How to Remove the Gemini Sparkle Watermark from AI Images (Lossless, 2026)

The Gemini Sparkle Watermark

Google's Gemini image generator (and the image features inside NotebookLM) stamps a small four-pointed sparkle "✦" watermark in the bottom-right corner of every image it produces. It's subtle, but it marks the image as AI-generated and looks out of place in presentations, thumbnails, and design work. The same sparkle appears on output from Google's newer image model — see our dedicated guide to removing the Nano Banana watermark from Gemini images.

There are actually two different watermarks on Gemini images, and it's important to understand the difference:

• The visible sparkle ✦ — the small glyph in the corner. This is what you can see and what this guide removes.
• SynthID — an invisible watermark Google embeds in the pixel data itself. This is not something a browser tool removes, and this guide does not touch it.

The Gemini watermark remover targets the visible sparkle. For standard image sizes, it removes it losslessly — meaning the result is mathematically identical to the original un-watermarked image. Here's how that works.

Why It's Lossless: Alpha Reversal

This is the part that makes Gemini watermark removal special. Most watermark removers guess what was behind the watermark and fill it in (which leaves artifacts). The Gemini sparkle is different because of how Google applies it.

The sparkle is composited onto the image using alpha blending — the watermark is a semi-transparent white overlay. The math Google uses to blend it is:

watermarked = original × (1 - α) + 255 × α

Where α (alpha) is the transparency of each watermark pixel. Because this is a known, reversible equation, we can solve for the original pixel value:

original = (watermarked - α × 255) / (1 - α)

If we know the exact alpha value of every pixel in the watermark, we can perfectly reconstruct the original image underneath. No guessing, no artifacts — the recovered pixels are mathematically exact.

The Alpha Templates

To reverse the blend, the tool needs the precise alpha map of the sparkle. Gemini uses two sparkle sizes depending on the image resolution:

Image short side	Sparkle size	Margin (right, bottom)
≤ 1024px	48 × 48 px	32px, 32px
> 1024px	96 × 96 px	64px, 64px

The tool ships with pre-measured alpha templates for both sizes. It detects your image's resolution, picks the right template, locates the sparkle position from the known margins, and reverses the alpha blend pixel by pixel. The result is a clean image with zero quality loss.

What About Non-Standard Sizes?

If your image has been resized, cropped, or re-compressed, the sparkle may not sit at the expected coordinates or size — the alpha-reversal math no longer lines up. For these cases the tool has a fallback:

1. Sparkle detection — it scans the bottom-right corner (a 150px region) for bright pixels and uses connected-component analysis to find the sparkle's center, filtering by area (50–800 px) and shape (roughly square aspect ratio)
2. Inpainting — once located, it fills the sparkle area by sampling the surrounding darker pixels and blending them over the bright watermark glyph

The inpainting fallback isn't mathematically lossless like alpha reversal, but it produces a clean, artifact-free result for resized or non-standard images. The tool tries alpha reversal first and only falls back when needed.

Step-by-Step: Remove the Gemini Watermark

1. Go to notebooklmremover.org/gemini-image
2. Drop your Gemini-generated image (PNG, JPG, or WebP)
3. The tool detects the resolution, applies the matching alpha template, and reverses the sparkle
4. Preview the before/after side by side
5. Download the clean image — choose to keep the original format or convert to PNG/JPG/WebP

Everything runs in your browser using the Canvas API with the Gemini image watermark remover. Your image is never uploaded to a server.

Which Sizes Are Lossless?

Alpha reversal works perfectly on Gemini's standard output resolutions, including common ones like 1024×1024, 1536×1024, 2816×1536, and other native generation sizes. As long as the image hasn't been resized after generation, you'll get a mathematically perfect result.

If you've already edited or resized the image, use the original Gemini export for the best (lossless) result, then resize afterward if needed.

A Note on SynthID

It's worth repeating: this tool removes the visible sparkle, not the invisible SynthID watermark Google embeds in the pixel data. SynthID is designed to survive cropping, compression, and editing, and detecting or removing it is a different (and much harder) problem that browser tools don't address. If your goal is simply to clean up the visible mark for a presentation or design, the sparkle removal is exactly what you need.

Other NotebookLM & Gemini Content

If you're working with other AI export formats, NotebookLM Remover has dedicated tools:

• Infographics — removes the "NotebookLM" text watermark via gradient fill
• Video exports — removes the logo overlay and end card
• PPTX presentations — cleans embedded slide watermarks
• PDF slides — removes watermarks and rebuilds a clean PDF

Frequently Asked Questions

Is the removal really lossless?

For standard Gemini output sizes, yes — mathematically lossless. The tool reverses the exact alpha-blending equation Google uses to apply the sparkle, recovering the original pixels precisely. For resized or non-standard images, it uses an inpainting fallback that is clean but not mathematically exact.

Does it work on images from Gemini, NotebookLM, and Google AI Studio?

Yes, as long as the image carries the standard Gemini sparkle watermark in the bottom-right corner. The same alpha-blend approach applies across Google's image generation products.

Will it remove the watermark from a screenshot of a Gemini image?

A screenshot changes the resolution and pixel alignment, so the lossless alpha reversal won't line up. The tool will fall back to inpainting, which still produces a clean result but isn't pixel-perfect. For best results, use the original downloaded image.

Is my image uploaded anywhere?

No. All processing happens locally in your browser using the Canvas API. There's no server upload, no account, and no data collection.