DuoUmiWild - ComfyUI Wildcard Node

A powerful custom node for ComfyUI that supports wildcards from .txt files, advanced YAML-based selections with tags, {} randomization, and prefix/suffix injection.

Installation

1. Navigate to your ComfyUI custom nodes directory:

   cd ComfyUI/custom_nodes/
   

2. Clone or copy this repository:

   git clone https://github.com/yourusername/DuoUmiWild.git
   

3. Restart ComfyUI

The node will appear under the DuoUmiWild category in the node menu.

Usage

Basic Wildcard Syntax

Use __filename__ in your prompt to randomly select a line from wildcards/filename.txt:

a photo of __subject__, __style__

This will:

• Read a random line from wildcards/subject.txt
• Read a random line from wildcards/style.txt
• Replace the wildcards with the selected values

Range Selection Syntax

Select multiple random items from a wildcard file:

• __2$$colors__ - Select exactly 2 random colors
• __1-3$$animals__ - Select between 1 and 3 random animals
• __-5$$tags__ - Select up to 5 random tags
• __2-$$items__ - Select at least 2 random items

Examples:

__3$$artist__ style painting
__1-2$$mood__, __2-4$$colors__

Creating Wildcard Files

1. Create a wildcards folder in the DuoUmiWild directory (it's created automatically on first run)

2. Create .txt files with one option per line:

wildcards/subject.txt:

beautiful woman
handsome man
cute cat
majestic dragon

wildcards/style.txt:

photorealistic
anime style
oil painting
watercolor
digital art

3. Use comments to document your wildcards (lines starting with # are ignored):

# Portrait subjects
beautiful woman
handsome man
# Fantasy subjects
elf warrior
dwarf blacksmith

Using the Seed Parameter

The seed parameter ensures reproducible results:

• Same seed = same random selections
• Different seed = different random selections
• Connect to your KSampler's seed for consistent generations

Nested Folder Support

Organize wildcards in subfolders for better organization:

wildcards/
├── characters/
│   ├── heroes.txt
│   └── villains.txt
└── environments/
    ├── indoor.txt
    └── outdoor.txt

Use with folder path:

__characters/heroes__ fighting __characters/villains__ in __environments/indoor__

Or just use the filename (ignores folder structure):

__heroes__ vs __villains__

Recursive/Nested Wildcards

Wildcard files can contain other wildcards that will be expanded:

wildcards/combo.txt:

__subject__ in __style__
__quality__ portrait

Using __combo__ will expand the wildcards inside!

YAML Wildcards

DuoUmiWild supports advanced YAML-based wildcards with tag selection and prefix/suffix injection.

Tag-Based Selection:

<[Pose]>              # Select any entry tagged "Pose"
<[Hat][Fancy]>        # Select entries with BOTH "Hat" AND "Fancy" tags
<[Hat|Headband]>      # Select entries with "Hat" OR "Headband" tag

Curly Brace Randomization:

{option1|option2|option3}        # Randomly pick one
{2$$a|b|c|d}                     # Pick exactly 2
{1-3$$red|blue|green}            # Pick 1 to 3

Combined Example:

girl, {a|b|c|d|e}-size, <[Pose]>, wearing {<[Hat]>|<[Headband]>}

See YAML_GUIDE.md for complete YAML documentation!

Curly Brace Randomization

Use {} with | pipes to randomly select options:

{happy|sad|neutral} expression
{0-1$$dark skinned female}
wearing {red|blue|green} dress

Nodes Included

1. Wildcard Prompt Node

The main wildcard processing node.

Inputs:

• text: Your prompt with wildcards in __filename__ format, YAML tags <[Tag]>, and {} randomization
• seed: Random seed for reproducible wildcard selection (0 to max int)
• autorefresh:
  • No (default): Cache wildcard files for faster processing
  • Yes: Reload files each time (slower, but see edits immediately)

Outputs:

• processed_text: Your prompt with all wildcards replaced with random selections
• UI Display: The node displays the processed text as copyable string in the UI

2. Latent Ratio Selector Node

Creates empty latent images with predefined aspect ratios.

