Can you remove SynthID from video files today?
Can you remove SynthID from video files today? Not with SynthID Pass: video support is planned and not available yet. When creators seek to remove SynthID from video clips, they find that re-saving the file does not clear the mark, because SynthID sits in the content of the video rather than in file tags.
Google DeepMind explains on its SynthID overview page that the watermark is added directly to “an AI-generated image (or video segment)” at the moment of creation. The mark embeds imperceptible digital signals into the media itself. Unlike standard container metadata such as EXIF blocks or C2PA provenance manifests, these embedded signals do not sit in an optional file header.
Where does SynthID video come from across Google products? Google’s I/O post of May 20, 2025 lists Veo among the Google models whose content carries SynthID, and the same post said NVIDIA was watermarking videos from its Cosmos preview service. On October 7, 2026, Google reported that over 180 billion images and videos had been watermarked. Google’s Gemini help also says all AI-generated media created or edited with Gemini Apps, videos included, carries an invisible SynthID watermark.
Generative video clips often pair visual footage with synthesized speech, sound effects or music. The SynthID Detector reports the two separately: a positive video verdict names the track where it found the mark, video, audio, or both. For audio, Google uses SynthID in tools such as Lyria and NotebookLM audio overviews.
What modifications does Google say SynthID video survives?
SynthID video watermarks are built to stay in place across routine distribution steps on the web. According to the Google DeepMind SynthID documentation, the watermark is “added the moment content is created, and designed to stand up to modifications like cropping, adding filters, changing frame rates, or lossy compression.”
The following table outlines what Google states regarding common post-processing actions:
| Video Modification | What Google States | Notes and Limits |
|---|---|---|
| Spatial Cropping | Designed to survive (DeepMind) | Named in the SynthID overview |
| Adding Filters | Designed to survive (DeepMind) | Named in the SynthID overview |
| Changing Frame Rates | Designed to survive (DeepMind) | Named in the SynthID overview |
| Lossy Compression | Designed to survive (DeepMind) | Named in the SynthID overview |
| Color Adjustments | Not specifically covered by Google statement | Google names filters, not color grading as such |
| Container Swap (MP4 / MOV / WebM) | Not covered by a Google statement | A container swap without re-encoding leaves the frames themselves unchanged |
| Extreme Distortion | Not infallible (synthid.com FAQ) | Google does not define the limit |
| Low Frame Rate (< 10 fps) | Detector returns Unsure (portal client) | Client strings indicate insufficient frames per second for evaluation |
According to Google’s public FAQ on synthid.com, the watermark is not infallible, and SynthID is one layer in a broader approach to content provenance rather than a standalone guarantee.
Google does not publish how much compression or scaling a video can take before detection stops. For still images, see our guide to removing SynthID from an image and the SynthID remover overview.
How the SynthID Detector reports video verdicts
The official SynthID Detector portal provides browser-based verification for uploaded media. The portal opened for global English access on October 7, 2026. A detailed analysis of the detector web client on October 8, 2026, revealed how the portal processes video assets.
When a user uploads a video file, the detector inspects visual frames and audio tracks separately. Client bundle analysis shows that the detector presents one of several exact text strings depending on what it finds:
- Made with Google AI (Video): SynthID detected in the video track, while the audio track has no detected watermark.
- Made with Google AI (Audio): SynthID detected in the audio track, while visual frames have no detected watermark.
- Made with Google AI (Audio & Video): SynthID detected across both the visual video track and the accompanying audio track.
- Not made with Google AI: SynthID is not detected anywhere in the uploaded content.
In addition to positive and negative results, the detector client contains an Unsure if made with Google AI verdict state. This state is accompanied by specific reason strings when file parameters fall outside reliable thresholds. For video media, the client strings observed on October 8, 2026, define the following explicit uncertainty reasons:
- Less than 10 frames per second:
The __ has less than 10 frames per second. - Low resolution:
The __ resolution is too low. - Low quality:
The __ is too low quality. - Duration too long:
The __ is too long to be processed. - Duration too short:
The __ is too short to be processed. - Not enough details:
Not enough details to watermark. - Small modification:
We suspect the image has been tampered with or Google AI was only used for a very small part.
An unsure outcome is not a pass; it means the portal could not reach a confident reading for that file. To learn more about reading detector results, visit our SynthID Detector guide.
File limits and platform constraints across Google tools
Submitting video content for detection requires navigating specific technical limits across official interfaces. The standalone SynthID portal and the verification features in Google Gemini follow different operational rules.
The table below contrasts video specifications across official Google verification interfaces:
| Interface Parameter | Standalone SynthID Portal | Google Gemini Verification |
|---|---|---|
| Maximum File Size | 104,857,600 bytes (100 MiB) | 100 MB |
| Video Formats Accepted | MP4, MOV, WebM (client MIME check) | Not listed in help documentation |
| Duration Bounds | Minimum and maximum bounds enforced | Under 90 seconds per video clip |
| Verification Scope | Video and Audio tracks independently | Google AI SynthID and C2PA Content Credentials |
| Daily Quota | Account-level quota enforced | Approximately 10 videos (up to 5 min total) per 24 hours |
According to the official Google Gemini help documentation, Gemini checks both SynthID and C2PA Content Credentials. However, Gemini restricts video uploads to clips under 90 seconds. It also imposes a rolling 24-hour quota of approximately 10 videos totaling no more than 5 minutes.
On the standalone SynthID Detector portal, the web client limits uploads to 104,857,600 bytes (100 MiB). Files exceeding this threshold trigger an immediate “File is too large” message in the browser. In addition, the portal requires users to sign in with an authenticated account before submitting video files for scanning.
You can read our comparison of provenance methods in our synthid vs c2pa guide to understand how metadata manifests differ from pixel-level watermarks.
SynthID Pass product status and responsible use
SynthID Pass is an independent project currently in active development: a tool for removing the SynthID watermark, starting with images. Video support is planned for a later phase.
We outline our operational boundaries plainly:
- Video tools are planned: SynthID Pass does not offer an automated tool to remove SynthID from video files today.
- Nothing is uploaded today: Files selected in the drop zone above stay on your device; the drop zone only records interest.
- Independent project: SynthID Pass is entirely independent and is not affiliated with Google or its subsidiaries.
- No circumvention instructions: We do not publish removal recipes, circumvention pipelines, or links to bypass repositories.
- Responsible provenance standards: Watermarks serve an important public purpose in identifying generative media. Digital tools should never be used to misrepresent AI footage as human camera recordings or bypass mandatory disclosure rules.
- Getting updates: Early access opens on synthidpass.com first as video tools develop. You can try it when early access begins.