Skip to content

Add Latents

Elementwise sum of two latent tensors.

Category: ModularDiffusion/Transform

TL;DR

  • output = left_latent + right_latent elementwise.
  • Both inputs must share the same shape and latent space (i.e. come from the same pipeline type).
  • Common uses: blending two denoised latents, injecting controlled noise, residual additions between stages.

Typical workflow position

Generate Media Latents (A) ─┐
                            ├─→ [Add Latents] → Generate / Decode
Generate Media Latents (B) ─┘

Node preview

Add Latents

Inputs

Name Type Required Notes
left_latent LatentArtifact Yes
right_latent LatentArtifact Yes Must match left_latent shape.

Outputs

Name Type Notes
output_latent LatentArtifact left + right.

Tips & pitfalls

  • Both inputs must have the same shape. Resize or upsample one of the inputs first if they differ.
  • Keep latents within the same pipeline family. Each model uses a different latent space — mixing latents across families (e.g. Flux and SDXL) produces meaningless results.

See also