kevintsai1202/geminiwatermarkremoveskill

Overview

This skill removes the visible watermark that Google Gemini (Nano Banana) adds to generated images. It runs entirely locally and restores covered pixels using a reverse alpha-blend algorithm to approximate the original content. The tool is provided as a small JavaScript CLI with automatic and manual mask modes.

How this skill works

The algorithm applies reverse alpha blending: original_pixel = (current_pixel - α × watermark_color) / (1 - α). It detects the typical watermark mask size based on image resolution and applies a matching mask (48×48 with 32px margin for ≤1024px, 96×96 with 64px margin for >1024px). Users can run the CLI via npx or directly with Node.js, and tweak mode and alpha gain for better results.

When to use it

• You have a Gemini-generated image with a visible Nano Banana watermark.
• You prefer to remove the watermark locally without uploading images to a server.
• You need batch processing for multiple images in a folder.
• You want adjustable strength to recover underlying pixels when default results look faded.

Best practices

• Start with the default auto mode and inspect results before changing parameters.
• Use the gain parameter (1.0–3.0) to increase or reduce the reverse alpha effect; higher gain can overcorrect colors.
• Test on a copy of the original image to preserve the source file.
• Batch-process only after verifying the best mode/gain on a representative sample image.

Example use cases

• Quickly clean a single Gemini-generated PNG using npx without installing dependencies.
• Process a folder of generated images in a batch script to produce watermark-free versions.
• Fine-tune output for high-resolution images by switching to the large mask mode.
• Recover partially covered details in promotional assets where the watermark overlaps important areas.