🔧 SD3.5 Empty Latent Size Picker

A utility node for generating empty latent tensors in Stable Diffusion v3.5-compatible resolutions. This node allows for custom batch sizes, width/height overrides, and inverting aspect ratios, ensuring flexibility and compatibility in ComfyUI workflows.

Features

• Supported Resolutions: Choose from a predefined list of Stable Diffusion v3.5-compatible resolutions.
• Batch Size Control: Set a batch size to create multiple latent tensors at once.
• Width and Height Overrides: Use custom width and height values as needed, while maintaining Stable Diffusion compatibility constraints.
• Ratio Inversion Option: Quickly swap width and height to experiment with alternate aspect ratios.
• Multiple of 64 Constraint: Ensures output width and height are multiples of 64, which Stable Diffusion requires for optimal performance.

Parameters

• resolution: A dropdown list of pre-configured resolutions for Stable Diffusion v3.5, with a default of "1024x1024 (1.0)".
• batch_size: Integer value defining the number of latent tensors to generate. Default is 1.
• width_override and height_override: Optional custom dimensions. When set above 0, these values override the selected resolution dimensions.
• invert_ratios: Option to invert width and height for alternative aspect ratios.

Output

The node returns:

• LATENT: A tensor with dimensions [batch_size, 4, height // 8, width // 8], representing empty latent samples ready for further processing.
• width: Effective width, accounting for overrides and constraints.
• height: Effective height, accounting for overrides and constraints.

Example Code

Here’s an example of how to use this node programmatically:

import torch
from ComfyUI-SD3.5-Latent-Size-Picker.sd3_5_empty_latent import SD3_5EmptyLatent

# Initialize the SD3_5EmptyLatent node
latent_picker = SD3_5EmptyLatent()

# Set parameters
resolution = "1024x1024 (1.0)"
batch_size = 2
width_override = 768
height_override = 0
invert_ratios = "No"

# Execute to generate the latent tensor
latent_output, width, height = latent_picker.execute(
    resolution=resolution,
    batch_size=batch_size,
    width_override=width_override,
    height_override=height_override,
    invert_ratios=invert_ratios
)

print(f"Latent Shape: {latent_output['samples'].shape}, Width: {width}, Height: {height}")

Installation

To use this node in ComfyUI:

1. Clone the repository into the custom_nodes folder in ComfyUI:

   git clone https://github.com/mithamunda/ComfyUI-SD3.5-Latent-Size-Picker.git
   

2. Restart ComfyUI to enable the node in the interface.

License

This code is released under the MIT License, allowing for free use, distribution, and modification. See the LICENSE file for more details.