Skip to content
Longterm Wiki
Back

SynthID-Image: Image watermarking at internet scale

paper

Authors

Sven Gowal·Rudy Bunel·Florian Stimberg·David Stutz·Guillermo Ortiz-Jimenez·Christina Kouridi·Mel Vecerik·Jamie Hayes·Sylvestre-Alvise Rebuffi·Paul Bernard·Chris Gamble·Miklós Z. Horváth·Fabian Kaczmarczyck·Alex Kaskasoli·Aleksandar Petrov·Ilia Shumailov·Meghana Thotakuri·Olivia Wiles·Jessica Yung·Zahra Ahmed·Victor Martin·Simon Rosen·Christopher Savčak·Armin Senoner·Nidhi Vyas·Pushmeet Kohli

Credibility Rating

3/5
Good(3)

Good quality. Reputable source with community review or editorial standards, but less rigorous than peer-reviewed venues.

Rating inherited from publication venue: arXiv

A key industry paper on practical AI-generated content authentication; relevant to policy discussions around mandatory watermarking of synthetic media and technical approaches to mitigating AI-enabled misinformation.

Paper Details

Citations
10
1 influential
Year
2025

Metadata

Importance: 62/100arxiv preprintprimary source

Abstract

We introduce SynthID-Image, a deep learning-based system for invisibly watermarking AI-generated imagery. This paper documents the technical desiderata, threat models, and practical challenges of deploying such a system at internet scale, addressing key requirements of effectiveness, fidelity, robustness, and security. SynthID-Image has been used to watermark over ten billion images and video frames across Google's services and its corresponding verification service is available to trusted testers. For completeness, we present an experimental evaluation of an external model variant, SynthID-O, which is available through partnerships. We benchmark SynthID-O against other post-hoc watermarking methods from the literature, demonstrating state-of-the-art performance in both visual quality and robustness to common image perturbations. While this work centers on visual media, the conclusions on deployment, constraints, and threat modeling generalize to other modalities, including audio. This paper provides a comprehensive documentation for the large-scale deployment of deep learning-based media provenance systems.

Summary

SynthID-Image is Google DeepMind's system for imperceptibly watermarking AI-generated images to enable detection of synthetic content at scale. The paper describes the technical approach to embedding and detecting watermarks that survive common image transformations, and reports on its deployment across Google's image generation products serving hundreds of millions of users.

Key Points

  • Presents a production-scale invisible watermarking system for AI-generated images deployed across Google products including Gemini and Imagen.
  • Watermarks are designed to be robust against common image manipulations like cropping, compression, and color adjustments while remaining imperceptible.
  • Addresses the challenge of provenance verification for synthetic media without requiring centralized databases or metadata that can be stripped.
  • Evaluates watermark robustness, false positive/negative rates, and image quality trade-offs at internet scale.
  • Represents a practical deployment of AI content authentication technology relevant to misinformation and deepfake detection efforts.

Cited by 1 page

PageTypeQuality
AI Content AuthenticationApproach58.0

Cached Content Preview

HTTP 200Fetched Mar 20, 202698 KB
\\correspondingauthor

sgowal@google.com, pushmeet@google.com\\paperurl\\reportnumber

# SynthID-Image: Image watermarking at internet scale

Sven Gowal
Google DeepMind
Core contributor (randomized order)
Lead
Rudy Bunel
Google DeepMind
Core contributor (randomized order)
Florian Stimberg
Google DeepMind
Core contributor (randomized order)
David Stutz
Google DeepMind
Core contributor (randomized order)
Guillermo Ortiz-Jimenez
Google DeepMind
Core contributor (randomized order)
Christina Kouridi
Google DeepMind
Core contributor (randomized order)
Mel Vecerik
Google DeepMind
Core contributor (randomized order)
Jamie Hayes
Google DeepMind
Core contributor (randomized order)
Sylvestre-Alvise Rebuffi
Google DeepMind
Core contributor (randomized order)
Paul Bernard
Google DeepMind
Contributor
Chris Gamble
Google DeepMind
Contributor
Miklós Z. Horváth
Google DeepMind
Contributor
Fabian Kaczmarczyck
Google DeepMind
Contributor
Alex Kaskasoli
Google DeepMind
Contributor
Aleksandar Petrov
Google DeepMind
Contributor
Ilia Shumailov
Google DeepMind
Contributor
Meghana Thotakuri
Google DeepMind
Contributor
Olivia Wiles
Google DeepMind
Contributor
Jessica Yung
Google DeepMind
Contributor
Zahra Ahmed
Google DeepMind
Support
Victor Martin
Google DeepMind
Support
Simon Rosen
Google DeepMind
Support
Christopher Savčak
Google DeepMind
Support
Armin Senoner
Google DeepMind
Support
Nidhi Vyas
Google DeepMind
Support
Pushmeet Kohli
Google DeepMind
Lead
Sponsor

###### Abstract

We introduce SynthID-Image, a deep learning-based system for invisibly watermarking AI-generated imagery.
This paper documents the technical desiderata, threat models, and practical challenges of deploying such a system at internet scale, addressing key requirements of effectiveness, fidelity, robustness, and security.
SynthID-Image has been used to watermark over ten billion images and video frames across Google’s services and its corresponding verification service is available to trusted testers.
For completeness, we present an experimental evaluation of an external model variant, SynthID-O, which is available through partnerships.
We benchmark SynthID-O against other post-hoc watermarking methods from the literature, demonstrating state-of-the-art performance in both visual quality and robustness to common image perturbations.
While this work centers on visual media, the conclusions on deployment, constraints, and threat modeling generalize to other modalities, including audio.
This paper provides a comprehensive documentation for the large-scale deployment of deep learning-based media provenance systems.

###### keywords:

provenance, watermarking, robustness

## 1 Introduction

In the wake of powerful generative artificial intelligence (AI) systems and their accessibility in products such as Gemini, ChatGPT, Midjourney or ElevenLabs
including their corresponding open-source alternatives (Rombach et al., [2022](https://ar5iv.labs.arxiv.org/html/2510.09263#bib.bib74 ""); Touvron et al., [202

... (truncated, 98 KB total)
Resource ID: 10607c87667b587e | Stable ID: NmM5YjA3MW