1. Upload or paste: Drop a clip from your device or paste a direct video file URL (MP4, MOV, WebM, AVI · up to 5 min free, max 50 MB).; Drop a file or paste a video link (max 50 MB / 5 min free).
2. Frame-by-frame analysis: Our detection engine samples frames across the clip and scans each for face-swap artifacts and known generator signatures.; Scans keyframes in memory for face-swaps and synthetic artifacts.
3. Verdict & report: Real or Deepfake / AI — with separate deepfake and AI-generation scores, the likely generator, and how many frames were analyzed.; Confidence score, likely generator and a PDF report.

How to spot a deepfake video

A detector is the reliable check, but a few tells are worth knowing — especially on a clip you can only watch, not upload. Slow it down and look for these.

Key forensic indicators to inspect across faces, audio, and motion.

Blinking & eye movement: Deepfaked faces often blink too rarely, too regularly, or with eyes that do not quite track where the person is looking.
Lip-sync & mouth interior: Watch the mouth on hard consonants (b, p, m) and look inside it — teeth that blur or merge and lips that drift out of sync are classic face-swap tells.
Face edges & hairline: The boundary where a swapped face meets hair, ears, and jaw is the hardest thing to fake. Look for flicker, warping, or a faint seam, especially on head turns.
Lighting & shadows: A face lit differently from its surroundings, or shadows that fall the wrong way, means the face was composited in.
Skin texture: AI skin is often too smooth and even — missing pores, stray hairs, and the small asymmetries of real skin — or it shimmers subtly from frame to frame.
Teeth, jewelry & fine detail: Generators struggle with rigid, repeating detail: teeth that change count, earrings that morph, glasses frames that wobble.
Audio & voice: Cloned voices can sound flat or unnaturally smooth, and the audio may drift a beat off the lips. Check the sound and the mouth together.
Motion & physics: Fully AI-generated clips betray themselves in motion — too-smooth "gliding" movement, hands with the wrong number of fingers, or backgrounds that warp and melt as the camera moves.

Blinking & eye movement

Deepfaked faces often blink too rarely, too regularly, or with eyes that do not quite track where the person is looking.

Lip-sync & mouth interior

Watch the mouth on hard consonants (b, p, m) and look inside it — teeth that blur or merge and lips that drift out of sync are classic face-swap tells.

Face edges & hairline

The boundary where a swapped face meets hair, ears, and jaw is the hardest thing to fake. Look for flicker, warping, or a faint seam, especially on head turns.

Lighting & shadows

A face lit differently from its surroundings, or shadows that fall the wrong way, means the face was composited in.

Skin texture

AI skin is often too smooth and even — missing pores, stray hairs, and the small asymmetries of real skin — or it shimmers subtly from frame to frame.

Teeth, jewelry & fine detail

Generators struggle with rigid, repeating detail: teeth that change count, earrings that morph, glasses frames that wobble.

Audio & voice

Cloned voices can sound flat or unnaturally smooth, and the audio may drift a beat off the lips. Check the sound and the mouth together.

Motion & physics

Fully AI-generated clips betray themselves in motion — too-smooth "gliding" movement, hands with the wrong number of fingers, or backgrounds that warp and melt as the camera moves.

No single tell is proof: real compression, filters, and re-recording can mimic some, and the best fakes hide them all. When it matters, run the clip through a frame-by-frame detector instead of trusting your eye.

Note: Compression and re-encoding can mimic artifacts; run clips through frame-by-frame analysis.

Which AI video generators we detect

Our detector recognizes the fingerprints modern text-to-video and face-swap tools leave behind, reporting the most likely source.

Model	What it is	Signature we look for
ByteDance Seedance 2	Kinetic human motion & dance synthesis.	Foot-plant traction slippage against ground planes and limb-occlusion dissolve artifacts.
OpenAI Sora 2	Photorealistic text-to-video diffusion.	Fluid "mass-merge" where objects flow into one another; uncanny contact physics.
Google Veo 2	Cinematic video model by DeepMind.	Specular lighting disconnecting from scene geometry; drifting particle details.
Runway Gen-3	Cinematic video & motion generator.	Edge softening and lateral motion-blur smearing on rapid camera moves.
Kling 1.5	High-realism human motion generator.	Micro-jitter in fingers and subtle facial "breathing" between frames.
Alibaba Wan 2.1	Open high-definition diffusion video.	3D VAE decompression smearing under rotational motion.
MiniMax Hailuo	Text/image-to-video model.	Latent-decode smearing and temporal flicker in high-detail texture regions.
Luma Dream Machine	Fast camera motion & simulation model.	Texture crawl and warping across fine background structures.
Pika 2.0	Effects-driven viral video generator.	Non-physical "melt" and crush boundaries in signature effects.
Tencent Hunyuan	Large open-source video model.	Temporal flicker in textured regions across consecutive frames.

ByteDance Seedance 2

Kinetic human motion & dance synthesis.

Signature: Foot-plant traction slippage against ground planes and limb-occlusion dissolve artifacts.

