How to Remove the Gemini Watermark: Best Free Ways (For Your Own Images)
To remove the Gemini watermark from your own image, download the original file from Gemini and use a tool that subtracts the semi-transparent sparkle with reverse alpha blending, such as our Gemini Watermark Remover or the open-source projects below. That restores the pixels under the mark instead of painting over them. If the image was resized or recompressed, a healing or inpainting tool is the fallback; cropping works when you can spare the corner. Only the visible mark goes away: SynthID stays in the pixels.
Gemini images (including those made with the Nano Banana models) carry two marks. The visible one is a white four-pointed sparkle in the bottom-right corner. The invisible one is SynthID, which Google DeepMind embeds in all media generated by Google’s tools. According to Gemini Help, SynthID usually survives resizing, recolouring and compression, although after many alterations it may no longer be detected.
How we compared
- Method: exact reverse alpha blending (calculates the original pixels) or inpainting (invents plausible pixels from the surroundings).
- Where processing happens: in your browser, on your computer, or on a vendor’s server.
- Effort: automatic detection of the mark or manual work.
- Limits: only what the tool states.
- What remains: every option leaves SynthID in place; we say so for each.
Quick comparison (October 2026)
| Option | Method | Processing | Effort | Notes |
|---|---|---|---|---|
| PromoPilot Gemini Watermark Remover | Reverse alpha blending; fill-in as a fallback | Browser, no upload | Automatic, manual frame if needed | Up to 10 images, 25 MB each; PNG, JPEG, WebP in |
| GargantuaX/gemini-watermark-remover (open source) | Reverse alpha blending | Browser or Node.js, client-side | Automatic | Also a Chrome extension, userscript and CLI; video support |
| allenk/GeminiWatermarkTool (open source) | Reverse alpha blending + optional AI cleanup | Offline desktop app | Automatic, batch | GUI and CLI; MIT licence |
| Cropping | Cut the corner off | Any editor | Seconds | Changes framing and aspect ratio |
| Photopea Spot Healing Brush / GIMP Heal | Healing from surrounding pixels | Browser (Photopea) or desktop (GIMP) | Manual | Good on plain backgrounds, blurs fine detail |
| Cleanup.pictures | AI inpainting | Online service | Brush over the mark | Check free-tier conditions on their site |
The options, by what they are best at
PromoPilot Gemini Watermark Remover — best for untouched Gemini files, in the browser
Our tool looks for the sparkle in the bottom-right corner and subtracts it using the mark’s transparency maps, which come from an open-source project (MIT). Because each pixel under the mark was blended with white in a known proportion, the original value can be calculated back, so texture and fine detail survive. If the file was resized or recompressed, a faint outline may remain; then you can adjust the strength, move the frame by hand or switch to fill-in by the sparkle shape or by the whole frame. Everything runs in your browser and the image is never uploaded. You can add up to 10 images (PNG, JPEG or WebP, 25 MB each) and save as PNG or JPEG. Metadata (EXIF, XMP, C2PA) is copied to the new file by default when the format stays the same; the tool warns that a C2PA signature will no longer match edited pixels.
Open-source projects — best if you want to run it yourself
GargantuaX/gemini-watermark-remover (the former journey-ad repository redirects there) is a JavaScript tool that runs fully client-side with the same reverse alpha blending. Its README lists an online version, a Chrome extension and userscript that work on the Gemini page, a CLI and video support. allenk/GeminiWatermarkTool is an offline desktop program in C++ with a GUI and a command line, batch processing and optional GPU denoising; its author describes it as the original reverse-alpha implementation. Both repositories warn that Gemini’s mark changes over time, so use the latest version.
Cropping — best when the corner does not matter
The sparkle sits in the bottom-right corner, so cropping a strip from the bottom or right edge removes it without any special tool. You lose part of the frame and may change the aspect ratio; for social media formats that is often acceptable. Cropping does not remove SynthID either.
Healing tools in photo editors — best for resized copies and plain backgrounds
Photopea’s Spot Healing Brush replaces the area you paint over with content from its surroundings, and Photopea says it runs fully locally with no uploads. GIMP’s Heal tool does similar manual retouching on Windows, macOS and Linux. Cleanup.pictures is an online AI inpainting service for removing objects, text and watermarks; check its current free-tier terms on the site. These tools do not know what was under the mark, so they invent pixels: fine on sky or a wall, visible on faces, text and grass.
Official route: a plan without the visible mark
In its Nano Banana Pro announcement of 20 November 2025, Google said the visible sparkle stays on images from free and Google AI Pro users and is removed for Google AI Ultra subscribers and in the Google AI Studio developer tool. SynthID is still embedded in those images. Check Google’s current plan terms before you rely on this.
Step by step: removing the sparkle from your own image
- Get the original. Use the download button in Gemini, not a screenshot, and do not send the file through a messenger first: messengers and social networks resize and recompress images.
- Try exact removal first. Drop the file into a reverse alpha blending tool and wait for the status. If the mark is found, the corner is restored by calculation.
- Zoom into the corner. Compare before and after at high magnification. A faint light outline means the strength is too low, a dark one means it is too high.
- Fall back only if needed. On a plain background (sky, wall, gradient) a fill-in or healing tool hides the rest of the outline; on detailed areas keep the exact method.
- Choose the format. PNG keeps the result lossless; JPEG at quality 90–95 is much lighter for the web.
- Decide about metadata and labels. Keep or remove metadata deliberately, and label the image as AI-generated where the platform or the law asks for it.
How to check whether an image has SynthID
- Gemini app. Upload the image to Gemini and ask, for example, "Was this created with Google AI?". According to Gemini Help, Gemini can check images, video and audio, but currently recognises only content created by Google AI tools.
- SynthID Detector. Google DeepMind’s SynthID page says the SynthID Detector portal is now open to everyone and checks images, video and audio from Google and, per Google, partners such as OpenAI, NVIDIA and Kakao.
- Open signals. Our SynthID checker does not detect SynthID; it reads what is visible without Google’s detector: C2PA manifests, IPTC/XMP "digital source type" labels and EXIF fields, and shows the bottom-right corner enlarged. For a broader metadata scan, use the image watermark detector.
Common mistakes
- Working on a screenshot or a messenger copy. Scaling smears the mark; download the original from Gemini.
- Expecting the image to become "not AI". SynthID stays, and so may C2PA metadata unless you remove it, for example with the EXIF remover. Removing metadata is about privacy and file size, not about hiding origin.
- Ignoring labelling rules. Many platforms ask creators to label realistic AI content, and some laws (for example the EU AI Act) include transparency rules for synthetic media. Removing the sparkle does not change that.
What our tool does not do
- It does not remove or hide SynthID, and no tool on our site can.
- It is built for the Gemini sparkle. Marks from other generators or logos need the fill-in mode or a healing tool.
- Exact restoration needs the original Gemini file; on resized or heavily compressed copies the result is approximate.
- It does not process video. Up to 10 images per run, 25 MB each.
Related reading: how to tell if an image is AI-generated, how to remove metadata from photos and free tools for invisible characters in text.