ComfyUI Node
Nunchaku FLUX.2 Klein Enhancer
Adjusts active FLUX.2 Klein conditioning embeddings.
Nunchaku FLUX.2 Klein Enhancer
- conditioning
- CONDITIONING
◄active_scale1.00►
◄per_token_whiten0.00►
◄norm_equalize0.00►
◄early_layer_scale1.00►
◄mid_layer_scale1.00►
◄late_layer_scale1.00►
◄preserve_original0.00►
◄active_end_override0►
◄deviceauto►
◄debugfalse►
CategoryNunchaku/FLUX.2 Klein/Enhancer/Text
Inputs (11)
| Name | Type | Default | Description |
|---|---|---|---|
| conditioning | CONDITIONING | — | |
| active_scale | FLOAT | 1.000–10 | Multiplier on every active-token embedding. 1.0 = unchanged. The model was trained on Qwen3's natural distribution; values far from 1.0 push it off-distribution. |
| per_token_whiten | FLOAT | 0.00-1–5 | Amplifies per-token deviation from the sequence mean: (x - mean)*(1+w) + mean. >0 widens spread, <0 compresses. Was called 'contrast' in v1. |
| norm_equalize | FLOAT | 0.000–1 | Blend each token toward the per-sequence mean L2 norm. Flattens magnitude variance — fights Qwen3's natural emphasis. 0 = no effect. |
| early_layer_scaleopt | FLOAT | 1.000–5 | Klein-specific. Scale the first Qwen3 layer slice (low-level / structural features). Klein conditioning stacks 3 layers along the embed dim; this targets the first. |
| mid_layer_scaleopt | FLOAT | 1.000–5 | Klein-specific. Scale the middle Qwen3 layer slice (intermediate semantic features). |
| late_layer_scaleopt | FLOAT | 1.000–5 | Klein-specific. Scale the last Qwen3 layer slice (high-level / abstract semantic features). |
| preserve_originalopt | FLOAT | 0.000–1 | Linear blend back the unmodified active region. 0.0 = full enhancement, 1.0 = no change. |
| active_end_overrideopt | INT | 00–512 | Override the active-region end. 0 = auto-detect from attention_mask, falls back to full sequence length if mask missing. |
| deviceopt | COMBO | auto | 2 options: auto, cpu |
| debugopt | BOOLEAN | false | — |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| CONDITIONING | CONDITIONING | — |