Inputs:

• ratio_selected: Choose from portrait, landscape, square, or cinematic ratios
• batch_size: Number of latent images (1-64)
• randomize:
  • No: Use the manually selected ratio
  • Yes: Randomly select a ratio (ignores ratio_selected)
• randomize_from: When randomize is Yes, select from:
  • All: Any ratio
  • Portrait Only: Only portrait ratios (2:3, 3:4, 4:5, 9:16)
  • Landscape Only: Only landscape ratios (16:9, 21:9, etc.)
  • Square Only: 1:1 ratio
• seed: Seed for random ratio selection

Outputs:

• latent: Empty latent tensor
• ratio_used: String describing which ratio was used
• width: Image width in pixels
• height: Image height in pixels

Available Ratios:

• Portrait: 2:3 (832x1248), 3:4 (880x1176), 4:5 (912x1144), 9:16 (768x1360)
• Square: 1:1 (1024x1024)
• Landscape: 4:3 (1176x880), 16:9 (1360x768), 21:9 (1536x640), and more
• Cinematic: IMAX (1.43:1), Cinemascope (2.35:1), Anamorphic (2.39:1)
• Special: Golden Ratio (1.618:1)

Examples

Simple Example

Input:

a __quality__ photo of a __subject__

Wildcard files:

• wildcards/quality.txt: high-quality, stunning, professional
• wildcards/subject.txt: cat, dog, bird

Possible outputs:

• "a stunning photo of a cat"
• "a professional photo of a bird"
• "a high-quality photo of a dog"

Advanced Example

Input:

__1-2$$mood__ portrait of __subject__, __2-3$$style__, __lighting__

Possible output:

dramatic, moody portrait of elegant woman, oil painting, renaissance style, soft lighting

Features

• ✅ Random line selection from .txt files
• ✅ Range-based selection (select multiple items)
• ✅ Seeded randomization for reproducibility
• ✅ Comment support in wildcard files (lines starting with #)
• ✅ Inline comment removal
• ✅ Recursive/nested wildcard support (wildcards within wildcards)
• ✅ Nested folder organization for wildcard files
• ✅ YAML support with tag-based selection <[Tag]>
• ✅ YAML Prefix/Suffix random injection
• ✅ Curly brace randomization {option1|option2}
• ✅ Combined syntax support - Mix wildcards, YAML, and {}
• ✅ Latent ratio selector with randomization
• ✅ 17 preset aspect ratios (portrait, landscape, cinematic)
• ✅ Text preview as copyable string in the UI
• ✅ File caching with optional auto-refresh
• ✅ Automatic comma formatting
• ✅ Clean output (removes extra commas and whitespace)

File Structure

DuoUmiWild/
├── __init__.py          # Node registration
├── wildcard_node.py     # Main node implementation
├── README.md            # This file
└── wildcards/           # Your wildcard .txt files
    ├── subject.txt
    ├── style.txt
    ├── colors.txt
    └── ...

Tips

1. Organize your wildcards: Create separate files for different categories (subjects, styles, colors, moods, etc.)

2. Use descriptive filenames: Name files clearly so you can easily remember them in prompts

3. Comment your files: Add comments to organize and document your wildcard collections

4. Test with different seeds: Try different seed values to explore various combinations

5. Combine with other nodes: Connect the output to any node that accepts text/string input

Troubleshooting

Wildcard not found:

• Check that the file exists in the wildcards/ directory
• Ensure the filename matches exactly (case-sensitive on some systems)
• Verify the file has a .txt extension

No output / empty result:

• Check that your wildcard files aren't empty
• Ensure lines aren't all comments
• Look at the console for error messages

Same result every time:

• Verify you're changing the seed value
• Or connect to a random seed generator node

License

MIT License - Feel free to use and modify as needed.

Contributing

Contributions are welcome! Please feel free to submit pull requests or open issues on GitHub.

Credits

Inspired by the original UmiAI wildcard system for Stable Diffusion WebUI.