Google DeepMind’s SynthID is one of the most significant developments in AI content provenance. Unlike visible watermarks that you can see and remove, SynthID works at a level you cannot perceive at all. This guide explains exactly how SynthID works and what it means for creators using Google’s AI tools.

What Is SynthID?

SynthID is Google DeepMind’s proprietary invisible watermarking technology, designed to mark AI-generated content in a way that is imperceptible to humans but detectable by Google’s verification tools. It was first announced in August 2023 and has since been expanded across multiple Google AI products.

SynthID works by making subtle, mathematically computed modifications to the pixels of an AI-generated image. These modifications are so small that they are completely invisible to the human eye — you cannot see any difference between a SynthID-watermarked image and one without it. However, when analyzed by SynthID’s detection algorithm, the watermark pattern is clearly detectable.

Which Google Products Use SynthID?

As of 2026, SynthID is embedded in outputs from:

• Google Gemini: All AI-generated images from Gemini include SynthID
• Imagen 3: Google’s flagship image generation model includes SynthID by default
• Veo (video): Google’s AI video generator embeds SynthID in all video outputs
• Google Search AI overviews: AI-generated images in search results include SynthID
• Google Cloud Vertex AI: Images generated via the API include SynthID

How Is SynthID Different from a Visible Watermark?

The difference is fundamental:

• Visible watermarks are a layer added on top of the image — a logo, text, or overlay. They can be seen, selected, and removed using inpainting.
• SynthID modifies the actual pixel values of the image itself. There is no separate layer. The watermark IS part of the image data, woven invisibly into the visual information.

Does SynthID Survive Image Editing?

One of SynthID’s key design goals is robustness. According to Google DeepMind’s research, SynthID is designed to survive many common image modifications including:

• JPEG compression (even aggressive compression)
• Cropping (unless very aggressive)
• Rotation and minor geometric transforms
• Brightness and contrast adjustments
• Color grading and filters
• Screenshot captures

More aggressive modifications — like significant upscaling/downscaling, heavy AI enhancement, or extensive compositing — may degrade the SynthID signal.

Can SynthID Be Removed?

Removing SynthID is significantly more difficult than removing a visible watermark. Since the watermark is woven into the pixel data, standard inpainting tools cannot target it. Approaches that may partially disrupt the SynthID signal include:

• Aggressive JPEG compression at very low quality settings
• Heavy image noise addition and subsequent denoising
• Significant resolution reduction followed by upscaling
• Adversarial perturbation techniques (academic/research tools)

However, these approaches degrade image quality significantly and may not completely remove the watermark signal. For most practical purposes, SynthID in Google Gemini outputs is very difficult to reliably remove.

What About the Visible Gemini Watermark?

The visible watermark (the “Made with Gemini” logo or badge) on Gemini images is separate from SynthID and can be removed using standard AI inpainting. Visit our Gemini watermark removal tool to remove the visible watermark for free. This does not affect the underlying SynthID invisible watermark.