NotebookLM Video Watermark Remover
Remove NotebookLM video watermarks and "Made with Google" ending. Browser-local processing with a re-encoded clean MP4 output.
Video Watermark Remover — Features
Remove bottom-right watermark
Trim "Made with Google" ending
100% browser-local processing
The NotebookLM Video Watermark Remover is designed to remove branding from Google NotebookLM-generated videos. Using FFmpeg WebAssembly, the tool processes videos locally in your browser, applies preset cleanup coordinates for common 720p and 1080p NotebookLM exports, and trims the final 2.5 seconds containing "Made with Google" when ending trim is enabled. The cleaned video is re-encoded as MP4 while the audio track is copied where possible, and your file is never uploaded to a server.
How to Use
- Upload your NotebookLM-exported MP4 video file (up to 500MB supported)
- The tool applies NotebookLM export presets with the delogo filter and trims the ending according to your settings
- Download the cleaned, re-encoded MP4 with the audio track copied where possible
Why Choose This Tool
- 100% Browser-Local Processing — Uses FFmpeg WASM technology, your video never leaves your device
- NotebookLM presets — Targets common 1080p and 720p export coordinates without manual area selection
- Controlled output — Re-encodes video with libx264 for cleanup while copying the audio track where possible
Supported Files
Supports MP4 format, up to 500MB. Best for standard 720p or 1080p NotebookLM exports.
Limitations
Processing requires re-encoding and may take several minutes. Non-standard resolutions may not fully remove watermarks. Large files may fail due to browser memory limits.
The NotebookLM video watermark remover at a glance
Two preset rectangles cover the badge — 770 x 62 px at 1080p, 513 x 41 px at 720p — so anything exported at another resolution will miss. Files up to 500 MB on desktop, 100 MB on mobile.
| Input format | MP4 (H.264), one file at a time |
|---|---|
| Maximum file size | 500 MB on desktop, 100 MB on mobile browsers |
| Removal area, 1080p export | x=1104, y=656, 770 x 62 px — about 2.3% of a 1920x1080 frame |
| Removal area, 720p export | x=736, y=437, 513 x 41 px — the same rectangle, scaled |
| Ending trim | Last 2.5 seconds (the "Made with Google" end card). On by default, switchable off |
| Output encoding | libx264, CRF 18, faststart. Audio stream copied, never re-encoded |
| Files uploaded | None. FFmpeg WebAssembly runs inside your browser tab |
| Account required | None. No sign-up, no email, no export limit |
| Price | Free. No output watermark, no paid tier |
How the NotebookLM video watermark remover works
FFmpeg compiled to WebAssembly applies its delogo filter over a preset rectangle, rebuilding the covered pixels by interpolating inward from the rectangle border on every frame. No AI model, no upload — it runs in the browser tab.
What the delogo filter actually does to the badge
It paints over the badge by interpolating from the pixels immediately around it. The video watermark remover runs FFmpeg compiled to WebAssembly and applies delogo over a fixed rectangle: FFmpeg samples the ring of pixels on the rectangle border, then fills the interior by blending inward from those edges, frame by frame, for the whole clip.
That is the important thing to understand about the result. There is no AI model reconstructing what was behind the logo, and no memory of earlier frames. On a flat or gently graded background the patch is invisible. Where a hard edge, moving footage or a line of text crosses the rectangle, you will see a soft smear instead of the correct content, because the filter has nothing to interpolate from except the border.
How the rectangle gets picked, and why only two sizes exist
The NotebookLM video watermark remover runs a probe pass first (ffmpeg -i input.mp4 -f null), reads the duration and frame height out of FFmpeg's log, then picks one of exactly two presets: the 1080p rectangle when height is 1080 or greater, the 720p rectangle otherwise.
The two presets are the same rectangle in relative terms — both span 57.5% to 97.6% across the frame and 60.7% to 66.5% down it, which is why a Video Overview looks identical at either resolution. The consequence is that only 1920x1080 and 1280x720 land exactly. Nothing detects the badge; the coordinates are declared in advance.
