How to get Gemini images without a watermark
There are two legitimate paths here, and they are not the same thing. One changes which Gemini tier generates the image so no visible mark is added in the first place. The other removes a supported visible mark that is already on an existing Free- or Pro-tier image. Neither path touches SynthID, and this guide does not claim otherwise.
"Unwatermarked" in this guide means specifically "no visible corner mark." It does not mean "no AI provenance signal at all." Keep that distinction in mind through every step below, because conflating the two is the single most common mistake in comparisons of Gemini's watermark behavior.
Get an unwatermarked image in three steps
1. Check which tier produced the image
Free and Pro tier Gemini images ship with a visible corner mark plus SynthID. Google AI Ultra and AI Studio and API images ship with SynthID only, so there is no visible mark to work around in the first place. See Gemini watermark by tier for the full breakdown.
If you generated the image yourself, you already know which surface you used. If someone sent it to you or you found it elsewhere, look at the image itself: a visible mark near a corner means Free or Pro tier. No visible mark is consistent with Ultra or API output, but it can also mean the mark was cropped out, covered, or the file was resized before it reached you. Absence of a visible mark on a file you didn't generate yourself is a clue, not proof.
2. Use Ultra or the API for no visible mark upfront
If you can choose the generation surface, Google AI Ultra and the AI Studio and API endpoints do not add a visible corner mark, according to Google's own tier documentation. This avoids the removal question entirely for new generations, at the cost of whatever tier requirements, developer setup, or usage terms apply to that surface.
This route only helps for images you have not generated yet. It changes nothing about a file that already exists with a visible mark on it, and it does not affect SynthID either way; SynthID applies regardless of which of these two paths you take. Because Google AI Ultra's price is disputed across public sources, decide on this route based on what the tier actually does rather than a specific number — see the tier guide for why this guide will not quote a figure.
3. Use the browser eraser for a supported visible mark
If the image already has a Free- or Pro-tier visible mark, open the local browser eraser, upload the original untouched file, and let the detector validate a supported candidate before downloading the result. If the detector cannot confirm a supported mark, the safer outcome is an unchanged file rather than a forced edit.
Bring the original download if you can get it. A JPG, PNG, or WebP straight from Gemini keeps the mark's expected size, position, and edge profile intact; a screenshot, a re-saved copy, or a heavily cropped version can shift or distort those enough that the detector correctly declines to guess. The full step-by-step walkthrough covers file requirements in more detail, and the troubleshooting checklist covers what to check when a file is skipped.
Which path fits your situation?
Pick the route that matches what you're starting from
Answer the first question that applies to you.
Do you already have a Free- or Pro-tier image with a visible mark? — Yes
Use the browser eraser on the original file. If it gets skipped, work through the troubleshooting checklist before assuming the tool failed.
Are you about to generate a new image and can pick the surface? — Yes
Generate on Google AI Ultra or through AI Studio or the API instead of Free or Pro. No visible mark will be added, per Google's tier documentation, though SynthID still applies.
Do you need the file to carry no AI provenance signal at all? — Yes
Neither path here does that. SynthID persists across every tier by design, and this project does not detect or remove it. Disclosure obligations follow the content, not the visible mark.
Is the file a screenshot, a resize, or already cropped? — Yes
Both routes get harder from here. Retrieve the original Gemini download if it still exists; a supported removal candidate depends on the mark's expected position and edge profile matching the source file, not a copy of it.
Cost and control point in opposite directions between the two paths. Switching to Ultra or the API buys you a clean file the moment it's generated, but it depends on a tier or developer access you may not already have, and its price is exactly the figure this guide declines to state without independent verification. Running the browser eraser costs nothing and needs no account, but it only ever targets a visible mark it can confidently validate on a file you already have; a low-confidence or unsupported image is left unchanged rather than forced through an edit. If your image doesn't need a visible mark removed at all — because it already came from Ultra or the API — the honest answer is that you don't need either path, and google-ai-studio-watermark explains that case directly.
Frequently asked questions
Does this tool remove SynthID so the image has no watermark at all?
No. This project only targets the visible corner mark on supported Gemini and Veo output. It does not detect or remove SynthID, and no visible corner mark does not mean no invisible signal.
Is using Google AI Ultra or the API the only way to avoid a visible mark?
It is the tier-level way, per Google's own tier documentation. The alternative is generating on Free or Pro and then using a focused removal tool on the visible mark specifically, which is an evidence-gated process, not a tier setting.
Will resizing or cropping the image remove the watermark?
Not reliably, and it can make a supported removal tool less confident too, since cropping and resizing change the mark's expected position and edge profile.
Primary sources
- Nano Banana Pro and visible watermark policy — Google
- Gemini image generation guide — Google AI for Developers
- Gemini Watermark Remover source and usage — Gemini Watermark Remover on GitHub