Question 1

How accurate is the face swap detector?

Accepted Answer

There is no honest single accuracy number across face-swap tools, edits and platforms. Original files preserve the most evidence; heavily compressed or re-encoded uploads are harder. Treat the result as strong evidence to weigh alongside context, not absolute proof.

Question 2

Can it detect face swaps in both photos and videos?

Accepted Answer

Yes. Photos go through the image pipeline (provenance, pixel-level signals and visual reasoning); videos are sampled across their full duration and each detected face is reviewed for swap artifacts, alongside a cross-modal reasoning pass.

Question 3

Is this video real or AI? Can I check a video call recording?

Accepted Answer

Yes. If you suspect a live video call used a real-time face swap, record a short clip and upload it to the video detector. Boundary artifacts around the jawline and hairline often survive in recordings.

Question 4

Is the face swap check free?

Accepted Answer

Yes — no signup is required. Service usage limits apply to keep the tool available for everyone.

Question 5

What happens to my upload?

Accepted Answer

Uploads are used only to analyze the current request and are not retained by DeepFakeCheck. We never train models on user uploads.

Question 6

What can I do if my own face was swapped into an image or video?

Accepted Answer

Keep the original file as evidence and run it through the detector to document the manipulation signals. In the US, the TAKE IT DOWN Act now obliges platforms to remove non-consensual intimate imagery, including AI face swaps, after a valid report.

Question 7

Is the viral 'three-finger test' a reliable way to catch a live face swap?

Accepted Answer

It still helps today: holding fingers or a hand in front of the face breaks many real-time face-swap tools, which is why the test went viral. But detection specialists already warn it is losing reliability as tools learn to handle occlusion. Use gesture tests as a first screen, then record a clip of the call and upload it to the video detector for an evidence-based check.

Question 8

Does the EU AI Act require deepfakes to be labeled?

Accepted Answer

Partly, from August 2, 2026: whoever deploys a deepfake must disclose that the content was artificially generated or manipulated (Article 50(4)). The separate duty for AI providers to embed machine-readable markings (Article 50(2)) applies from day one to tools launched on or after August 2, 2026, while tools already on the market before that date have until December 2, 2026. Scammers ignore labeling duties either way, so independent detection remains necessary. General information, not legal advice.

AI換臉檢測