What is a SynthID remover and how does it work?
A SynthID remover, in the sense people search for, would have to change pixel values. SynthID Pass does not do that. The tool above strips file headers: EXIF, GPS, XMP, IPTC and C2PA. When an AI model generates an image, it can stamp markers in two separate places: the file header and the pixel grid.
Understanding these two layers helps creators inspect and manage digital provenance across their workflows:
| Layer | Where it lives | What reads it | What the owner says about durability |
|---|---|---|---|
| Metadata records | EXIF headers, IPTC tags, C2PA manifests in the file container | Photo apps, social platforms, Content Credentials viewers | Lost when a file is re-saved without them, for example in a screen capture |
| Pixel watermark | SynthID pattern in the pixel values | Google’s SynthID Detector, Gemini, Search, Chrome | Google says it is designed to remain detectable after filters, color changes and lossy compression |
| Visible watermarks | Logos or badges drawn into the picture | People looking at the image | Visible to anyone, so not a hidden signal |
Metadata tools clean text records stuck onto the file wrapper. For example, Content Credentials from the Coalition for Content Provenance and Authenticity attach signed manifests. If a user posts a photo to a chat app that strips tags, the manifest drops away, but the pixels stay untouched.
SynthID does not use file tags. Google DeepMind introduced the technology on August 29, 2023. The system weaves a pattern into the pixel lattice of images and video frames, as well as the audio wave. Because the mark lives in the media content, stripping headers leaves the watermark behind. A true watermark tool must process the pixel values directly.
What SynthID Pass cleans
SynthID Pass strips image container metadata in your browser: EXIF, GPS, XMP, IPTC and C2PA content credentials. The file is not uploaded, and the compressed pixel data is not rewritten.
That is a different job from removing SynthID. Google’s watermark lives in pixel values. Google says it is designed to remain detectable after filters, color changes and lossy compression. A file with the headers removed can still carry SynthID. Check the result on the SynthID Detector.
- SynthID Pass is an independent project and has no affiliation with Google, DeepMind, or Alphabet.
- The cleaner runs in the browser. Nothing is uploaded or stored.
- It removes container metadata. It does not edit pixels and does not remove SynthID.
- Video and audio files are not processed.
Which AI generators embed SynthID watermarks?
Watermark use has expanded across major commercial AI platforms. On October 7, 2026, Google DeepMind reported that over 180 billion images and videos and 240,000 years of audio carry SynthID watermarks across partner networks.
Several main model families add SynthID when rendering files:
- Google Gemini and Imagen: Google adds SynthID to images generated with its Imagen models, including images made in the Gemini app, according to Google DeepMind.
- Google Veo: Google’s video model adds SynthID to each generated video segment.
- Google Lyria and NotebookLM: Audio from Lyria and podcast clips from NotebookLM carry an inaudible SynthID mark that Google says withstands noise, MP3 compression and speed changes.
- Partner companies: NVIDIA began watermarking videos from its Cosmos preview service with SynthID in 2025, per Google’s I/O 2025 post. Google’s October 2026 post names OpenAI, NVIDIA and Kakao as partners whose media the detector checks.
Files from these tools carry embedded marks that stay with the media as it moves across social sites and storage drives.
How to verify content on Google’s SynthID Detector
Anyone can check whether an asset carries an active watermark by uploading files to the public SynthID Detector portal. Google opened the site to newsrooms in May 2025 and launched it globally to the public on October 7, 2026.
The detector evaluates uploaded media and returns verdicts observed in the client on October 8, 2026:
- Watermark detected: Client title displays “Made with Google AI” alongside the subtitle “SynthID detected in all or part of the uploaded content” (specifying Video, Audio, or Audio & Video tracks). This indicates the file came from a participating model.
- Watermark not detected: Client title displays “Not made with Google AI” with the subtitle “SynthID is not detected anywhere in the uploaded content”. This means the file came from an untracked model, was made before integration, or heavy edits weakened the signal. It does not prove the image was made by a human.
- Unsure if made with Google AI: Client title displays an unsure status. Observed client reasons include frame rates under 10 fps, low resolution, extreme file length, or suspected alterations.
The public detector accepts files including JPG, PNG, WebP, AVIF, HEIC, MP4, MOV, MP3, and WAV up to a cap of 100 MiB per file. Each account has usage quotas, and the portal’s terms forbid automated or adversarial use.
Responsible use and legal boundaries for watermark processing
Using the metadata cleaner brings real ethical and legal duties. Provenance tags tell platforms and viewers something about a file. Stripping them to mislead users or dodge laws brings serious risk.
Rules on synthetic media disclosure are tightening. Article 50 of the EU AI Act (Regulation (EU) 2024/1689) requires providers of generative AI systems to mark their output in a machine-readable way, and requires deployers who publish deepfakes to disclose that the content is artificial. In the United States, Section 1202 of the Digital Millennium Copyright Act forbids removing copyright management information when done to enable or conceal copyright infringement.
Do not use the cleaner to break platform rules, impersonate real people, or hide artificial origins where disclosure is required. These notes do not constitute legal advice.
Explore the SynthID Pass resource cluster
Our technical resource cluster explains how digital watermarks function across specific tools, file types, and verification surfaces. Each guide reviews provenance handling in depth across modern AI generation pipelines:
- SynthID Watermark Remover: Learn how pixel watermarks differ from corner logos and C2PA metadata tags.
- Remove SynthID from Image: Image formats, what Google says the image watermark survives, and how image verdicts read.
- Gemini SynthID Remover: Specific details on Google Gemini, Imagen, and Nano Banana output files.
- ChatGPT SynthID Remover: How OpenAI’s partner role influences ChatGPT image provenance.
- Remove SynthID from Video: Video detection mechanics across Veo models and the 10 fps limit.
- How Does SynthID Work: How the watermark is added and detected in images, audio, video and text.
- SynthID vs C2PA: Complete comparison of signed cryptographic metadata versus pixel watermarks.
- SynthID Detector Guide: Practical guide to synthid.com verdicts, quotas, and account limits.