😺NKD Klein Reference Region:
The NKDKleinReferenceRegion node is an experimental tool designed to direct a single reference image to a specific zone on a canvas. This is achieved by painting a mask for the desired zone and selecting the reference image to apply. The node confines the influence of the reference image to the masked zone and can optionally strengthen its effect within that area. It operates by adjusting the attention between the canvas and the reference image, making it independent of resolution. This node is particularly useful for artists who want to control the placement and influence of reference images on their artwork. It is important to chain one node per region/reference and place it on the model line between Presampling and your sampler to ensure proper functionality.
😺NKD Klein Reference Region Input Parameters:
model
The model parameter represents the AI model being used. It is essential for processing the reference image and applying the attention biases as specified by the node. This parameter does not have specific minimum or maximum values as it depends on the model architecture being used.
latent
The latent parameter is the latent representation of the image being processed. It is passed through the node to maintain consistency in the image generation process. This parameter is crucial for mapping the region specified by the mask.
mask
The mask parameter is a crucial input that defines the specific zone on the canvas where the reference image's influence will be applied. It is a binary mask where the zone is marked, and its absence or an empty mask results in the node being a no-op, meaning it will not perform any operation.
reference_index
The reference_index parameter specifies which reference image to apply to the masked zone. It is an integer value with a default of 1, indicating the first reference image. This parameter allows you to select the appropriate reference image when multiple references are available.
region_weight
The region_weight parameter determines the strength of the reference image's influence within the masked zone. It is a float value with a default of 1.0, a minimum of 0.0, and a maximum of 2.0. A higher value increases the influence of the reference image within the zone.
outside_suppression
The outside_suppression parameter controls the suppression of the reference image's influence outside the masked zone. It is a float value with a default of 1.0, indicating full suppression. Adjusting this value can allow some influence of the reference image outside the zone.
region_hardness
The region_hardness parameter defines the hardness of the transition between the masked zone and the rest of the canvas. It is a float value with a default of 0.0, indicating a soft transition. Increasing this value results in a harder edge between the zone and the surrounding area.
😺NKD Klein Reference Region Output Parameters:
model
The model output is the modified AI model with the applied attention biases. This output is crucial as it reflects the changes made by the node to direct the reference image's influence to the specified zone on the canvas.
latent
The latent output is the unchanged latent representation of the image. It is passed through the node to ensure continuity in the image generation process and can be used in subsequent nodes for further processing.
😺NKD Klein Reference Region Usage Tips:
- Ensure that the mask accurately represents the zone where you want the reference image to influence the canvas. An incorrect mask can lead to undesired results.
- Experiment with the region_weight and outside_suppression parameters to achieve the desired level of influence and suppression for the reference image.
- Chain multiple NKDKleinReferenceRegion nodes if you need to apply different reference images to different zones on the canvas.
😺NKD Klein Reference Region Common Errors and Solutions:
"NKD Klein Reference Region: empty mask — node is a no-op."
- Explanation: This error occurs when the mask is either missing or does not contain any marked zones, resulting in the node not performing any operation.
- Solution: Ensure that the mask is correctly defined and contains the desired zone for the reference image's influence.
"NKD Klein Reference Region: token grid mismatch (seq=%d, canvas=%d, refs=%d) — skipping this step."
- Explanation: This error indicates a mismatch in the token grid dimensions between the sequence, canvas, and references, preventing the node from applying the attention biases.
- Solution: Verify that the dimensions of the sequence, canvas, and references are compatible and correctly configured. Adjust the model or input parameters as necessary to resolve the mismatch.