Have the original? Start there.
Upload the file you received or downloaded directly from its source. Avoid taking a screenshot or re-saving it just for the scan: that can change both embedded information and pixel patterns.
Check a photo for evidence of AI generation or editing. Read its source information, inspect available AI signals, and see what the file can actually tell you.
A useful check starts with realistic expectations. Visual signals remain Inconclusive during beta. Trusted AI source evidence can support a stronger finding.
Upload the original if you have it. We identify the media and show the matching report.
No files selected.
Loading secure verification…
Uploaded originals are kept privately for future analysis until you delete them. The report expires separately. Use “Delete uploads and report” on your scan page. Privacy details.
Private upload. Keep this browser’s access token to manage deletion.
Images up to 20 MB · 50 megapixels
Videos also work here: up to 200 MB / 2 min.
Have a Live Photo pair? Check both components.
An original photo, a social media download and a screenshot may look alike. They don’t necessarily carry the same information into a scan.
Upload the file you received or downloaded directly from its source. Avoid taking a screenshot or re-saving it just for the scan: that can change both embedded information and pixel patterns.
You can inspect a supported downloaded image, but resizing and compression may weaken useful signals. Missing camera fields after sharing do not mean the photo was generated.
The scan describes the file you upload, not the unseen original. A screenshot adds interface elements and may lose credentials; a crop leaves part of the image outside the check. Ask for the source file when the answer matters.
Read each finding alongside its limits. The report identifies checks that were unavailable, so an untested feature doesn’t look like a clean result.
Camera fields, editing software and embedded export information can help explain how a file was made.
How to interpret itEditable context. A camera name or Photoshop export alone cannot settle whether an image contains AI.
When supported, C2PA verification checks signed source declarations, including statements about AI generation or editing.
How to interpret itA declaration needs valid, trusted credentials to support a strong finding. Missing or broken credentials are not evidence of AI.
Where text recognition is available, on-image labels can provide clues about a possible AI source.
How to interpret itLabels can be copied, cropped or added later. Basic Scan does not verify every invisible vendor watermark.
A passive image model checks learned visual patterns that may be associated with generated images.
How to interpret itCalibration is pending in this beta. These signals alone remain Inconclusive and do not produce a measured AI probability.
AI generation, local manipulation and source authenticity are related, but separate questions. A result about one does not settle the others.
A trusted credential may identify generation or AI editing. Check which category the declaration concerns. An AI-editing declaration does not automatically mean every part of the image was generated.
This can mean the available evidence is insufficient or the visual model is not calibrated. Ask for the original file, find the earliest available publication, and check the publisher’s explanation.
An absent label or credential is one finding, not an authenticity verdict. It may never have been present, may have been removed during sharing, or may need a verifier that did not run.
Use the guide that matches the question left open by your scan.
Review documented provider scopes and access status. Use the OpenAI image verifier when enabled, or follow an official external route. Pending integrations remain clearly marked.
Run a focused C2PA check that keeps missing credentials, validation failures and signer trust separate.
Compare file history and supporting evidence when a result conflicts with what you know about the photo.
Inspect the downloaded image, then trace the original publisher, date and caption. A genuine photograph can still be shared with a false story.
Look for source evidence, but don’t infer someone’s identity or intentions from a detector result. Retouching, studio lighting and compression can make real portraits look artificial.
Ask the supplier for the source file and an editing disclosure. An AI scan does not establish ownership, commercial usage rights or whether a pictured product exists.
Practical answers for original photos, edited pictures and the copies that travel around the internet.
Upload a supported original image, run Basic Scan, then read the evidence and coverage together. Start with the original file when possible. A trusted declaration of AI generation can support a stronger finding; uncalibrated visual signals or missing metadata cannot. For an important photo, also verify who published it and where the original came from.
You can upload a supported image regardless of its claimed source. That does not mean every generator or model version is reliably detectable. We do not publish generator-specific accuracy claims for this beta, and Basic Scan cannot promise to identify which tool made an image.
Whole-image generation and local editing are different questions. A trusted source declaration may disclose AI editing, but a general image signal does not reliably locate an edited region or rule out a face swap. Specialized face-swap detection is not included in Basic Scan. Read manipulation coverage separately from generation coverage.
Yes, if it is a supported image file within the upload limits. However, the screenshot may no longer contain the original metadata or Content Credentials, and its pixels have changed. Treat a screenshot result as evidence about that copy, not a verification of the original image.
No. Perspective, motion blur, retouching and compression can make real photos look unusual, while generated images can look convincing. We do not use an opinion about whether a scene looks plausible as a standalone verdict. An unusual detail is a reason to investigate the source, not to accuse its creator.
JPG, PNG, WebP, AVIF, HEIC/HEIF, BMP, single-page TIFF, GIF and APNG are accepted, up to 20 MB and 50 megapixels. GIF, APNG and animated WebP use bounded frame sampling; very long or complex animations may be rejected. HEIC image collections use the primary image, with that limitation reported. RAW and PDF are not supported. This uploader also accepts short videos and routes them to video analysis.
Yes. Basic Scan does not require an account, and usage limits may apply. It provides the evidence available to the enabled checks. Deep Scan is not publicly available yet, and the beta does not promise a definitive answer or calibrated probability for every image.
Original uploads are retained privately for future analysis until you delete them. Reports expire after 45 minutes, independently of the originals. Use Delete uploads and report on your scan page, including after report expiry. Keep the scan URL and this browser’s stored access token; clearing browser storage can remove your access.
Upload one image and explore the evidence behind it.