An AI image detector, a Content Credentials validator and a provider watermark verifier answer different questions. The useful comparison starts with the evidence you need, not a single “best” score.
This comparison was checked on September 25, 2026. AIFakeScan behavior is based on our executed 60-file case study and implementation checks. The Google, Adobe and Sightengine entries are limited to their current official documentation unless a row explicitly says it was tested.
Quick selection table
| Tool | Primary evidence | Best fit | What a negative result cannot prove | Evidence basis here |
|---|---|---|---|---|
| AIFakeScan AI Image Detector | Local visual model plus available file evidence | Triage an unknown image and see separate evidence with limitations | That the file is authentic or was never AI edited | Tested locally on 60 recorded file versions |
| AIFakeScan C2PA Checker | Embedded Content Credentials, validation and local trust status | Inspect provenance, signer status, declared actions and basic metadata | That a credential-free file is real | Tested locally on the same 60 files |
| Google Gemini verification | SynthID for Google AI plus supported Content Credentials | Ask whether Google AI created or edited an image | That another generator was not used | Vendor documentation; no compatible source in our experiment |
| Adobe Content Authenticity Inspect | Content Credentials and recoverable matching credentials | Inspect credential history in Adobe's web workflow | That missing credentials establish authenticity | Interface availability and vendor documentation; no upload result recorded |
| Sightengine genai API | Visual pixel classification through a hosted API | Programmatic visual screening when an external provider is acceptable | Provenance, signer identity or universal correctness | Vendor documentation; no paid API run in this comparison |
AIFakeScan: combine file evidence without collapsing it
The general AI Image Detector runs a local visual model and reports available file evidence. The dedicated C2PA Checker uses a provenance-only profile and does not run the pixel classifier or a paid detector.
In our 60-file case study, the C2PA checker found valid but locally untrusted generated-content manifests in four generated originals. The 16 transformed versions derived from those originals had no embedded credential. The visual model's mean for the generated group changed from 0.9968 on originals to 0.5226 on JPEG quality 60 versions. Those observations demonstrate why the two evidence types should stay separate; they do not rank AIFakeScan against another detector.
Public Basic Scan files are processed on AIFakeScan's server infrastructure and deleted after processing with a cleanup backstop. The checker is not browser-local. Review the current privacy policy before uploading sensitive media.
Google Gemini: use it for the official SynthID route
Google's help page says signed-in eligible Gemini users can upload one file and ask whether it was created or edited by Google AI. Gemini currently recognizes SynthID for content from Google AI tools and can also summarize supported Content Credentials.
This is the appropriate route when the narrow question is about Google's watermark. It is not a universal watermark test. A “not detected” SynthID result does not rule out other AI systems, and Google documents an unclear outcome for some files.
Use the SynthID checker route to prepare the file, copy a precise prompt and record the returned outcome. AIFakeScan does not decode SynthID itself.
Adobe Content Authenticity: inspect credential history
Adobe's web Inspect experience is designed for Content Credentials. Adobe documentation describes viewing origin, edit history and generative-AI information where credentials are available, including recovery methods for some matching credentials.
The public site was reachable during this review, but we did not record an upload result. Its entry in the table is therefore a documented capability, not a measured result. It is useful as a second interface for credential review, especially in Adobe workflows.
Sightengine: a hosted visual API
Sightengine documents a hosted `genai` model that examines pixel content and returns an AI-generated score, with per-generator information in some access levels. The vendor states that metadata is not used for that visual score and offers a test tier and paid production plans.
We did not send the 60-file set to Sightengine, so this article does not compare its scores or accuracy with AIFakeScan. Use it when an external API and its current data handling, limits and pricing fit your workflow.
Match the tool to the question
| Your question | Start here | Add when needed |
|---|---|---|
| Does this file carry verifiable origin or edit history? | C2PA checker | Adobe Inspect for another credential interface |
| Was this made or edited by Google AI? | Gemini SynthID verification | C2PA and source review |
| Does the pixel content resemble learned generated-image patterns? | AI image detector | A second visual provider and controlled file-version comparison |
| Is the depicted event or caption true? | Source investigation and corroboration | Reverse image search, original publisher and contextual evidence |
No tool in this table proves factual truth. Provenance can record who signed a history; a watermark can identify a compatible provider; a visual model can produce a probabilistic signal. Keep the findings separate until you write the final conclusion.
Privacy, access and cost notes
| Tool | File processing | Access and price note checked September 25, 2026 |
|---|---|---|
| AIFakeScan | Uploaded to AIFakeScan's Cloudflare-hosted processing service; short-lived private result | Public Basic Scan during beta; check current limits on the tool page |
| Gemini | Uploaded to Google under the account and product terms | Sign-in and an eligible account are required; Google documents daily verification quotas |
| Adobe Inspect | Uploaded or supplied to Adobe's web inspection workflow | Public web access was available; account and feature availability can change |
| Sightengine | Sent to Sightengine's hosted API | Vendor documents a free testing tier and paid production plans; check current pricing |
Frequently asked questions
Which AI image detector is the most accurate?
This comparison does not make that claim. A defensible accuracy comparison needs a versioned, representative labeled dataset, identical inputs, predefined metrics and complete failures. The tools here also answer different questions.
Should I use C2PA or a visual detector first?
If the original file is available, check provenance first because a validated declaration can be direct evidence about file history. Then use visual and contextual checks for questions the credential does not answer.
Can one tool check every invisible AI watermark?
No public tool in this comparison establishes universal coverage across proprietary watermark systems. Use the provider's compatible verifier and record exactly which system ran.
Sources
- Verify AI-generated images, videos and audio — Google
- Content Credentials Verify — Content Authenticity Initiative
- Content Credentials overview — Adobe
- C2PA and Content Credentials Explainer — C2PA
- AI-generated Image Detection API — Sightengine
- AIFakeScan 60-row experiment record — AIFakeScan