🐇 Watermark

Adding a watermark to a single image is a five-second job in any editor. Adding it to every frame of a video is the job that makes you swear at the editor instead. Watermark (class BatchWatermarkSingle) is WhiteRabbit's answer: a GPU-accelerated watermark overlay that takes a batch of images, composites a watermark onto each one, and does the watermark's scaling with the same gamma-correct TorchLanc resampler the pack uses everywhere else. It works on a single image, but it's built for the hundreds-of-frames case.

The honest pitch is in the author's own words: "very quick, especially when compared to doing the same task in pro editing tools." If you've ever exported a video from a local render and then sat in a video editor waiting for a batch watermark job, you know the pain this removes. It's the watermark, applied at render time, so you never touch the footage twice.

How it works

You pick a watermark file (PNG recommended - its transparency is preserved), and the node resizes it to a width relative to the target image, then alpha-composites it at the chosen position with the chosen transparency. The sinc_window and precision controls govern the Lanczos resize quality, and everything runs on the GPU.

The inputs that matter

• image - the frames to watermark (single or batch).
• watermark - select or upload the watermark image. Starts with an example.png so you can test the node before dropping in your own mark.
• position - bottom-right (default), bottom-left, top-right, top-left, or center.
• scale - target watermark width = image width × (scale/100), aspect preserved. 70 means the mark is 70% as wide as the frame - generous; most people land lower.
• transparency - alpha multiplier, 0–100. 100 = the file's own alpha, unchanged.
• rotation - rotate the mark (bicubic), with the canvas expanding so nothing clips (PIL-style).
• padding_x / padding_y - pixel padding from the chosen edge. Ignored when position is center.
• optical_padding + optical_strength - this is the nice one. Corner alignment with a wide or rotated mark looks off-center to the eye because the visual center isn't the geometric center. Optical alignment nudges placement toward the mark's visual center so equal padding looks equal. Off by default; try it.
• max_batch_size - chunk the batch for VRAM control; 0 = whole batch at once.

Installing it

WhiteRabbit pack install, done once:

cd ComfyUI/custom_nodes
git clone https://github.com/Artificial-Sweetener/comfyui-WhiteRabbit
cd comfyui-WhiteRabbit
python -m pip install -r requirements.txt

or ComfyUI Manager → "WhiteRabbit". The only pip dependency is torchlanc (the GPU resampler this node uses); the pack targets ComfyUI's v3 node API.

Where people get burned

• Using a JPG watermark. JPGs have no alpha channel, so the transparency is garbage and you get a white box around your mark. The tooltip is not kidding: PNG.
• Padding with optical alignment off. A mark with a wide empty margin (say, a text logo) will look inset oddly at corners because the visible glyph isn't centered in its box. Turn on optical_padding and set optical_strength to taste.
• Huge scale on high-res frames. The mark is sized relative to image width, so a setting that looks subtle on a 512-wide frame can look like a billboard on a 4K one.

Watermark every frame of a 10-second clip in the time it takes the batch to render, and you'll wonder why you ever exported first and watermarked second.