NotebookLM Infographic Watermark Remover

The NotebookLM Infographic Watermark Remover is specifically designed to clean branding watermarks from the bottom-right corner of NotebookLM-generated infographics. The tool uses advanced connected component analysis algorithms to scan the bottom area of images, precisely identifying watermark position and boundaries, then uses vertical gradient interpolation technology to sample from surrounding background and intelligently fill the watermark area. The entire process happens in the browser Canvas API without uploading images to servers, maintaining original resolution and color quality.

How to Use

1. Upload your NotebookLM-generated infographic (supports PNG, JPG, WebP formats)
2. The tool automatically scans the bottom-right area, detecting and locating watermark boundaries
3. Download the watermark-removed image with original resolution and quality maintained

Why Choose This Tool

• Smart Detection — Connected component algorithm automatically identifies watermarks, no manual area selection needed
• Natural Fill — Gradient interpolation technology ensures filled areas blend naturally with background
• Quality Maintained — Native Canvas processing, no image compression, preserves original colors and details

Supported Files

Supports PNG, JPG, WebP formats. Best for NotebookLM-generated infographics.

Limitations

Only processes bottom-right watermarks. Non-standard watermark positions or complex backgrounds may require manual adjustment.

Step 1 — locating the mark by shape, not by matching a picture of it

The NotebookLM infographic watermark remover reads a box in the bottom-right corner sized at whichever is larger, 350 px or 22% of the image width, by 80 px or 8% of the height, capped at half the width and a quarter of the height. It takes the median grey of that box as the background, then builds a mask of pixels that sit more than 80 grey levels away from it — darker or lighter, which is why the same code handles dark text on a pale card and white text on a dark one. If that pass finds nothing usable it retries at 60 levels, then at 45. Each connected blob in the mask is measured and thrown out if it is smaller than 20 px of area (35 px above 1200 px, 60 px above 2000 px), shorter than 8 px (10 px, then 14 px at those same sizes), taller than 80% of the scan box, more than 12 times longer than it is tall, or centred in the top quarter of the box. What survives is the wordmark and its icon.

Step 2 — merging the icon with the text and padding the box

The largest surviving blob is assumed to be the wordmark. Anything whose vertical centre sits within 35% of the scan box height of that blob is merged into the same rectangle, which is how the small logo glyph and the words next to it end up in one region instead of two. The merged rectangle is then grown by 15 px horizontally and 8 px vertically — 25 px and 14 px on images over 2000 px — to catch antialiased edges and the faint halo that survives a lossy re-encode.

Step 3 — filling with a per-column vertical gradient

For each column of the cleared rectangle the infographic watermark remover averages an 8 px strip directly above it and an 8 px strip directly below (14 px on large images), then interpolates linearly between the two down that column. Because it works column by column rather than filling one flat colour, it follows horizontal colour shifts and the soft vertical gradients NotebookLM uses on its infographic cards. This is arithmetic, not a generative repaint: it is fast, deterministic, and it produces the same output every run — but it only reconstructs what a straight-line blend can describe. If the pixels underneath were a photograph or a dense chart, that limit is real and is spelled out below. If nothing at all is detected after all three threshold passes, the NotebookLM infographic watermark remover falls back to clearing a box starting at 80% of the width and 94% of the height, covering 20% by 6% of the image, which is where the mark sits on standard exports.

A lecturer printing an infographic as a course handout

NotebookLM will happily turn a reading list into a one-page visual summary at 2752 px on the long edge, which is enough for a crisp A4 print. At that size the corner brand mark prints too, at roughly 20 mm wide, and on a handout distributed to a class it reads as an advertisement rather than as attribution. Clearing it takes about a second in the browser and the print resolution is unchanged, because nothing is re-encoded except the rectangle that was cleared.

A newsletter writer putting the infographic above the fold

Email clients and Substack render an infographic at 600-700 px wide, so the brand mark shrinks to a smudge that is legible enough to notice and too small to read. Removing it before upload also avoids the second problem: if you crop the corner off instead, you lose the bottom of the last data row. The gradient fill keeps the card edge intact so the image still ends where the design intended.

A consultant merging the infographic into a client deck

An infographic rarely travels alone. If the same notebook also produced a slide deck, the mark is on every page of that too — those go through the PDF slides remover or the PPTX remover, which run this same detection pass over each rendered page. Doing all three in one sitting means the deck, the handout and the exported video carry one consistent look rather than three different corners.

Anyone on the free or Plus tier deciding whether to upgrade

Google's only supported way to get an unmarked export is the AI Ultra plan; the free tier and the $19.99/month Pro tier both stamp every infographic, slide deck and video. The in-app slide editor added in February 2026 will not delete the mark either, because it is composited into the background layer rather than sitting on the canvas as an object you can select. If the mark is the only reason you were considering the upgrade, it is worth running one export through here first — the honest comparison is in Ultra vs free.

