ComfyUI Node: 🐇 Watermark

🐇 Watermark:

The BatchWatermarkSingle node is designed to apply a single-position watermark to batches of images efficiently. This node is particularly useful for AI artists who need to add watermarks to multiple images simultaneously, ensuring consistency and saving time. It leverages GPU processing to handle images in batches, supporting both single images and larger collections. The watermark is applied with customizable parameters such as scale, rotation, and position, allowing for precise control over its appearance. The node also preserves the original transparency of the watermark image, ensuring that the final output maintains the desired visual quality. By using advanced techniques like TorchLanc for resizing and optional torch.compile for optimization, this node provides a robust solution for watermarking tasks in AI art projects.

🐇 Watermark Input Parameters:

image

This parameter accepts the images to be watermarked, which can be in the form of a single image with dimensions (H,W,C) or a batch of images with dimensions (B,H,W,C). The values should be in the range [0,1], and the processing is done on the GPU to ensure efficiency.

watermark

This parameter allows you to select or upload the watermark image, with PNG being the recommended format due to its support for transparency. The watermark's transparency is preserved, ensuring that it blends seamlessly with the images.

position

This parameter determines where the watermark will be placed on the images. Options include "bottom-right", "bottom-left", "top-right", "top-left", and "center". When "center" is selected, padding is ignored. The default position is "bottom-right", and rotation is applied with clipping, meaning the canvas size is not expanded.

scale

This integer parameter controls the scaling of the watermark based on the width of the image. The target watermark width is calculated as the image width multiplied by (scale/100). The aspect ratio of the watermark is preserved. The default value is 70, with a minimum of 1 and a maximum of 100.

max_batch_size

This integer parameter specifies the maximum number of images to process in a single batch, which helps manage VRAM usage. A value of 0 means the entire batch is processed at once. The default is 0, with a range from 0 to 4096.

sinc_window

This integer parameter sets the Lanczos window size used when resizing the watermark. A higher value results in a sharper image but may introduce more ringing artifacts. The default is 3, with a range from 1 to 8.

precision

This parameter defines the resampling compute data type, with options including "fp32", "fp16", and "bf16". The default is "fp32", which offers the safest quality, while "fp16" and "bf16" can provide faster processing on many GPUs.

🐇 Watermark Output Parameters:

output_image

The output of this node is a batch of images with the watermark applied. The images retain their original dimensions and are returned in the same format as the input, either as a single image (H,W,C) or a batch (B,H,W,C). The watermark is applied according to the specified parameters, ensuring a consistent and professional appearance across all images.

🐇 Watermark Usage Tips:

• To maintain the quality of the watermark, use a high-resolution PNG image with transparency.
• Adjust the scale parameter to ensure the watermark is visible but not overpowering, especially for images of varying sizes.
• Use the max_batch_size parameter to optimize performance based on your system's VRAM capacity, preventing memory overflow issues.

🐇 Watermark Common Errors and Solutions:

"image must be a torch.Tensor with shape (H,W,C) or (B,H,W,C) in [0,1]."

• Explanation: This error occurs when the input image is not in the expected format or range.
• Solution: Ensure that the image is a torch.Tensor with the correct dimensions and that its values are normalized between 0 and 1.

"Select a watermark image from the list (or upload one)."

• Explanation: This error indicates that no valid watermark image has been selected or uploaded.
• Solution: Choose a watermark image from the available list or upload a new one, ensuring it is in a supported format like PNG.

"Unsupported channel count C={C}. Expected 1, 3 or 4."

• Explanation: The input image has an unsupported number of channels.
• Solution: Convert the image to have 1 (grayscale), 3 (RGB), or 4 (RGBA) channels before processing.