → Krita Bridge → [your workflow]

Usage Examples

Case 1: Simple loading

# No configuration needed
# Node automatically loads latest image

Case 2: Using alpha mask

# Connect 'mask' to a mask node (Mask Composite, etc.)
# Krita image's alpha channel becomes usable mask

Case 3: Iterative Krita ↔ ComfyUI workflow

1. Draw in Krita
2. Export → input/krita/sketch.png
3. ComfyUI detects and generates
4. Get result
5. Refine in Krita
6. Re-export → Node loads new version

Case 4: Inpainting with Krita mask

# Workflow:
# Krita Bridge (image + mask) → Inpaint Model → VAE Decode
# Krita's alpha mask defines inpainting area

Technical Details

File Detection

The node:

  1. Scans input/krita/ folder for all .png files
  2. Finds file with most recent modification
  3. Compares timestamp with last loaded file
  4. Reloads if change detected

Alpha Channel Handling

RGBA image (with transparency):

  • RGB channels → image output
  • Alpha channel → mask output (0-1 values)

RGB image (no transparency):

  • RGB → image output
  • Uniform white mask → mask output

Automatic Update

The IS_CHANGED function returns current timestamp, forcing ComfyUI to:

  • Re-evaluate node on every execution
  • Detect new files in real-time
  • Update image automatically

Krita Integration

Krita Configuration

  1. Set default export folder

    • Settings > Configure Krita > General
    • Set default folder: ComfyUI/input/krita/

  2. Keyboard shortcut for quick export

    • Settings > Configure Krita > Keyboard Shortcuts
    • Assign key to Export
    • Example: Ctrl+Shift+E

  3. Export format

    • Format: PNG
    • Compression: as preferred
    • Important: Enable "Save alpha channel" if using mask

Optimal Workflow

┌─────────┐    Export PNG    ┌──────────────┐
│  Krita  │ ───────────────> │ input/krita/ │
└─────────┘                   └──────────────┘
                                      │
                                      ▼
                              ┌──────────────┐
                              │ Krita Bridge │
                              └──────────────┘
                                      │
                     ┌────────────────┴────────────────┐
                     ▼                                 ▼
                ┌────────┐                        ┌──────┐
                │ image  │                        │ mask │
                └────────┘                        └──────┘

Extract Widget From Node 🔧

Click to expand full documentation

Extracts specific widget values from any node in the ComfyUI workflow.

Features

  • Targeted extraction of specific widgets by name
  • Compatible with all ComfyUI nodes
  • Multiple extraction: extract several widgets at once
  • Auto mode: extracts all widgets if none specified
  • Smart handling of dictionaries and nested values
  • "on" filter: ignores disabled widgets

Parameters

Required Parameters

Parameter Type Description Default
node_name STRING Node type name to target (e.g., "Power Lora Loader") Power Lora Loader
widget_names STRING Widget names to extract (comma-separated) lora_name, strength_model

Hidden Parameters

Parameter Type Description
extra_pnginfo EXTRA_PNGINFO Workflow PNG metadata
prompt PROMPT Current workflow data
unique_id UNIQUE_ID Node unique ID

Outputs

Parameter Type Description
STRING STRING Extracted values (one per line, separated by \n)

Usage Examples

Case 1: Extract specific widgets from Lora Loader

node_name = "Power Lora Loader"
widget_names = "lora_name, strength_model"

Output:

my_lora_v1.safetensors
0.85

Case 2: Extract multiple parameters from KSampler

node_name = "KSampler"
widget_names = "seed, steps, cfg"

Output:

123456789
20
7.5

Case 3: Extract all widgets (auto mode)

node_name = "CheckpointLoaderSimple"
widget_names = ""  # Empty = extract all

Case 4: Extract from multiple nodes of same type

node_name = "Power Lora Loader"
widget_names = "lora_name"

Output (if 3 Lora Loaders in workflow):

lora1.safetensors
lora2.safetensors
lora3.safetensors

Technical Details

Node search

The node performs a case-insensitive search on node_name:

  • "power lora" will find "Power Lora Loader"
  • "ksampler" will find "KSampler" and "KSamplerAdvanced"

widget_names format

Widget names must be comma-separated:

"widget1, widget2, widget3"

Spaces are automatically stripped.

Nested values handling

The node intelligently handles nested dictionaries:

Simple structure:

{
  "lora_name": "my_lora.safetensors",
  "strength_model": 0.85
}

Nested structure (Power Lora Loader):

{
  "loras": {
    "on": true,
    "lora_name": "my_lora.safetensors",
    "strength_model": 0.85
  }
}

The node automatically extracts values from both structures.

