Skip to content

Create Noise Latents

Generates a random-noise latent tensor matching the connected pipeline's expected shape — the typical starting point for text-to-image / text-to-video flows.

Category: ModularDiffusion/Create

TL;DR

  • Drop this between the Pipeline Builder and Generate Media Latents for any pure text-to-image / text-to-video flow.
  • width / height are pixel-space; the node handles VAE down-scaling internally.
  • num_frames appears only when the connected pipeline produces video (LTX, LTX2, WAN). It's hidden otherwise.
  • seed controls reproducibility — same seed + same pipeline + same dimensions → identical latent.

Typical workflow position

Pipeline Builder → [Create Noise Latents] → Generate Media Latents → Decode Media Latent

Node preview

Create Noise Latents

Inputs

Name Type Required Notes
pipeline Pipeline Config Yes From the Pipeline Builder. Determines shape, VAE scale factor, and whether num_frames is shown.

Outputs

Name Type Notes
output_latent LatentArtifact Unpacked, normalized (~N(0,1)) noise tensor. Feed to Generate Media Latents.

Parameters

Name Type Default Notes
width int (pixels) 1024 Pixel-space width. Internally divided by the VAE scale factor.
height int (pixels) 1024 Pixel-space height.
num_frames int 41 Number of video frames. Hidden for image pipelines.
seed int random Reproducibility.
num_inference_steps int 20 Only shown for SDXL — used by SDXL to scale the initial noise. Other pipelines ignore this.

Tips & pitfalls

  • width / height must respect VAE divisibility. Most VAEs require multiples of 8 or 16. Pick standard dimensions (512, 768, 1024, …) to keep shapes valid.
  • Same seed across pipelines ≠ same image. Latent shape and VAE space differ per model; the seed is only meaningful within one pipeline type.
  • For Image-to-Image or rediffusion, use Encode Media Latent instead. Encoding an existing image gives you a conditioned starting point; optionally feed it through Generate Media Latents with add_noise=True.

See also