What Is SynthID?
Google DeepMind's invisible watermark that marks every AI-generated image — and why it's separate from the sparkle logo.
SynthID vs. the Visible Watermark
SynthID is an invisible digital watermark developed by Google DeepMind that is embedded directly into the pixel data of AI-generated images during creation. Unlike the visible sparkle (✦) logo, SynthID cannot be seen by the human eye and is practically impossible to remove.
Google applies two distinct watermarks to Gemini-generated images:
- Visible watermark (✦ sparkle logo) — A semi-transparent graphic in the bottom-right corner. This is what you see and what our tool removes.
- SynthID (invisible watermark) — An imperceptible signal embedded into the image's pixel data at generation time. It cannot be seen by the human eye.
These two systems serve different purposes and operate independently.
How SynthID Works
SynthID was developed by Google DeepMind and introduced in 2023. It works by making imperceptible modifications to the pixel values of AI-generated images during the generation process itself — not applied after the fact as a post-processing step.
Key characteristics:
- Invisible to humans — The modifications are below the threshold of human perception
- Robust to edits — SynthID survives screenshots, cropping, resizing, compression, color adjustments, and most common image edits
- Detectable by AI — Specialized models can read the embedded signal to determine if an image was AI-generated
- Baked into generation — It's part of the image generation pipeline, not a separate overlay
Can SynthID Be Removed?
Practically, no. Unlike the visible sparkle logo (which is a simple alpha-composited overlay), SynthID is woven into the fundamental pixel data. Removing it would require:
- Knowledge of the exact embedding algorithm (which is proprietary)
- The specific parameters used for each image
- Modifications that would significantly degrade image quality
Standard image editing (cropping, resizing, JPEG compression, filters) does not remove SynthID.
What Does This Mean for Watermark Removal?
When you use our AI Watermark Cleaner to remove the visible Gemini sparkle logo:
- The visible ✦ logo is removed — your image looks clean
- The invisible SynthID remains intact — the image is still traceable as AI-generated
This is by design. Our tool is intended to clean up the visual appearance of your images while respecting the integrity of AI content identification systems.
SynthID Beyond Images
As of 2026, SynthID has expanded beyond images. Google now applies similar invisible watermarking to:
- Text — Generated by Gemini language models
- Audio — Generated by audio AI models
- Video — Generated by video AI models
This reflects a broader industry trend toward AI content provenance and transparency.
Comparison
| Visible Watermark (✦) | SynthID | |
|---|---|---|
| Visibility | Visible to everyone | Invisible to humans |
| Location | Bottom-right corner | Entire image |
| Application | Post-composited overlay | Baked into generation |
| Removable? | Yes (reverse alpha blending) | Practically no |
| Survives edits? | No (cropping removes it) | Yes (robust to most edits) |
| Purpose | User-facing disclosure | Machine-detectable provenance |
Keep Reading
SynthID explains what stays in the file. These related pages explain what the visible watermark is, how removal works, and what people usually ask next.