ComfyUI-SDXL-EmptyLatentImage

A simplified ComfyUI extension node that provides resolution selection with usage statistics for generating empty latent images.<br> <img src="misc/ss_resolution_list.jpg" alt="Node image" style="width:300px; height:auto;">

Features

Core Functionality

• Smart Resolution Selection: Choose from pre-defined resolutions loaded from JSON files
• Category-based Organization: Automatic categorization with priority order: Custom → SDXL → SD15
• Resolution Display: Format [Category] Width x Height (aspect_ratio) with decimal aspect ratios
• Channel Selection: Support for both 4-channel (SDXL/SD1.5) and 16-channel (SD3/Flux) latent generation

Usage Statistics & Visual Feedback

• Usage Tracking: Automatic recording of resolution usage frequency
• Visual Indicators: Dynamic marks displayed after resolution labels:
  • ★ Favorite resolutions (manually registered)
  • 🔥 Frequently used resolutions (≥3 uses)
  • 🕒 Recently used resolutions (last 5 used)

Simple Configuration

• Environment Variables: Override settings with SDXL_* environment variables
• Basic JSON Config: Simple config.json with essential settings only

Resolution Display Order

Resolutions are displayed in the following priority order:

1. Custom resolutions (user-defined)
2. SDXL resolutions
3. SD15 resolutions

Within each category, resolutions are sorted by aspect ratio (portrait → square → landscape).

JSON Format

Resolution files should follow this structure:

[
    {
        "width": 1024, "height": 1024
    },
    {
        "width": 832, "height": 1152
    },
    {
        "width": 1152, "height": 832
    }
]

Pre-included Resolution Sets

• sdxl_resolution_set.json: Optimal SDXL training resolutions
• sd_resolution_set.json: Standard SD1.5 resolution set
• custom_resolution_set.json: Your custom resolutions (create this file to add your own)

Configuration

Environment Variables

Override default settings:

• SDXL_MAX_RESOLUTION: Maximum allowed resolution (default: 8192)
• SDXL_MIN_RESOLUTION: Minimum allowed resolution (default: 64)
• SDXL_MAX_BATCH_SIZE: Maximum batch size (default: 64)
• SDXL_DEFAULT_CHANNELS: Default channel count - 4 or 16 (default: 4)
• SDXL_TRACK_USAGE: Enable/disable usage tracking (default: true)

config.json (Optional)

Create or modify config.json for persistent settings:

{
  "max_resolution": 8192,
  "min_resolution": 64,
  "max_batch_size": 64,
  "default_channels": 4,
  "track_usage": true
}

Managing Favorite Resolutions

Adding Favorites

To mark resolutions as favorites, manually edit usage_stats.json:

{
  "favorites": [
    "[CUSTOM] 1024 x 1024 (1.00)",
    "[SDXL] 832 x 1216 (0.68)"
  ],
  "usage_count": {},
  "recent": []
}

Usage Statistics

The extension automatically tracks:

• Usage Count: Number of times each resolution is used
• Recent History: Last 5 used resolutions
• Favorites: Manually marked favorite resolutions

Example usage_stats.json after usage:

{
  "favorites": [
    "[CUSTOM] 1024 x 1024 (1.00)"
  ],
  "usage_count": {
    "[SDXL] 1024 x 1024 (1.00)": 5,
    "[SDXL] 832 x 1216 (0.68)": 2
  },
  "recent": [
    "[SDXL] 832 x 1216 (0.68)",
    "[SDXL] 1024 x 1024 (1.00)"
  ]
}

Install

1. Navigate to the custom_nodes folder where ComfyUI is installed
2. Clone the repository:

git clone https://github.com/shingo1228/ComfyUI-SDXL-EmptyLatentImage

3. (Optional) Add custom resolutions by creating custom_resolution_set.json:

cd ComfyUI-SDXL-EmptyLatentImage
# Create your custom resolution file
echo '[{"width": 1024, "height": 1024}, {"width": 512, "height": 768}]' > custom_resolution_set.json

4. (Optional) Customize settings in config.json
5. Restart ComfyUI to load the extension

Channel Support

This extension supports multiple latent channel configurations:

4-Channel Mode (Default)

• Compatible with: SDXL, SD1.5, and similar models
• Tensor shape: [batch_size, 4, height/8, width/8]
• Use case: Traditional Stable Diffusion workflows

16-Channel Mode

• Compatible with: SD3, Flux, and newer models
• Tensor shape: [batch_size, 16, height/8, width/8]
• Use case: Next-generation AI models requiring expanded latent space

Usage

Simply select "4" or "16" from the channels dropdown in the node interface. The default can be configured via environment variables or config.json.

Architecture

This extension uses a simplified single-class design:

• Single File: All functionality in sdxl_empty_latent.py
• Simple Caching: Basic file modification time checking
• Minimal Configuration: Environment variables + optional JSON config
• Essential Features Only: Focus on core resolution selection, usage tracking, and channel flexibility
• Multi-Model Support: Compatible with SDXL, SD1.5, SD3, Flux, and future models

The simplified architecture makes the code easy to understand, modify, and maintain while providing all essential features for resolution management in modern ComfyUI workflows.