Invite-only preview · paid checkout is not open. See availability

Accuracy, providers & limits

Reviewed 16 November 2026

Each claim below belongs to a specific tool and test. We do not combine provider figures into an overall MyImageLabs accuracy score.

Local metadata, editing and exports are available. Public automatic AI and reverse-search checks remain disabled; the full automated Trust Report is prelaunch.

AI image detection: Sightengine

Sightengine reports 98.3% accuracy in a 2024 benchmark evaluating 80,000 real and text-to-image AI images. This is a provider-reported benchmark result, not measured MyImageLabs accuracy or a guarantee for your image.

Sightengine's article, dated 26 September 2024, attributes the result to researchers from the Universities of Kansas and Rochester using the ARIA dataset. The cited result concerns real images and text-to-image outputs; it does not establish the same performance for all edited images, screenshots, newer generators or customer uploads.

Read Sightengine's benchmark write-up · AI detection documentation

A confidence score returned for one image is not a measured accuracy rate. False positives and false negatives are possible. MyImageLabs has not completed an independent end-to-end accuracy benchmark.

Face manipulation: a separate Sightengine model

The optional deepfake check uses Sightengine's separate face-manipulation model. We have not verified a separate published accuracy percentage for this feature. The 98.3% GenAI benchmark figure does not apply to this check.

Deepfake model documentation

Reverse search: search engines through SerpApi

Our integration requests exact and visual matches from Google Lens through SerpApi. The multi-engine option also requests Bing and Yandex results for supported public image URLs. Local files are not automatically published to enable those additional engines; unavailable or failed engines must be distinguished from searches returning no matches.

We have not verified a published matching-accuracy or original-source discovery percentage for these integrations. API uptime and request-success rates do not measure whether a match is correct or whether all copies were found. Results depend on each engine's indexed content and the submitted image.

Search matches are leads, not proof of theft, ownership or identity. No matches do not establish that an image is genuine. The earliest dated result found may not be the original source.

SerpApi Google Lens documentation · SerpApi Yandex documentation

EXIF and metadata: ExifReader

Our current build uses ExifReader 4.54.0 to read supported embedded fields from JPEG, PNG and WebP files locally. Available fields may include camera, dates, editing software and GPS. Support varies with the file and metadata present; we do not claim to extract every possible field.

We have not verified a published extraction-accuracy percentage. Metadata can be changed or removed, so reading a field does not authenticate its contents. The library supports other formats, but that does not mean MyImageLabs currently accepts them.

ExifReader documentation and source

Reports, editing and saved work

MyImageLabs' own reporting code presents available findings, source links and limitations. PDF export does not independently validate a finding. Local edits create a copy; the clean-copy tool re-encodes a PNG without the original embedded metadata, but visible identifying information remains. Browser storage and downloaded bundles preserve work; they do not certify it.

How we report performance

We identify the provider, benchmark scope, date and source alongside any accuracy figure. A published result applies to its test conditions, not all images or every feature. We do not present software regression-test counts as evidence of detection accuracy. Claims will be reviewed as providers and our implementation change.

Questions or corrections: hello@northernmedia.com.au.