Frequently Asked Questions Fresh
Comprehensive FAQ about Google SynthID verification.
General Questions
What is SynthID?
SynthID is a digital watermarking technology developed by Google DeepMind that embeds invisible markers into content generated by Google AI models. These watermarks are imperceptible to humans but can be detected by SynthID verification tools. It applies to images, videos, and audio generated by Google AI.
Why was SynthID created?
As AI-generated content becomes increasingly difficult to distinguish from human-created content, SynthID provides a transparency mechanism. It allows anyone to verify whether content was produced by Google AI models, helping combat misinformation and maintain trust in digital media.
How does SynthID work?
SynthID embeds invisible digital watermarks directly into the content at the time of generation. These watermarks are:
- Invisible to the human eye/ear
- Embedded within the content itself (not attached as separate metadata)
- Resilient to common transformations (compression, resizing, recoloring, format conversion)
- Machine-detectable by the SynthID verification system
Is SynthID free to use?
Yes. SynthID verification is free through the Gemini App, subject to daily usage quotas (~20 images, 10 videos/5 min, 3 hours audio per day).
Watermarks and Detection
What is the difference between watermarks and metadata?
| Feature | Watermark (SynthID) | Metadata |
|---|---|---|
| Location | Embedded in the content itself | Attached as separate data to the file |
| Visibility | Invisible to humans | Visible in file properties |
| Resilience | Survives editing, compression, format changes | Can be stripped by editing software or re-saving |
| Reliability | High — difficult to remove without degrading content | Low — easily removed intentionally or accidentally |
Does the SynthID watermark survive editing?
Yes. SynthID watermarks are designed to survive common transformations including:
- Rescaling (resizing up or down)
- Cropping (removing parts of the image)
- Recoloring (adjusting hue, saturation, brightness)
- JPEG compression (saving at different quality levels)
- Format conversion (e.g., PNG to JPEG)
However, extremely heavy editing or complete regeneration of the content may degrade or remove the watermark.
Can SynthID detect content from non-Google AI systems?
No. SynthID only detects watermarks embedded by Google AI models. Content generated by other AI systems (Midjourney, DALL-E, Stable Diffusion, Runway, ElevenLabs, etc.) will return a "no watermark detected" result.
A "no watermark" result means the content was not created by Google AI. It does not mean the content is authentic or human-created.
What if I get an "unclear" result?
An "unclear" result means SynthID cannot make a definitive determination. This may happen when:
- The file is too small or low resolution
- The content was too heavily edited after generation
- Only a very small portion was AI-generated
- The file format does not carry enough signal
In these cases, use alternative detection methods: visual artifact inspection, reverse image search, and EXIF metadata analysis.
Verification Process
Can I verify multiple files at once?
No. SynthID verifies one file at a time. Upload and verify each file individually.
What file types can I verify?
- Images: JPEG, PNG, WebP, and other common image formats
- Videos: MP4 and other common video formats (max 90 seconds)
- Audio: MP3, WAV, and other common audio formats (max 1 hour)
- Not supported: Text documents, PDFs, spreadsheets
What are the daily verification limits?
| Content Type | Daily Limit |
|---|---|
| Images | ~20 checks |
| Videos | 10 checks / 5 minutes total |
| Audio | 3 hours total |
What accounts can use SynthID verification?
- Personal Google accounts (free, 18+)
- Google Workspace accounts (if enabled by admin)
- Supervised accounts may have limited access
Does SynthID verify text content?
No. SynthID currently only verifies images, videos, and audio. Text content verification is not supported.
Privacy and Data
Does Google use my uploaded files for training?
For personal accounts, conversations may be reviewed by human annotators to improve product quality. Do not share sensitive information in Gemini conversations.
For Workspace accounts, conversation data is not used to train Google AI models. Workspace data is subject to enterprise-grade data processing agreements.
How long does Google keep my verification data?
Your verification data follows your Gemini Activity settings:
- Auto-delete: Choose 3, 18, or 36 months
- Manual delete: Delete specific conversations or all activity anytime
- Pause activity: Stop saving new activity entirely (disables Drive access)
Manage these settings at myactivity.google.com.
Can my Workspace admin see my verification activity?
Workspace administrators have access to audit logs and can see Gemini usage patterns. Refer to your organization's IT policy for specifics about monitoring and data access.
Technical Questions
Why can't SynthID detect all AI-generated content?
SynthID only detects watermarks that Google specifically embeds in content generated by its AI models. Other AI companies use different (or no) watermarking systems. There is currently no universal AI content detection standard.
Can someone remove a SynthID watermark?
SynthID watermarks are extremely difficult to remove without significantly degrading the content quality. They are embedded throughout the content at a level that is intertwined with the content itself. Standard editing operations (crop, resize, compress, recolor) do not remove the watermark.
Is SynthID 100% accurate?
SynthID has high accuracy for content generated by Google AI models, but no verification system is perfect. Factors that can affect accuracy include:
- Heavy editing after AI generation
- Very low resolution files
- Extreme compression
- Only a small portion being AI-generated
For the most reliable results, verify original, uncompressed files whenever possible.