OpenAI Sora 2

Photorealistic text-to-video diffusion.

Signature: Fluid "mass-merge" where objects flow into one another; uncanny contact physics.

Google Veo 2

Cinematic video model by DeepMind.

Signature: Specular lighting disconnecting from scene geometry; drifting particle details.

Show 7 more generators (Runway, Kling, Alibaba, MiniMax, Luma, Pika, Tencent)▼

Runway Gen-3

Cinematic video & motion generator.

Signature: Edge softening and lateral motion-blur smearing on rapid camera moves.

Kling 1.5

High-realism human motion generator.

Signature: Micro-jitter in fingers and subtle facial "breathing" between frames.

Alibaba Wan 2.1

Open high-definition diffusion video.

Signature: 3D VAE decompression smearing under rotational motion.

MiniMax Hailuo

Text/image-to-video model.

Signature: Latent-decode smearing and temporal flicker in high-detail texture regions.

Luma Dream Machine

Fast camera motion & simulation model.

Signature: Texture crawl and warping across fine background structures.

Pika 2.0

Effects-driven viral video generator.

Signature: Non-physical "melt" and crush boundaries in signature effects.

Tencent Hunyuan

Large open-source video model.

Signature: Temporal flicker in textured regions across consecutive frames.

Verify against all generators

Deepfake vs. AI-generated: what’s the difference?

People use the words interchangeably, but they are two different things — and we score them separately.

	Deepfake (manipulated real footage)	AI-generated (fully synthetic)
What it is	Real video of a real person, altered — usually a face swap or lip re-enactment.	The entire clip is invented by a model from a prompt; no camera ever filmed it.
Typical use	Impersonating a specific person — a catfish "proof" video, a fake executive, a celebrity.	A person or scene that never existed — fake influencers, scam ads, "historical" hoaxes.
How we detect it	Face-swap seams, blending artifacts, and inconsistencies around the edited face.	Generator fingerprints and synthesis artifacts across the whole frame.
Where it shows in your result	The Face-Swap / deepfake score.	The Synthetic-AI score.

Deepfake (manipulated real footage)

Face-swap / voice clone

What it isReal video of a real person, altered — usually a face swap or lip re-enactment.

Typical useImpersonating a specific person — a catfish "proof" video, a fake executive, a celebrity.

How we detect itFace-swap seams, blending artifacts, and inconsistencies around the edited face.

AI-generated (fully synthetic)

Diffusion / synthetic

What it isThe entire clip is invented by a model from a prompt; no camera ever filmed it.

Typical useA person or scene that never existed — fake influencers, scam ads, "historical" hoaxes.

How we detect itGenerator fingerprints and synthesis artifacts across the whole frame.

Frequently asked questions

6 Answers

How does deepfake video detection work?

The video is split into sampled frames and each frame is scanned for face-swap artifacts, generation fingerprints, and frequency patterns left behind by synthetic-media tools. A video is flagged when suspicious signals persist across frames — a single glitchy frame is not enough.

Which AI video generators can it detect?

The detector recognizes signatures from major video generators including ByteDance Seedance 2, OpenAI Sora 2, Runway Gen-3, Kling 1.5, Google Veo 2, Alibaba Wan 2.1, MiniMax Hailuo, Pika 2.0, and Luma Dream Machine, alongside classic face-swap deepfakes of real footage.

What is the difference between “deepfake” and “AI-generated”?

AI-generated means the whole video was synthesized by a model like Sora or Runway. A deepfake is real footage of a real person that has been manipulated — typically a face swap or re-enactment. We report both scores separately.

How accurate is it?

It is probabilistic, not definitive. Heavy compression, screen recordings, filters, very low resolution, and brand-new generators can all reduce accuracy. Treat the result as one strong signal alongside other context — where the video came from, who sent it, and whether the source can be verified.

Do you store my video or personal data?

Checked videos are deleted right after your scan — they are analyzed ephemerally in volatile memory and discarded. We keep a record of each check (scores, your IP address, approximate location, and basic device info) for rate limiting, fraud prevention, and legal compliance.

Is it free? Are there limits?

Yes. Free: 8 scans a day with a free account. Video clips up to 5 minutes and 50 MB are free, with non-expiring credit packs available for longer clips and high-volume investigations.

How does deepfake video detection work?

Which AI video generators can it detect?

What is the difference between “deepfake” and “AI-generated”?

How accurate is it?

Show 2 more questions▼

Do you store my video or personal data?▼

Is it free? Are there limits?▼

Yes. Free: 8 scans a day with a free account. Video clips up to 5 minutes and 50 MB are free, with non-expiring credit packs available for longer clips and high-volume investigations.

Protect your media and detect AI impersonations in real time

Analyze unknown video clips, detect deepfakes, and verify authentic identity processed in memory and deleted after your scan, with PDF scan reports.

Deepfake detection is probabilistic, not definitive. Consider results alongside other forensic indicators.

Scan a video free View forensic plans & pricing →