Nodes/Skoogeer-Noise/Latent Channel Linear Transform
ComfyUI Node

Latent Channel Linear Transform

A ComfyUI node in latent/channel with 15 inputs and 1 output.

By ttulttul·Created 8 months ago·Updated 2 months ago· 14
Latent Channel Linear Transform
  • latent
  • mask
  • latent
operation
seed0
sign_flip_prob0.50
tile_size0
block_size0
alpha0.50
selection_modeall
selection_fraction1.00
selection_count0
selection_orderhighest
selection_indices
mix1.00
match_statsfalse
Categorylatent/channel

Inputs (15)

NameTypeDefaultDescription
latentLATENTLatent to transform in channel-space.
operationCOMBOLinear channel-space transform to apply.
seedINT00–18446744073709550000Seed for deterministic channel selection and random transforms.
sign_flip_probFLOAT0.500–1Probability of flipping channel signs when using signed_permute.
tile_sizeINT00–512When >0 and operation is signed_permute, apply a new permutation per tile (latent pixels).
block_sizeINT00–4096Channels per orthogonal block. 0 means full rotation.
alphaFLOAT0.50-4–4Shear strength for low_rank_shear.
selection_modeCOMBOallHow to choose which channels are transformed.
selection_fractionFLOAT1.000–1Fraction of channels to select when selection_count is 0.
selection_countINT00–4096Exact number of channels to select (overrides selection_fraction when >0).
selection_orderCOMBOhighestWhether to pick high or low variance/roughness channels.
selection_indicesSTRINGComma-separated channel indices to select when selection_mode=indices.
mixFLOAT1.000–1Blend factor for the transformed channels (0=off, 1=full).
match_statsBOOLEANfalseMatch per-channel mean/std after the transform (stabilizes latent stats).
maskoptMASKOptional mask to limit the transform to masked areas.

Outputs (1)

NameTypeDescription
latentLATENT