ComfyUI Node
😺NKD Klein Reference Weight
Controls the attention WEIGHT of one reference image, independently of the others — so a secondary reference can assert itself, or an over-dominant one can be diluted. (Distinct from Presampling's Reference Strength, which sets positional anchoring.) Pick the reference by index (0 = ref_0, 1 = ref_1, …), set its weight (1.0 = no change, <1 dilutes, >1 reinforces, up to 2), and optionally a per-step curve that multiplies it. Chain one node per reference you want to control; untouched references keep their default influence. Works by scaling that reference's attention tokens, so it's independent of resolution and reference count.
😺NKD Klein Reference Weight
- model
- model
◄reference_index1►
◄weight1.00►
◄schedule—►
Category😺NKD Nodes/Klein
Inputs (4)
| Name | Type | Default | Description |
|---|---|---|---|
| model | MODEL | — | |
| reference_index | INT | 10–7 | Which reference this node controls. 0 = the main reference (ref_0), 1 = ref_1, … in the order added. |
| weight | FLOAT | 1.000–2 | Attention weight for this reference. 1.0 = no change, below 1 dilutes (0 = removed), above 1 reinforces (up to 2). A connected curve multiplies this per step; with no curve it applies flat across all steps. |
| scheduleopt | FLOAT | Optional per-step curve in [0,1] for THIS reference that MULTIPLIES its weight (e.g. the FLOAT output of NKD Sigmas Curve). With weight 2.0, a 1→0.5 curve runs the effective weight from 2.0 to 1.0, a 1→0 curve from 2.0 to 0 (removed). Leave unconnected for flat weight. |
Outputs (1)
| Name | Type | Description |
|---|---|---|
| model | MODEL | — |