Check AI Images and Videos with SynthID Detector

AI-generated images, videos, and audio now look realistic enough that visual inspection often fails. Google launched SynthID in 2023 to embed an imperceptible watermark into media created by its models. On 7 October 2026 it opened the SynthID Detector to everyone at synthid.com. The tool now also reads watermarks from partner companies.

Anyone with a Google, Apple, or OpenAI account can upload a file and receive a clear result: watermark detected or not detected. This is not a general AI classifier. It only finds the specific SynthID signal. A negative result does not prove a file is human-made.

What SynthID Detector checks

SynthID embeds a hidden signal into the pixels of images and video frames, and into the waveform of audio. Humans cannot see or hear it. The signal is designed to survive common edits such as compression, cropping, resizing, and adding noise.

Google states it has applied the watermark to more than 180 billion images and videos plus 240,000 years of audio since 2023. Built-in checks in Search, the Gemini app, and Chrome already handle over one million requests daily. The new public site adds a direct upload path and expands partner coverage.

Current partner detection includes Google (Gemini, Imagen, Veo, Lyria, and related tools), OpenAI, NVIDIA, and Kakao. Apple support is listed as coming soon. The detector highlights which portions of a longer video or audio file carry the watermark when it finds one.

Supported formats are broad. Images include JPEG, PNG, BMP, WEBP, AVIF, HEIC, HEIF, TIFF, and GIF. Video covers MP4, MOV, and WEBM. Audio includes WAVE, MP3, OGG, FLAC, AAC, and M4A. Text is not supported on this portal.

Step-by-step: check a file

1. Open synthid.com in a browser.
2. Sign in with a Google, Apple, or OpenAI account. No email-only option is available.
3. Upload one image, video, or audio file that matches a supported format.
4. Wait a few seconds for the scan.
5. Read the result. “Watermark detected” means the file was likely generated or edited with a SynthID-enabled model from Google or a listed partner. The tool may identify the source company and mark segments in video or audio.
6. “Watermark not detected” means no confident SynthID signal was found. The file may still be AI-generated by a non-partner model, generated before a partner adopted SynthID, or altered enough to remove the mark.

You can also ask the Gemini app whether an uploaded image or video was created or edited by Google AI. That check looks for SynthID and Content Credentials. The dedicated site is useful when you want partner coverage beyond Google or a simple upload workflow.

BrenTech demonstrated the public site shortly after launch:

Limits you must understand

A positive detection is strong evidence that a supported AI system created or edited the media. A negative detection is not proof of authenticity. Many popular generators do not use SynthID. Local models, older versions of partner tools, and non-partner services leave no mark for this detector to find.

The tool does not distinguish between content that was fully generated by AI and content that was only edited with AI. It also does not analyze visual style or artifacts the way some third-party classifiers do. Login is required, and usage is subject to Google’s terms.

Screenshots of watermarked images can still trigger detection in some cases because the mark lives in the pixels. Heavy editing, re-encoding through certain pipelines, or deliberate removal attempts can reduce or eliminate the signal. Google does not claim the watermark is impossible to remove.

For on-device AI context protection on Android, Google is separately rolling out AISeal, a hardware-isolated vault. That architecture stores personal assistant context and is unrelated to media watermark detection. See the Android AISeal guide for details on what it currently covers.

Practical tips

• Check the original file when possible. A screenshot or re-encoded version may weaken the signal even if the original was watermarked.
• Combine the SynthID check with other signals. OpenAI’s verify tool at openai.com/verify looks for both C2PA Content Credentials and SynthID on its outputs. C2PA readers can inspect signed provenance metadata when it is present.
• Use the result for context, not as the sole basis for high-stakes decisions. Journalists, platforms, and researchers still need additional verification methods.
• Expect the partner list to grow. Apple is already named as forthcoming. More companies may adopt the same watermark standard.
• Built-in Gemini, Search, and Chrome checks remain available for quick looks without visiting the dedicated site.

Conclusion

The public SynthID Detector at synthid.com gives anyone a free, direct way to check whether an image, video, or audio file carries a watermark from Google or its current partners. Upload the file after signing in, and the tool reports whether the specific SynthID signal is present and, when found, which partner’s system likely produced it.

The result is useful evidence when it is positive. It is incomplete when negative. Treat the detector as one verification step alongside other provenance tools and common-sense checks. As more companies embed the same watermark, the coverage of a single public site will increase.

Sources

• Google expands SynthID Detector for AI content — Google Blog
• SynthID Detector
• SynthID — Google DeepMind
• Google’s SynthID AI Detection Tool is Now Available For All — BrenTech (YouTube)