Fake photo check
Is this photo real?
A picture is going round and something feels off. Start with the automatic check below, then work through the five-minute checklist — together they catch far more fakes than either alone.
Free · no sign-up · your file never leaves your device
Drop the photo you’re unsure about, click to choose, or paste
JPEG · PNG · WebP · AVIF · HEIC (Safari) — checked on your device
Real, fake, or real but misleading?
“Is this photo real?” is really three questions, and it helps to separate them.
- Was it made by a camera at all? This is what AI detection answers. Generated images from Midjourney, ChatGPT, Gemini, FLUX and others can now look exactly like snapshots.
- Was it altered? A real photo with a face swapped, an object removed or a sign rewritten is a different kind of fake. Some edits are AI-made and some are old-fashioned retouching.
- Is the caption true? Most viral misinformation uses entirely real photos with the wrong date, place or story. No pixel analysis can catch that; only checking the source can.
The tool on this page addresses the first question directly and helps with the second. The checklist covers the third.
What the automatic check looks at
GPTTrace inspects the file, not just the picture. It looks for Content Credentials — signed records that cameras from Leica, Sony, Nikon and Google Pixel, and generators from OpenAI, Adobe and Microsoft, embed in images. It reads EXIF metadata, which in a genuine phone photo includes the device model, lens, exposure and timestamp. It looks for generator fingerprints such as Stable Diffusion settings or the IPTC “made by AI” label. Then it analyses the pixels with a neural classifier trained on thousands of generators, plus frequency-domain and error-level forensics.
Each finding is listed with whether it pushed toward “AI” or toward “real”. A photo with full camera metadata and a low classifier score is very likely a genuine capture; an image with no metadata, a generator-standard size and a high classifier score very likely is not. Most re-shared images fall somewhere in between, which is why the reasons matter more than the percentage.
Common situations
- A shocking news photo on social media — see the fake news photo checker for verification steps journalists use.
- A dating or social profile that seems too perfect — the fake profile picture checker covers AI faces and stolen photos.
- A face in a video call or clip that looks off — try the deepfake detector.
- A receipt, ID or document photo — AI now renders convincing paperwork; see the AI receipt detector.
Whatever the situation, keep the file you were sent rather than a screenshot of it. Every re-save throws away evidence.
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.
Five-minute checklist for a suspicious photo
- Run the detector above on the best copy you have, and read the reasons, not just the number.
- Find the first appearance. Use a reverse image search (Google Lens, Bing, TinEye) and sort by date. AI images often trace back to a generator community, an art account or a satire page.
- Check the context. Is the event, place and date confirmed by anyone else? Real news is photographed by many people from many angles.
- Zoom in on the details that generators still get wrong: text on signs and clothing, jewellery and glasses, hands holding objects, reflections, shadows that disagree with the light.
- Ask for the original. A person who took a photo can send the file from their camera roll. GPTTrace will show its camera make, model and capture time.