EmptySD3LatentImage

The EmptySD3LatentImage node is a vital starting point for Stable Diffusion 3 (SD3) workflows in ComfyUI, initializing a blank latent image tensor in the SD3 latent format. This node helps ensure compatibility and provides a controlled, uniform base for prompt-driven art generation and testing.

Overview

EmptySD3LatentImage generates an empty (uniform value or zeros) latent tensor tailored to the SD3 model’s requirements. It is commonly used to kickstart image synthesis workflows where a base latent space is needed, especially for testing, reproducibility, or workflows that do not depend on a real input image. It interacts mainly as the source for sampler, decoder, and image transformation nodes downstream.

Visual Example

EmptySD3LatentImage Comfyui node
Figure 1 - EmptySD3LatentImage Comfyui node

Official Documentation Link

https://comfyui.dev/docs/guides/Nodes/emptysd3latentimage/

Inputs

Parameter Data Type Input Method Default
width Integer Direct input or slider 1024
height Integer Direct input or slider 1024
batch_size Integer Direct input 1

Outputs

Output Name Data Type Description
LATENT Tensor/Dict Generated empty SD3-compatible latent tensor (16-channel, filled with constant or zeros, batchable)

Usage Instructions

  1. Add EmptySD3LatentImage from the node browser to your workspace.
  2. Set width and height (multiples of 64, typically 1024).
  3. Set batch_size as needed (default 1 for single image).
  4. Connect LATENT output to a KSampler or DecodeSD3 node.
  5. Run the workflow. The node will produce an empty latent starting point for image generation.

Advanced Usage

  • Batch generate latent images for parallel processing by increasing batch_size.
  • Experiment with aspect ratios for special effects or dataset matching.
  • Use with SD3-specific samplers for best results, ensuring dimensions are multiples of 64.
  • Combine with conditioning nodes for guided generation.

Example JSON for API or Workflow Export

{
  "id": "empty_sd3_latent_1",
  "type": "EmptySD3LatentImage",
  "inputs": {
    "width": 1024,
    "height": 1024,
    "batch_size": 2
  }
}

Tips

  • Always use image dimensions that are multiples of 64 for SD3 compatibility.
  • Larger batch sizes save time when initializing multiple workflows, but increase memory usage.
  • Troubleshooting generation issues? Ensure latent size matches model requirements.

How It Works (Technical)

The node constructs a latent image tensor (multi-dimensional array) of shape [batch_size, channels, height, width], filled with a constant (e.g., 0.0609) or zeros. The output matches the SD3 format (usually 16 channels). This tensor is passed downstream, where samplers or decoders use it as a canvas for generative operations.

Github Alternatives

  • Bobs Latent Optimizer for ComfyUI – custom nodes to generate optimized empty latent images for SD3 and other models, with automatic aspect ratio and dimension calculation.
  • liusida/ComfyUI-SD3-nodes – SD3-compatible nodes for ComfyUI including wrappers and improved usability for SD3 latent image tasks.
  • ComfyUI Hunyuan Latent Nodes – alternative latent generation nodes compatible with SD3, FLUX, Qwen and Hunyuan models, offering dimension auto-adjust and video support.

Videcool workflows

The EmptySD3LatentImage node is used in the following Videcool workflows:

FAQ

Q1: Can I use this node with non-SD3 models?
A1: It is tailored for SD3 latent format; other models may require their own format.
Q2: Why are image dimensions restricted to multiples of 64?
A2: This ensures compatibility with SD3 downsampling and upsampling layers.
Q3: What does "empty" mean in this context?
A3: The latent tensor is initialized with zeros or a small constant, providing a uniform starting point for generation.

Common Mistakes and Troubleshooting

  • Non-multiple-of-64 dimensions; will cause errors or unexpected behavior—always check!
  • Setting batch_size too high can exceed system memory/VRAM.
  • Using the node outside SD3 pipelines may result in incompatible latent format.
  • For blurry or non-diverse outputs, verify the sampler and prompt interactions downstream.

Conclusion

The EmptySD3LatentImage node is a foundation for SD3 image generation in ComfyUI, enabling reproducibility, testing, and creative control. Its reliable initialization of the SD3 latent space allows users to build complex workflows with clarity and precision.

More information