"on" filter

If a dictionary contains "on": false, its values are ignored.

Advanced Use Cases

1. Extract prompts from workflow

node_name = "CLIPTextEncode"
widget_names = "text"

2. Retrieve used seeds

node_name = "KSampler"
widget_names = "seed"

3. List all loaded models

node_name = "CheckpointLoader"
widget_names = "ckpt_name"

4. Extract control parameters

node_name = "ControlNetLoader"
widget_names = "control_net_name, strength"

Collection Categorizer (LLM) 🗂

Click to expand full documentation

A ComfyUI node that automatically scans your folders and categorizes content with a local LLM (Ollama).

Features

  • Automatic scanning of folders (video files, archives, documents)
  • Smart categorization via local LLM (Ollama)
  • Recursive scanning of subfolders (optional)
  • Reproducible seed for identical results
  • Custom Ollama models supported
  • Auto-save JSON output
  • Compatible with Collection Manager
  • 100% local - no external APIs

Parameters

Main Parameters

Parameter Type Description Default
ollama_model COMBO LLM model to use (or "custom") qwen2.5:7b
custom_ollama_model STRING Model name if "custom" selected -
folder_path STRING Path to folder to scan -
scan_subfolders BOOLEAN Recursively scan subfolders False
save_json BOOLEAN Auto-save JSON output True
collection_title STRING Collection title Ma Collection
content_type COMBO Content type (or "custom") films

Optional Parameters

Parameter Type Description Default
custom_type_name STRING Type name if "custom" selected -
custom_categories STRING Custom categorization criteria (multiline) (empty = LLM decides)
seed INT Seed for reproducible results 0 (random)

Outputs

Parameter Type Description
json_output STRING The complete collection JSON
summary STRING Categorization summary

Usage Examples

Case 1: Movies with automatic categories

content_type = "films"
custom_categories = ""  # Empty
# LLM decides categories (Genre, Era, etc.)

Case 2: TV Series with custom criteria

content_type = "series"
custom_categories = "Genre, Year, Studio"
# LLM categorizes by these criteria

Case 3: Custom type

content_type = "custom"
custom_type_name = "Documentaries"
custom_categories = "Theme, Duration"

Case 4: Custom model

ollama_model = "custom"
custom_ollama_model = "mistral:7b"

Case 5: Reproducible results

seed = 42
# Always same result with same parameters

Installation

Requirements

  1. Ollama installed and running

    # Download: https://ollama.ai
ollama --version

  2. Python requests (for local HTTP API)

    pip install requests --break-system-packages

  3. At least one LLM model

    ollama pull qwen2.5:7b

Supported File Types

Videos .mp4, .mkv, .avi, .mov, .wmv, .flv

Archives .cbz, .cbr, .zip, .rar

Documents .epub, .pdf, .mobi

Folders Subfolders are treated as individual items (unless scan_subfolders is enabled)

Recommended Models

Model Size Speed Quality Usage
qwen2.5:7b 7B ⚡⚡⚡ ⭐⭐⭐⭐ Recommended
gemma3:12b 12B ⚡⚡ ⭐⭐⭐⭐⭐ Best quality
llama3.1:8b 8B ⚡⚡⚡ ⭐⭐⭐⭐ Very reliable
gemma3:4b 4B ⚡⚡⚡⚡ ⭐⭐⭐ Fast

Output Format (JSON)

{
  "title": "My Collection",
  "icon": "🎬",
  "type": "Films",
  "filename": "films.json",
  "categories": [
    {
      "id": 1,
      "name": "Science Fiction",
      "subcategories": [],
      "games": ["Blade Runner", "The Matrix"]
    },
    {
      "id": 2,
      "name": "Comedy",
      "subcategories": [],
      "games": ["Superbad", "The Hangover"]
    }
  ]
}

🗺 ToDo

  • Extract Widget From Node : aiming at nodes using name, title or ID

📄 License

MIT License

👤 Author

Tetsuoo

🙏 Credits

  • Claude AI ❤ - AI assistant extraordinaire
  • ComfyUI - Amazing node-based interface
  • Ollama - Local LLM runtime
  • Krita - Open-source digital painting software

About

Custom nodes pack for ComfyUI

Resources

Readme

License

MIT license
Activity

Stars

0 stars

Watchers

0 watching

Forks

1 fork
Report repository

Releases

No releases published

Packages

No packages published

Languages