Nodes/ComfyUI-Tensor-Prism-Node-Pack/Noise Injection Merge (Tensor Prism)
ComfyUI Node

Noise Injection Merge (Tensor Prism)

A ComfyUI node in Tensor_Prism/Merge with 13 inputs and 1 output.

By Arctenox·Created 11 months ago·Updated 6 months ago· 2
Noise Injection Merge (Tensor Prism)
  • model_A
  • model_B
  • merged_model
merge_ratio0.50
noise_strength0.050
noise_patterngaussian
seed42
layer_scaling_factor1.5
adaptive_threshold0.30
noise_decay0.90
perlin_frequency2.0
focus_attentiontrue
focus_mlpfalse
preserve_normstrue
CategoryTensor_Prism/Merge

Inputs (13)

NameTypeDefaultDescription
model_AMODELFirst model to merge
model_BMODELSecond model to merge
merge_ratioFLOAT0.500–1Base merge ratio (0=full A, 1=full B)
noise_strengthFLOAT0.0500–0.5Overall noise injection strength
noise_patternCOMBOgaussianType of noise pattern to inject
seedINT420–18446744073709550000Random seed for reproducibility
layer_scaling_factoroptFLOAT1.50.5–5Scale factor for layer-based noise (deeper=more noise)
adaptive_thresholdoptFLOAT0.300–1Similarity threshold for adaptive noise
noise_decayoptFLOAT0.900.5–1Decay factor per layer (1.0=no decay)
perlin_frequencyoptFLOAT2.00.5–10Frequency for Perlin noise
focus_attentionoptBOOLEANtrueFocus noise on attention layers
focus_mlpoptBOOLEANfalseFocus noise on MLP layers
preserve_normsoptBOOLEANtruePreserve tensor norms after injection

Outputs (1)

NameTypeDescription
merged_modelMODEL