What the four options change in the FFmpeg command
Trim ending subtracts 2.5 seconds from the duration before encoding, which drops the "Made with Google" end card. It is skipped automatically if the clip is shorter than 2.5 seconds.
Remove first-frame top watermark adds a second delogo rectangle across the top of the frame (y=15, 800 x 60 px at 1080p) gated to the first 0.1 seconds only, so it masks the opening title card without touching the rest of the video.
Cinematic mode applies a light colour grade — eq=contrast=1.05:brightness=0.02:saturation=1.1. It does not move or resize the removal rectangle.
Frame rate is a resample: 15 FPS for speed, 30 FPS by default, or original. Everything is then encoded with libx264 at CRF 18 (CRF 24 and the ultrafast preset on phones when the file is over 30 MB), with the audio track copied through untouched.
When people reach for a NotebookLM video watermark remover
Instructors, internal comms teams and video editors, mostly — people whose Video Overview cannot ship with a Google badge sitting across the lower right of every single frame and a 2.5-second "Made with Google" card at the end.
An instructor posting a Video Overview inside Canvas or Moodle
You generate a six-minute overview from a reading pack and embed it in a course module. The badge occupies a 770-pixel-wide strip in the lower right of every single frame, for the entire runtime — anything your slides put in that band is partly covered, and the mark is on screen far longer than the end card. The 2.5-second end card is the other problem: in a module with autoplay, the next item starts on a Google splash screen rather than on your material.
Internal comms clearing a video for the company intranet
Most brand guidelines prohibit third-party logos in official internal material, and an onboarding video carrying a Google mark implies a partnership that does not exist. Comms or legal sends it back. Worth being clear about what this fixes: it stops the video from implying an endorsement. It does not change who wrote the source documents, and it does not make the video less AI-generated.
Cutting an overview into a longer edit in Premiere or Resolve
When a Video Overview becomes one segment inside a longer sequence, the end card is dead footage sitting between your cuts, and the corner badge appears and disappears at every in and out point. Viewers read that flicker as an editing mistake rather than as branding. Trimming and cleaning the source before you drop it on the timeline means one encode instead of an overlay patch you have to rebuild every time you re-cut.
A conference booth loop or a webinar explainer clip
A 30-second overview looping on a trade-show screen or dropped into a webinar as a lower-third explainer puts another vendor's logo on your stand for the entire event. If you are a Google partner that may be fine; if you are not, or if you compete in the same category, it reads badly to anyone walking past.
What the NotebookLM video watermark remover will not do
It removes the visible badge, not SynthID — Google's invisible watermark survives the libx264 re-encode intact, so the output stays identifiable as AI-generated. The rectangle is also chosen from frame height, never detected, so unusual exports miss.
This removes the visible logo. It does not remove SynthID.
SynthID is Google's invisible watermark, embedded in the pixels of every frame and in the audio waveform. Google states it is designed to survive cropping, filters, frame-rate changes and lossy compression, and it is applied across Google's generative AI consumer products — so assume your export carries it. One libx264 pass at CRF 18 will not touch it. Independent testing published in July 2026 needed roughly 300 successive re-compressions plus a 20% crop before detection finally failed, by which point the video was unwatchable. Anyone can upload your output to the Gemini app and ask whether it was made with Google AI; for video, Gemini reports which segments carry the mark. Plan on the output remaining identifiable as AI-generated.
The coordinates are preset. Unusual exports will miss.
The rectangle is chosen from frame height, not from finding the badge. A 2560x1440 export gets the 1080p rectangle applied at 1440p — too far left and too high, leaving the badge fully visible. An 854x480 file gets the 720p rectangle for the same reason. Two current export types are outside the presets entirely: Short Video Overviews, which are roughly 60 seconds in 9:16 vertical, and Cinematic Video Overviews, which are built on Veo and available only on Google AI Ultra. Both place branding somewhere the 16:9 rectangle does not reach. If Google changes the export layout, the preset goes stale until it is updated — changes are logged on the changelog.
The output is a re-encode, not a copy of your file
libx264 at CRF 18 is close to visually transparent, but it is still a second generation: fine detail and grain are re-quantised, and the file size can land above or below the source depending on how the original was encoded. Frames are also resampled if you leave the frame rate on 30 FPS and your source is 24 or 25. The audio track is stream-copied, so sound is bit-identical to the source. If you want that guarantee for the whole file, the audio trimmer never re-encodes at all — but it only cuts, it cannot remove anything visual.
