Learn
An AI photo detector analyzes the visual content of an image and estimates whether it was generated by artificial intelligence. It gives you a useful signal when a photo’s origin is unclear, rather than a certificate that the picture is real or fake.
Start with the best-quality version you can obtain. An original download usually carries more visual information than a small screenshot. Our AI photo detector accepts one image at a time; upload a local file. A public image URL can also be used when the active detector supports it.
Choose a JPG, PNG or WebP file from 1 KB to 10 MB. Link input is available only with a compatible detector. Sign in with Google to use your free monthly allowance. Avoid sending private or sensitive images unless you are authorized to share them for analysis.
Select Analyze photo. The detector evaluates visual patterns in the image, then returns a confidence score. Keep the page open while the request completes. A failed check does not consume paid credits.
Review the score and its assessment, then compare it with the source, the original file and other available evidence. If the result is uncertain, do not turn it into a definitive claim.
Check an unfamiliar image from a message, feed or search result before passing it on. Use the AI photo detector alongside a source check: find the earliest available copy and read the surrounding context.
Review supplied images when editing a newsletter, article or website. Ask contributors about generation and editing tools. A detector can support that conversation, but cannot replace a clear editorial policy.
Use the score as one piece of evidence when a profile photo, product image or visual claim seems unusual. Image generation, face manipulation and misleading captions are different problems; one score cannot resolve all three.
Yes. AI photo detection can produce false positives and false negatives. A real photo may receive a high score, and a generated image may receive a low one. Heavy compression, resizing, screenshots, filters, illustration styles and newly released generators can change the signals available to the model.
The percentage shows confidence in a classification, not the share of AI pixels. Real and AI-generated classifications need at least 80% confidence for the corresponding assessment. Edited, unknown and lower-confidence results remain uncertain.
The image was classified as real with high confidence. Check its source and context before relying on it.
The image may be edited, or the evidence is insufficient. Keep the result uncertain and look for the original.
The image was classified as AI-generated with high confidence. Use the heatmap and source as supporting evidence.
AI photo detector
Upload the photo to receive a confidence score and assessment. The AI photo detector can highlight evidence of generation, but it cannot give absolute proof. Check the original source before drawing a conclusion.
Look for a trustworthy original source, consistent context and any documented creation history. Visual clues can help, but unusual hands, text or lighting are not reliable proof on their own. Combine these observations with a detection score.
You can submit a screenshot saved as JPG, PNG or WebP. Screenshots, cropping and recompression can remove useful signals, so the original image is preferable when available. A low-resolution result should be interpreted cautiously.
This release uses a general AI-image generation model. It does not include a dedicated face-swap or identity verification service. An overall AI score should not be presented as a confirmed deepfake diagnosis.
You can submit images from these tools without knowing the source first. The current detector does not return verified generator attribution. Results and region maps need context, and new generator versions can affect detection.
Sign in with Google for 3 free checks per month. Buy a credit pack when you need more checks.
This site does not save the original image, filename or image URL. The image is sent to the configured detector. Returned region masks stay private for 24 hours for guests and 7 days for accounts. Signed-in history retains the assessment; shared reports exclude the original and mask and can be revoked.
An uncertain result means the returned category or its confidence does not support our likely-real or likely-AI assessment. It is a useful outcome when the evidence does not support a strong conclusion. Try to obtain a better-quality original and verify its source.