If you have asked "why does Gemini have a watermark", you are not alone. The small sparkle on Gemini images and the invisible signal behind it both exist for the same reason: so people can tell when content was made with AI. Here is the full picture, including text and video.
Short answer
| Content | Visible mark | Invisible mark |
|---|---|---|
| Images | Yes, a sparkle near the corner on many images | Yes, SynthID |
| Videos (Veo) | Yes, "Veo" text on most videos | Yes, SynthID |
| Text | No | Yes, SynthID for text |
| Audio | No | SynthID in supported features |
Why Gemini adds watermarks
- Transparency: a visible mark tells anyone at a glance that an image is AI-generated.
- Misinformation: invisible marks help platforms and fact-checkers spot AI images that are shared as real.
- Regulation: rules such as the EU AI Act's Article 50 require AI providers to mark generated content in a machine-readable way.
- Trust: clear labels make it easier for businesses and users to adopt AI tools responsibly.
Google explains its approach to invisible watermarking on the SynthID page.
The Gemini image watermark
The Gemini image watermark is a white, semi-transparent sparkle, usually near the bottom-right corner. Gemini has used an older, brighter version close to the corner and a newer, fainter one placed further in.
| Visible sparkle | SynthID | |
|---|---|---|
| Can you see it? | Yes | No |
| Where it lives | One corner | Spread through the whole image |
| Survives cropping? | No | Designed to |
| Removable from your own images? | Yes, with a Gemini watermark remover | No |
Some paid plans and developer tools may not add the visible mark, but SynthID is still applied. People sometimes type "gemini water" when they mean this mark; it is the same thing.
Does Gemini leave watermarks in text?
Does Gemini leave watermarks in text? Yes, invisibly. Google has used SynthID for text in Gemini responses. Unlike a visible logo, it is a statistical pattern in the words the model chooses, which a matching detector can recognise. You cannot see it, and it does not change the meaning of the text. Wikipedia's text watermarking page explains the idea in simple terms.
| Assistant | Text watermark |
|---|---|
| Gemini | SynthID for text |
| ChatGPT | Not deployed; see our ChatGPT watermark guide |
| Claude | Newer models add an invisible text watermark, announced in 2026 |
How to remove SynthID?
People search how to remove SynthID, but it is built specifically to stay in content through everyday edits like cropping, compression and filters. It is the part of the watermark system meant to keep working when the visible label is gone. We do not provide ways to strip or defeat it, and no Imaginoa tool attempts to. If your aim is simply a cleaner-looking image or video, removing the visible mark from your own content is enough, and SynthID does not affect how it looks.
Can you remove a watermark?
Can you remove a watermark in general? It depends on whose content it is and what the mark is for.
| Situation | OK to remove? |
|---|---|
| Visible mark on AI images or videos you created | Generally yes, following the AI tool's terms |
| Stock photos (Getty, Shutterstock and others) | No; license the image instead. See our stock photo guide |
| Someone else's photo or video | No, without permission |
| Trial watermarks from paid editing software | No; they mark unlicensed exports |
| Invisible provenance like SynthID or C2PA | Not something to strip |
What this means for you
- Removing the sparkle from your own Gemini images is fine for cleaner thumbnails, slides and posts.
- The image stays identifiable as AI through SynthID, so be open about AI use where it is expected.
- Tools: image remover, video remover, and the Veo3 watermark remover guide.
How Gemini's watermarks have changed
| Change | What it meant |
|---|---|
| SynthID introduced for images | An invisible mark that survives common edits |
| Visible sparkle added to Gemini images | A quick, human-readable AI label |
| SynthID extended to text, audio and video | Provenance across more types of content |
| Newer, fainter sparkle placed further from the corner | Less intrusive, but harder for older removal tools to find |
So why does Gemini have a watermark in two forms? Because each covers a gap in the other: people can see the sparkle, while machines can read SynthID even after the sparkle is gone.
How Gemini compares with other AI tools
| Tool | Visible image mark | Invisible signal |
|---|---|---|
| Gemini | Sparkle on many images | SynthID |
| ChatGPT | Not on most images | C2PA metadata; newer images reported to carry a pixel mark |
| Sora | Moving Sora logo on downloads | C2PA metadata |
| Veo | "Veo" text on most videos | SynthID |
More detail: ChatGPT, Sora and Veo 3.
Cleaning a clip instead of a picture? See the Gemini video watermark remover guide.
FAQ
Why does Gemini have a watermark?
To show that images, videos and text were made with AI, to help spot misinformation, and to meet transparency rules.
Does Gemini leave watermarks in text?
Yes, invisibly. Google uses SynthID for text in Gemini, a pattern in word choices that detectors can recognise.
Can the Gemini image watermark be removed?
The visible sparkle can be removed from your own images. SynthID stays.
How do I remove SynthID?
SynthID is designed to stay through normal edits, and we do not provide ways to remove it.
Does every Gemini image have a visible watermark?
Many do, but some paid plans and developer tools may not show it. SynthID is still applied.