One file at a time, and it is slower than you expect
There is no batch queue and no folder upload for video. More to the point, x264 running in WebAssembly has no GPU path and is several times slower than native FFmpeg — a ten-minute 1080p video routinely takes longer to process than it takes to watch. Selecting 15 FPS roughly halves that. This is the honest cost of not uploading your file to a server.
Troubleshooting the video watermark remover
The most common report is a badge still visible after processing, and the cause is almost always resolution: anything that is not 1920x1080 or 1280x720 gets one of the two preset rectangles applied in the wrong place.
The logo is still visible in the corner after processing.
Your export is not 1920x1080 or 1280x720. The rectangle is preset, so a 1440p, 480p or 9:16 vertical file gets a rectangle in the wrong place. Check the resolution in your file properties. Re-export at 1080p from Gemini Notebook, or scale the file to 1920x1080 in any editor first, then run it through again.
There is a soft blurry rectangle where the logo used to be.
That is delogo working as designed — it interpolates inward from the rectangle border, so a busy or moving background gives it nothing coherent to interpolate. There is no setting that fixes this. A frame-aware editor (After Effects content-aware fill, or a desktop object-remover) does better on moving backgrounds, at the cost of installing software or uploading your file.
The browser tab froze or crashed on a 400 MB file.
FFmpeg WebAssembly holds the input, the output and its working buffers in the tab's memory at once, so peak usage runs to several times the file size, against a hard 4 GB ceiling in 32-bit WebAssembly. Close other tabs, choose 15 FPS, or split the video before processing. On phones, anything over 30 MB is automatically dropped to CRF 24 and the ultrafast preset to survive the smaller memory budget.
The motion looks slightly stuttery compared with the original.
The default output is 30 FPS. A 24 or 25 FPS source has frames duplicated unevenly to reach 30, which shows up as judder on pans. Select "Quality (original FPS, slower)" in the frame rate dropdown and re-run.
Cinematic mode did not move the removal area, despite the label.
Cinematic mode applies a colour grade only — eq=contrast=1.05:brightness=0.02:saturation=1.1. The delogo rectangle is unchanged. If the badge is in the wrong place for your file, the fix is matching the resolution, not this toggle.
The last words of the audio are cut off.
Ending trim removes a flat 2.5 seconds. If the narration runs right up to the end card, switch the trim off and cut the tail yourself in an editor, or accept the 2.5 seconds of branding. The trim is a fixed duration, not a detection of where the end card starts.
Privacy, copyright and acceptable use
Your video is never uploaded — FFmpeg WebAssembly runs in the tab and the site has no upload endpoint to send it to. Removing the badge changes nothing about who owns the source material or your duty to disclose AI-generated content.
Process files you generated or are authorised to edit
Use this on your own exports, or on files you have written permission to process. Google's terms govern what you may do with Gemini Notebook output, and the sanctioned route to an unbranded export is a Google AI Ultra subscription, which removes the branding at generation time rather than afterwards. This is a post-processing utility. It does not grant you a licence to anything, and it makes no judgement about whether your particular use is permitted.
Nothing is uploaded — you can check that yourself
Press F12, open the Network panel, drop in a video and run it. You will see the page, the FFmpeg WebAssembly core and the fonts load, and no request carrying your file. The site is a static build with no upload endpoint and no server-side processing to send anything to. That is also why processing speed depends on your machine rather than on our queue. Full detail is on the privacy page.
Removing branding does not change ownership or disclosure duties
Taking the badge off does not make the video yours if the source material was not, and it does not exempt you from AI-disclosure rules. YouTube requires creators to disclose realistic altered or synthetic content; academic honesty policies and a growing number of platform terms expect the same. SynthID stays in the file regardless of what this video watermark remover does to the corner. If you are publishing rather than editing for internal use, disclose that the video is AI-generated and verify the claims in it against your original sources. Our boundaries are set out on the acceptable use page.