Create Empty Latents
Generates a zero-filled latent tensor of the shape expected by the connected pipeline.
Category: ModularDiffusion/Create
TL;DR
- Same shape behavior as Create Noise Latents, but the tensor is all zeros (no random seed).
- Useful as the destination for Latents Composite Mask, or as a starting point when you want fully controlled noise injection downstream.
- For ordinary text-to-image / text-to-video generation you almost always want Create Noise Latents instead.
Typical workflow position
Pipeline Builder → [Create Empty Latents] → Latents Composite Mask → Generate Media Latents
Node preview

Inputs
| Name | Type | Required | Notes |
|---|---|---|---|
pipeline |
Pipeline Config |
Yes | Determines latent shape and whether num_frames is shown. |
Outputs
| Name | Type | Notes |
|---|---|---|
output_latent |
LatentArtifact |
Zero-filled latent tensor in the pipeline's canonical latent space. |
Parameters
| Name | Type | Default | Notes |
|---|---|---|---|
width |
int (pixels) | 1024 |
Pixel-space width. |
height |
int (pixels) | 1024 |
Pixel-space height. |
num_frames |
int | 41 |
Number of video frames. Hidden for image pipelines. |
No seed parameter — output is deterministic by construction.
Tips & pitfalls
- Pair with noise before generating. An all-zero latent needs noise before denoising — enable
add_noise=Trueon Generate Media Latents, or inject noise via composite / math nodes first. Without noise it has nothing to denoise from.
See also
- Create Noise Latents — the usual text-to-image / text-to-video starting tensor.
- Latents Composite Mask — primary consumer.