SynthID stays in the image

This infographic watermark remover removes the visible corner brand mark and nothing else. Google also embeds SynthID, an invisible watermark carried in the pixel signal itself rather than in metadata, and it is engineered to survive cropping, re-compression, colour changes and screenshots. A cleaned infographic is still identifiable as Google-generated by a SynthID detector. If your goal is to pass an image off as human-made, this NotebookLM infographic watermark remover does not do that and no honest tool claims to.

Photographic and high-detail backgrounds leave a visible patch

The fill is a linear blend between the strip above the mark and the strip below it. Under a flat card, a soft gradient, or a solid band, that blend is indistinguishable from the original. Under a photograph, a texture, a dense bar chart or text that runs under the mark, it is not — you get a smooth rectangle where detail used to be, obvious at 100% zoom. There is no setting that fixes this, because the information under the mark is genuinely gone from the exported PNG. If your infographic template puts the mark over imagery, regenerating with a plain-footer template is the better move.

It clears exactly one rectangle, in one corner

Detection only ever looks at the bottom-right region and only ever returns a single merged rectangle. A second mark elsewhere on the page, a logo you added yourself, a footer URL, or a mark that a non-standard template placed centre-bottom will all be left alone. A translucent four-point sparkle is a different watermark with a different removal path — that one is solved by arithmetic rather than by filling, in the Gemini image remover. It also cannot tell a brand mark from a legitimate caption that happens to sit in that corner — see the troubleshooting note about erased chart elements.

Phones downscale large images

On a device whose browser reports iPhone, iPad or Android, any image over 2048 px on either edge is scaled down to fit inside 2048 px before processing. This is deliberate: a 2752 x 1536 canvas plus its pixel buffer is enough to get a mobile tab killed mid-run. If you need the full 4K file back, run it in a desktop browser, where no downscale is applied.

A smooth grey or beige rectangle is left where the watermark was

The pixels under the mark were not a flat or gradient background. The fill averaged an 8 px strip above and below and blended between them, so a photo, texture or chart underneath becomes a smear. Nothing in the infographic watermark remover tunes this away. Either regenerate the infographic with a footer style that keeps the corner clean, or patch the rectangle by hand in an image editor — the cleared area is a known size, 216 x 44 px on standard 1536 x 2752 exports.

Part of my chart or a caption in the corner got erased

Two causes. Either a real page element in the bottom-right scored as the largest dark blob and was treated as the wordmark, or all three threshold passes failed and the fallback fired, which blindly clears the region from 80% width and 94% height covering 20% by 6% of the image. If your design puts content in that corner, process the file before you add it, or accept the fallback box and repair it afterwards.

The watermark is still there and nothing changed

The mark has to be inside the bottom-right scan box and has to differ from its background by at least 45 grey levels — the loosest of the three passes. A very low-opacity mark on a mid-grey card can fall under that, and a mark placed centre-bottom by a non-standard template is outside the search area entirely. Check where yours actually sits using this breakdown of watermark positions before assuming the file is broken.

The result came back much smaller than the file I put in

You are on a phone or tablet. The mobile path caps the longer edge at 2048 px to avoid an out-of-memory crash, so a 2752 px export returns at 2048 px. Nothing else about the image changes. Run the same file in a desktop browser to keep the original resolution.

I uploaded a WebP and got a much larger PNG back

Expected. Only JPG input is re-encoded as JPG; everything else, WebP included, is written out as PNG. PNG is lossless and has no quality slider, so a photographic infographic can easily triple in size. Convert the PNG back to WebP afterwards if file size matters — do it as the last step, never before removal, since a lossy re-encode blurs the edges the detector keys on.

My newer exports say "Created with Gemini Notebook" — does that still work?

Yes. Google renamed NotebookLM to Gemini Notebook on 16 July 2026 and the wording on the mark changed with it. Detection never reads the letters; it groups dark or light blobs by position and merges anything on the same row, so a longer wordmark simply produces a wider rectangle. Both the old and new mark are handled by the same code path.

Use it on exports you made yourself

The intended use is straightforward: you generated the infographic from your own sources in your own notebook and you want your own document without a vendor badge on it. Removing a brand mark from someone else's work, from a stock image, or from anything you do not have the right to alter is a different act and this NotebookLM infographic watermark remover does not make it acceptable. A longer treatment of where the line sits is in this piece on the legal question.

Nothing is uploaded, and you can verify that yourself

The image is read with the FileReader and Canvas APIs inside your tab. There is no upload endpoint, no queue, no temporary storage and no retention policy to trust, because there is no server copy to retain. You do not have to take that on faith: open DevTools, switch to the Network panel, drop a file in, and watch that no request carries the image. That is a stronger guarantee than a privacy page, and it is the reason this works offline once the page has loaded.

Removing a badge does not remove a disclosure obligation

A visible mark and a disclosure requirement are two separate things. Several large platforms require you to label AI-generated or materially altered media at upload, and the EU AI Act places transparency duties on whoever publishes such content, not on whoever generated it. Clearing a corner of an image changes neither. Treat this as a formatting step for material you already have the right to publish, and disclose AI involvement wherever your platform, client or jurisdiction expects it.