Verification
Fake news photo checker
Breaking news brings a flood of images, and some are generated, edited or recycled from another event. Check the file here, then trace it back the way professional fact-checkers do.
Free · no sign-up · your file never leaves your device
Drop the news photo, click to choose, or paste
JPEG · PNG · WebP · AVIF · HEIC (Safari) — checked on your device
Three ways a news photo is fake
Professional fact-checkers sort misleading images into roughly three groups, and each needs a different check.
Generated
An image model created the whole scene: an explosion near a landmark, a politician being arrested, a flooded city. These spread fastest when they confirm what people already fear. The detector on this page is built for them: it reads generator credentials and runs a neural classifier that recognises the statistics of generated pixels.
Manipulated
A genuine photo with something added, removed or changed — a weapon placed in a hand, a sign rewritten, a crowd enlarged. Modern manipulation is often done with AI editing tools. If the editor recorded it, GPTTrace reports an “edited with generative AI” credential. If not, crop to the suspicious region and check it separately.
Mis-captioned
The most common kind: a real, unedited photo presented with the wrong time, place or story. No detector can catch this, because nothing about the pixels is false. Reverse image search, geolocation and contacting the source are the only defences.
Reading a detector result during a breaking story
Images circulating on social platforms have usually been re-encoded several times, stripping metadata and blurring fine detail. Expect fewer “proof” results and more probabilities. A high classifier score on a dramatic image with no traceable source is a strong reason not to share it. A low score does not make an image trustworthy — it may simply be a real photo from somewhere else.
Look at the list of reasons. Camera metadata that names a phone and a capture time from the day of the event supports authenticity; a generator-standard size such as 1024×1024 with no camera data points the other way. When the result is inconclusive, the steps above will settle it more reliably than another tool.
Tools fact-checkers use
Beyond this detector: reverse image search engines for the earliest copy, satellite and street-level imagery for geolocation, sun-position calculators for checking shadows against the claimed time, and archive services for finding deleted original posts.
How accurate is it? Our measured numbers
We test GPTTrace on labelled image samples and publish the results, including where it does badly. It is tuned to keep false accusations rare, so it misses some AI content rather than flag real work.
image check: AUC 0.938 (cross-validated)
679 labelled samples (399 AI, 280 human), run 2026-10-08. At the “Likely AI” line it caught 64% of AI samples and wrongly flagged 5% of human ones.
| Source | Truth | Samples | Result at “Likely AI” |
|---|---|---|---|
| gemini-nano-banana | AI | 40 | 40% caught |
| midjourney-v6 | AI | 40 | 68% caught |
| midjourney-v5 | AI | 40 | 73% caught |
| flux-dev | AI | 40 | 13% caught |
| flux-schnell | AI | 40 | 48% caught |
| sdxl | AI | 40 | 100% caught |
| gpt-image | AI | 40 | 30% caught |
| kling | AI | 39 | 97% caught |
| leonardo-stablecog | AI | 40 | 98% caught |
| bitmind-imagine-mix | AI | 40 | 80% caught |
| fullsize-photos | Human | 40 | 10% wrongly flagged |
| open-images-photos | Human | 40 | 5% wrongly flagged |
| lfw-faces | Human | 40 | 0% wrongly flagged |
| caltech-objects | Human | 40 | 3% wrongly flagged |
| coco-photos | Human | 40 | 0% wrongly flagged |
| ffhq-faces | Human | 40 | 0% wrongly flagged |
| celeba-faces | Human | 40 | 18% wrongly flagged |
Data sources and method: methodology & accuracy.
How fact-checkers verify a news photo
- Save the highest-quality copy and run it through the detector. Note whether it carries Content Credentials from a camera or news agency.
- Reverse-search it with two engines. Look for the earliest date and the original uploader.
- Geolocate: match buildings, signs, terrain and road markings against map and street-level imagery.
- Chronolocate: check shadows, weather and daylight against the claimed time; compare with other footage of the same event.
- Contact the source and ask for the original file and how it was taken.