Google SynthID Detector Opens to Everyone: What It Can and Can't Verify, and API Outputs

Sources checked on 2026-10-09. This article summarizes public official information, which may change. Verify against the linked pages.

On 2026-10-07 Google opened SynthID Detector to everyone at synthid.com. According to Google's official blog, anyone can check whether an image, video or audio file was made with AI from Google or its partners, including OpenAI, NVIDIA, Kakao and, "soon," Apple. It is available globally, in English only for now.

What changed

• Wider access. In 2025 Google released an early version through a waitlist for journalists, media professionals and researchers. It is now open to everyone.
• Scale. Google says that since 2023 it has watermarked over 180 billion images and videos, plus 240,000 years of audio.
• Built-in checks. Search, the Gemini app and Chrome already had verification. Google says these features now handle over 1 million requests a day.
• Formats. Per TechCrunch, images include JPG, PNG, WEBP and HEIC, video includes MP4, MOV and WEBM, and audio includes WAV, MP3, FLAC and M4A, among others.
• Sign-in. Per Ars Technica, you must sign in (Google, OpenAI and Apple accounts are supported) and get roughly 10 checks a day. Google engineers told the outlet the cap helps stop people from using the checks to build watermark-removal tools. We found this only in secondary reporting.

It is not a general AI detector

This is a common point of confusion. Search Engine Journal quotes the site: "This is not a general AI detector." It can only detect media from the listed companies that adopted SynthID. The site's FAQ also says no signal is foolproof and recommends collecting multiple data points before deciding.

Google's help page is more specific about reading results. It describes the Gemini app's verification feature, which uses the same SynthID technology:

• Watermark detected: all or part of the image or video was created or edited by Google's AI models. Note "created or edited": AI retouching and full generation can look the same.
• Not detected: the content was not created or edited by Google AI, but it could have been made by other AI systems.
• Unclear: very simple or abstract content ("not enough details to watermark") or very small edits can produce an inconclusive result.
• Modified files: the watermark usually survives resizing, recoloring and compression, but after many alterations it may no longer be detected.

OpenAI's own verification page separately checks for C2PA metadata and SynthID watermarks, and says it is built to detect images and audio generated with ChatGPT, the OpenAI API or Codex.

For developers: do API outputs carry a watermark?

Only what the official documents state:

Google

• The Gemini API image generation docs state that all generated images include a SynthID watermark.
• The Veo docs say videos created by Veo are watermarked with SynthID, and the Gemini Omni docs say all generated videos include SynthID.
• Per the Google Cloud announcement, creations from Veo, Imagen and Lyria use SynthID by default.
• Text is different: a reply on Google's developer forum says Gemini API text output is not SynthID-watermarked. We saw only the forum reply, not formal documentation, so treat it as indicative.

OpenAI

• Per OpenAI's announcement, images generated through ChatGPT, Codex or the OpenAI API carry SynthID, and a 2026-07-31 update says audio was added.
• Per OpenAI's help page, API outputs are covered except models planned for imminent deprecation, such as gpt-image-1, gpt-image-1.5 and sora-2. Coverage can also vary by product, model, export path and file type.
• OpenAI's text watermark is its own textGrain, not SynthID.

When you call models through an OpenAI-compatible API such as BazaarLink, Google and OpenAI state that content from their AI tools receives SynthID. Whether a given file still carries the watermark after passing through a third-party API is something to confirm against official documentation, and we suggest spot-testing representative files with the official verification tools. BazaarLink itself does not offer watermark detection.

Related models on BazaarLink

These are the Google and OpenAI image and video models actually listed in the BazaarLink model list on 2026-10-09 (/api/v1/models?output_modalities=all). See each model page for specs and pricing.

Note that OpenAI's help page names gpt-image-1 among the models planned for imminent deprecation that fall outside coverage. For the others, check each vendor's documentation; this article does not vouch for any particular model.

How to verify, and the limits

1. Go to synthid.com, sign in and upload a file. Use the original, high-quality copy, as the site recommends; screenshots and repeated conversions reduce accuracy.
2. If the file came from an OpenAI tool, use OpenAI's verification page.
3. Also check C2PA Content Credentials, which record origin and edit history but can be lost to a screenshot or file conversion.
4. Keep the original file and a record of your checks, and do not rely on a single result.

Build these limits into your workflow. A missing watermark does not mean content is not AI-made, because many models, including open-weight ones you can run yourself, do not use SynthID. Services that claim to remove watermarks have also appeared on the market, so a watermark is a signal rather than forensic proof. Ars Technica suggests that a stronger long-term approach may be to use cryptographic methods such as C2PA to label real content at capture.

What it means for labeling and platform rules

SynthID gives platforms and brands one more way to check: you can self-check assets before publishing and cross-check unfamiliar images or clips. But it only covers participating vendors and does not replace the disclosure that platform rules require. If your product generates or forwards AI content, follow each platform's and your local jurisdiction's labeling rules, and keep generation records.

Sources (checked 2026-10-09)

• Google blog: Google expands SynthID Detector for AI content
• Google blog (2025 early version): SynthID Detector: Identify content made with Google's AI tools
• Google help: Verify AI-generated images, videos, and audio
• Google DeepMind: SynthID; tool: synthid.com
• Gemini API docs: image generation, Veo, Gemini Omni
• Google Cloud: Veo 3, Imagen 4 and Lyria 2 announcement
• OpenAI: verification page, Advancing content provenance, Provenance signals help page (OpenAI pages return 403 to automated requests; open them in a browser)
• Secondary reporting: TechCrunch, Ars Technica, Search Engine Journal (returns 403 to automated requests)
• C2PA: c2pa.org

Further reading

• All models and live pricing: /models

TWD billing · Taiwan invoices · leading AI models · OpenAI-compatible API