An image or photo
JPG, PNG, WebP, HEIC, AVIF and more
Up to 20 MB and 50 megapixels
YOUR FIRST SCANHow AI Fake Scan works.
How AI Fake Scan works.
From a question
to a closer look.
Bring an image or a short video. We inspect the available evidence and explain its limits, so you can decide what to check next.
01 / CHOOSEStart with the file.
Start with the file.
Keep its context.
Use the original download when available. Keep a note of who shared it, when it appeared and what it claims to show. That context helps you follow up on the scan.
A short video
MP4, MOV, WebM, MKV, AVI and more
Up to 200 MB and 2 minutes
Choose one file from your device, then select Scan for AI evidence. File decoding, image dimensions and video duration are validated after upload. Both uploaders accept images and videos. For a still image and paired MOV, use the Live Photo detector. Links to social posts and live streams are not accepted.
02 / INSPECTWhat happens
What happens
after you press scan?
Once the upload finishes, you move to a private progress page. Keep that tab open while the file waits in the queue and the enabled checks run.
The file is checked
We check whether the media can be processed and inspect available embedded information. Camera and editing-software fields provide context; they do not settle whether AI was used.
Source evidence is inspected
Where supported, Content Credentials are checked for valid, trusted declarations. Visible AI labels may add clues when text recognition is available. Missing credentials or labels remain neutral findings.
The visual checks follow the media
Images receive available passive visual checks. Videos are sampled across time, with selected moments revisited and audio-track information checked separately. Basic Scan does not classify face swaps or synthetic speech.
The report puts the findings together
The report separates generation, manipulation and AI-audio coverage, then explains the supporting evidence and missing checks. Passive signals remain Inconclusive until calibration is validated.
03 / UNDERSTAND
Read past the headline.
Start with the finding, then check which category it concerns and what actually ran. A result about visual generation cannot answer an untested question about a voice.
Likely AI
A trusted AI source declaration can support this finding in the current beta. Read whether it describes generation or editing; it does not necessarily mean every part of the media was generated.
Inconclusive
The evidence, calibration or coverage is insufficient for a responsible binary answer. This is the current outcome for passive-only findings. It does not mean “half real,” “half fake” or a failed upload.
No strong AI evidence
This label requires the necessary calibrated checks to finish below their thresholds. It is not the default when markers are absent, and it is not available from uncalibrated passive checks in this beta.
Coverage tells you how far the answer goes.
These labels describe the checks, not the truth of the content. Read them alongside “What the scan found” and “Limits of this result.”
- Complete
- What it meansThe required checks for this category finished within the scan’s defined scope.
- How to use itRead the finding and limitations. Complete coverage is not proof of authenticity or a check of every video frame.
- Partial
- What it meansOnly part of the intended checking scope was completed.
- How to use itLook for the missing checks before using the result to make a decision.
- Not tested
- What it meansThis type of detection did not run.
- How to use itKeep that question open. For example, Basic Scan does not classify AI speech.
- Not applicable
- What it meansThat category does not apply to this file, such as audio in a video with no audio track.
- How to use itDo not read this as a passed authenticity check.
- Failed
- What it meansA check could not produce a usable result.
- How to use itRead the reported limitation. A failed detector is not evidence for or against AI.
A score is not the whole answer. An AI evidence score is not a verified probability that the file is fake. Check the evidence source and category before interpreting any number.
04 / FOLLOW UPTurn the finding
Turn the finding
into a better question.
Detection is one part of checking a claim. The next useful step depends on what the report found and why you needed the scan.
If AI use is declared
Ask what was generated or edited, and whether that use was disclosed. An AI-assisted illustration and a fabricated event recording make very different claims.
If evidence is limited
Request the original file, look for the earliest publication and compare independent sources. A cleaner copy may preserve context that a screenshot or re-encoded download lost.
If a video moment stands out
Replay the surrounding action in your original clip. A review timestamp points to sampled evidence, not an exact boundary around an edit. Check the soundtrack separately.
GETTING STARTED, EXPLAINED
Questions about your first scan
Practical help with processing, uncertain results and private report access.
Do I need to sign up before my first scan?
Basic Scan does not require an account. Select one supported file and complete the verification prompt if one appears. Usage limits may apply. Deep Scan is not publicly available, so there is no paid scan option to unlock on this page.
What should I do while my scan is processing?
Keep the result page in the same browser tab. It updates as the task moves through the queue and processing. Video checks can take longer because frames and audio need to be extracted. There is no guaranteed completion time in this beta; avoid submitting the same file repeatedly while its first scan is still running.
Why can a scan finish without a yes-or-no answer?
Finishing the process and having enough evidence for a conclusion are different things. A file may have no trusted source declaration, a check may be unavailable, or the visual model may not be calibrated. The current beta keeps passive-only results Inconclusive rather than converting those signals into a measured AI probability.
Is the AI evidence score a probability that the file is fake?
No. An evidence score is not a verified probability that the file is fake, and AI use does not by itself establish deception. Read the source of the evidence, the category it concerns and the coverage. Uncalibrated passive signals do not provide a measured AI probability in this beta.
Can I send someone the result page link?
The URL alone does not grant access to the report. Access uses a private per-scan token stored in this browser, with session storage as a fallback. A different browser or device will not automatically have access. Keep the scan URL and avoid clearing its browser storage if you want to manage retained uploads later. There is no public report-sharing feature in the current workflow.
What if my upload or scan fails?
Read the error first. Check that the file is supported, non-empty and within the upload limits. A supported extension can still contain an unsupported codec or damaged data. For a connection error, try again once your connection is stable. If the file cannot be decoded, use another available source copy and retain the original. Failure is not an AI finding.
READY WHEN YOU AREStart with one question.