ComfyUI Node: 😺NKD Klein Reference Weight

😺NKD Klein Reference Weight:

The NKDKleinReferenceWeight node is designed to provide precise control over the attention weight of individual reference images within a sequence. This node allows you to adjust the influence of each reference image independently, enabling you to either amplify or diminish their impact on the final output. This is particularly useful when you have multiple reference images and need to ensure that a secondary reference asserts itself or to dilute an over-dominant one. Unlike Presampling's Reference Strength, which focuses on positional anchoring, NKDKleinReferenceWeight modifies the attention tokens directly, making it independent of resolution and reference count. By scaling the attention key/value tokens of a reference image, this node offers a robust mechanism to fine-tune the strength of each reference image, ensuring that your creative vision is accurately represented in the output.

😺NKD Klein Reference Weight Input Parameters:

model

This parameter represents the model to which the reference weight adjustments will be applied. It is essential for the node to function as it provides the context in which the reference images are processed.

reference_index

This integer parameter specifies which reference image the node will control. The index starts at 0 for the main reference (ref_0) and increments for additional references. The default value is 1, with a minimum of 0 and a maximum of 7. This allows you to target specific reference images in the sequence for weight adjustments.

schedule

The schedule is an optional float input that allows you to apply a per-step curve to the reference weight. This curve, ranging from 0 to 1, multiplies the reference's weight dynamically across the denoise steps. For example, with a weight of 2.0, a curve from 1 to 0.5 will adjust the effective weight from 2.0 to 1.0. If left unconnected, the weight remains constant throughout the process.

weight

This float parameter sets the attention weight for the selected reference image. A value of 1.0 indicates no change, values below 1.0 dilute the reference's influence (with 0 removing it entirely), and values above 1.0 reinforce it, up to a maximum of 2.0. The default value is 1.0, and the weight can be further modulated by a connected schedule curve.

😺NKD Klein Reference Weight Output Parameters:

model

The output model reflects the adjustments made to the reference weights. It incorporates the scaled attention tokens for the specified reference images, ensuring that the desired influence is applied throughout the processing steps. This output is crucial for achieving the intended balance and emphasis of reference images in the final result.

😺NKD Klein Reference Weight Usage Tips:

• To ensure a balanced influence of multiple reference images, use the reference_index to target specific images and adjust their weight accordingly.
• Utilize the schedule parameter to dynamically modulate the reference weight across processing steps, allowing for more nuanced control over the influence of reference images.

😺NKD Klein Reference Weight Common Errors and Solutions:

"NKD Klein Reference Region: empty mask — node is a no-op."

• Explanation: This error occurs when the mask input is empty or missing, leading to no operation being performed by the node.
• Solution: Ensure that a valid mask is provided if regional control is intended. If not, verify that the node is configured correctly for pure strength adjustments without a mask.