SynthID Detector
Check if an AI-generated image contains Google's SynthID watermark — the invisible signature Gemini and Imagen add to their outputs.
Images are processed in-memory and never stored. 10 scans per IP per hour.
How it works
Three steps from uploaded pixels to a verdict — no models to install, no setup.
Drop in any image
PNG, JPG, or WebP. We keep the file client-side while you preview it and stream the bytes to our detection service when you hit scan.
Spectral analysis
We run a 2D FFT over the image, isolate the mid-frequency bands SynthID uses, and compare the coefficients to a pre-built codebook of known watermark signatures.
Read the verdict
Get a yes-or-no answer plus a confidence score and raw details. Works whether SynthID is present, partially degraded, or absent entirely.
Why use this detector
Built specifically for SynthID. No generic "AI detector" hand-waving — just spectral matching against the real watermark.
Gemini and Imagen aware
Tuned against the public SynthID codebook used by Gemini, Imagen, and ImageFX — the models that actually stamp watermarks on their outputs.
Frequency-band matching
Unlike "AI detectors" that guess from style, we compare FFT coefficients to a known watermark codebook. The signal is either there or it isn't.
Resolution agnostic
Works on thumbnails, 4K renders, cropped chunks, and re-encoded JPEGs. Normalization happens before analysis so input size never changes the verdict.
Free, no sign-up
No account, no API keys, no credit card. Ten scans per hour are free and anonymous — enough for moderators, journalists, or curious users.
Common use cases
Who reaches for a SynthID detector and what they do with the result.
Verify AI provenance at moderation time
Trust and safety teams flag user-submitted images and want a fast answer on whether they came out of a Google model. A positive signal is strong evidence for policy decisions around disclosure or labeling.
Used by trust and safety teams, platform moderators, and content reviewers.
Probe watermarking robustness
Researchers studying invisible watermarking want to measure how resilient SynthID is to JPEG compression, resizing, inpainting, or style transfers. Run samples through the detector to chart confidence degradation.
Used by ML researchers, watermarking academics, and red-teamers.
Check if your art was scraped and re-generated
Digital artists suspicious that their work is being laundered through Imagen or Gemini prompts can test suspected look-alike images. A SynthID match confirms the pipeline.
Used by illustrators, photographers, and independent digital artists.
Perfect for
Anyone who cares whether a pixel came from a Google image model.
Sanity-check submitted photos before running a story. A SynthID hit is a strong signal that an image was synthesized rather than captured.
Investigative reporters, fact-checkers, editors.
Study watermarking behavior across models, measure robustness under adversarial transforms, and build datasets for papers on provenance.
Academics, grad students, applied ML teams.
Stock photo sites, marketplaces, and community platforms enforcing AI-disclosure rules can spot-check submissions with a deterministic signal.
Marketplaces, moderators, community operators.
Investigators correlating evidence for disinformation probes, legal discovery, or brand-abuse cases want machine-readable provenance hints, not guesses.
Forensic analysts, legal teams, OSINT investigators.