Face-swap boundary artifacts
Blending seams at the jawline and hairline, softened eye and teeth detail, and a face whose noise floor does not match the surrounding frame.
Maintained by DetectFake · Updated . See our methodology and limitations.
Video forensics
AI video detector for deepfakes and synthetic clips
Upload a video and DetectFake samples frames across its timeline, hunts for deepfake, face-swap, and AI-generation artifacts, and returns a verdict with a confidence score and the exact evidence behind it. Free to start, no install, and the original file stays on your device.
Analyze a video now 3 free scans every day
What our AI video detector looks for
Every scan reports individual forensic signals rather than a single opaque score, so you can weigh the evidence yourself.
Lighting and shadow mismatch
Key light direction on a subject that disagrees with the scene, missing contact shadows, and specular highlights that stay fixed while the head turns.
Generative texture signatures
Over-smooth skin, repeating fabric and hair patterns, melted background text, and warped hands or jewellery typical of diffusion-generated video.
Frame-to-frame inconsistency
Details that drift between sampled frames — changing earrings, flickering teeth, shifting collar geometry — a strong tell for synthesized motion.
How to detect AI video in 4 steps
- 1
Upload your clip
Drop an MP4, MOV, or WebM file up to 60 MB. Nothing is installed and no account is required to run your first scans.
- 2
Frames are sampled locally
Your browser extracts and downscales evenly spaced frames across the timeline, so the original video never leaves your device.
- 3
Forensic analysis runs per frame
Selected frames are assessed using local classification models and image statistics. Sampling can miss brief edits; the report identifies the model or limited fallback used.
- 4
Read the verdict and evidence
You get Authentic, Manipulated, or AI-Generated with a confidence score, a plain-language summary, and the signal list behind it.
Who uses video deepfake detection
Journalists and fact-checkers
Triage viral clips before publication and keep a reviewable record of which signals raised the alarm.
Trust & safety teams
Screen user-submitted video for synthetic media and non-consensual face swaps as a first-pass filter in a moderation queue.
Recruiters and verification
Screen recorded interviews and testimonial clips as one part of source verification. A scan does not verify identity or establish eligibility for an account.
Everyday scam checks
Test a celebrity endorsement, investment pitch, or family-emergency video message before you act on it.
Video detection limits worth knowing
No detector is infallible. Aggressive compression from social platforms, re-encoding, beauty filters, heavy grain, and screen recordings all destroy the fine detail that forensic analysis depends on, which pushes results toward "inconclusive". Very short clips give fewer frames to compare, and a genuine video shot in unusual lighting can legitimately look odd. DetectFake is built for triage: it shows its reasoning so a human can make the final call, and it is not a substitute for a formal forensic examination.
AI video detector FAQs
How does an AI video detector work?
DetectFake samples still frames across the timeline and runs forensic vision analysis on each one. It looks for generation and editing artifacts — inconsistent lighting and shadows, warped edges around faces and hands, unnatural texture or noise patterns, blending seams, and frame-to-frame inconsistency — then combines the per-frame signals into a single verdict with a confidence score.
Can you detect deepfakes and face swaps?
DetectFake screens selected still frames for AI-generation and manipulation signals. It can miss face swaps, short edits between samples and unfamiliar generators. It does not authenticate audio, perform a dedicated lip-sync test or verify every frame. Preserve the original clip and corroborate the result.
Which video formats and sizes are supported?
MP4, MOV, and WebM files up to 60 MB. Frames are sampled and downscaled in your browser before analysis, so large clips still work — free scans sample 5 frames, Silver samples 8, and Gold samples 12 for deeper temporal coverage.
Is the result proof that a video is fake?
No. A detector result is one screening signal, not proof of manipulation or authenticity. Compression, re-encoding, filters and screen recordings can obscure or imitate artifacts. Confidence is not a measured accuracy rate. Verify the source, context and original file independently.
Is my video uploaded or stored?
The original video remains on your device. Normal analysis runs locally on sampled, downscaled frames after model files download. If inference fails, those frames and the filename are sent to DetectFake’s server for a limited fallback assessment. Scan history, including a preview, is stored in your browser.
How many videos can I check for free?
Three scans per day on the free Regular plan. Silver raises this to 30 scans a day with 8-frame sampling, and Gold is unlimited with 12 sampled frames per video.
Check a video for AI right now
Three free scans a day, no signup. Need more? Silver adds 30 daily scans with 8-frame sampling and Gold gives unlimited 12-frame deep analysis.
Open DetectFake