Viral videos & deepfakes

Is this video
real?

A video is spreading. Something about it feels off — the motion, the face, the voice. You can check whether it's real, AI-generated, or a deepfake in seconds. No download required. Just paste the link.

Check the video now → How deepfake detection works

Types of fake video spreading online

Not all fake videos are created the same way. Understanding the type helps you know what to look for and how KweliAI's verdict applies:

Fully AI-generated video
Created entirely from a text prompt using tools like Sora, Runway, Pika, or Kling. No real footage involved. The people, places, and events in the video never happened.
Deepfake — face swap
A real video with a real person's face replaced by someone else's. Common in political misinformation, celebrity content, and non-consensual intimate imagery.
Deepfake — voice clone
A real video of a real person with their voice replaced by an AI-cloned version. Used to put words in someone's mouth they never said.
Manipulated real footage
Real video that has been selectively edited, mislabelled with a false date or location, or taken out of context. KweliAI detects AI generation — context manipulation requires separate verification.

KweliAI detects the first three categories — AI-generated video, deepfake face swaps, and AI-cloned audio. Context manipulation (real footage presented misleadingly) requires separate fact-checking with sources like Snopes, AFP Fact Check, or the BBC Verify team.

How to check if a video is AI-generated or a deepfake

If the video is posted on X, TikTok, YouTube, Instagram, Facebook, or Reddit, you can verify it in seconds without downloading anything.

Copy the post URL. Go to where you saw the video — X, TikTok, YouTube, etc. Open the specific post or video page. Copy the URL from your browser, or use the Share → Copy link option on mobile.
Paste it into KweliAI. Go to kweliai.com. Paste the URL into the scan box and press Scan. X and Reddit are free. TikTok, YouTube, Instagram, and Facebook require a Pro plan ($12/month).
Wait for the verdict. KweliAI downloads the video directly from the platform — you don't need to. It then runs three detection models simultaneously: AI video detection, deepfake face detection, and AI audio detection. Results typically arrive in under 10 seconds.
Read the clarity score. You receive a score from 0 (definitively real) to 100 (definitively AI-generated), a verdict label, and a confidence rating. High confidence verdicts are the most reliable.
"You can't download TikTok videos without a watermark and most YouTube videos can't be saved at all without third-party tools. KweliAI works directly from the post URL — you paste the link, we handle the rest."

Check the video now

Free for X and Reddit · Pro for TikTok, YouTube, Instagram, Facebook

Paste the link →

I can't download the video — can I still check it?

Yes. This is actually where KweliAI has a distinct advantage over most AI detection tools. Most competing tools require you to upload a video file. KweliAI works directly from the social media post URL.

This matters for several common situations:

TikTok videos — can only be downloaded with a watermark using the official app, and third-party download tools often violate TikTok's terms. KweliAI extracts the video server-side from the post URL without watermark complications.
X/Twitter videos — X removed the native video download option. KweliAI scans X videos directly from the post URL on the free plan.
Instagram Reels and posts — Instagram does not provide a download option. KweliAI can access and analyse Instagram media from the public post URL. Requires Pro plan.
YouTube videos — YouTube downloading requires third-party tools that exist in a legal grey area. KweliAI analyses YouTube content directly from the video URL without requiring a download. Requires Pro plan.

The only requirement is that the post is publicly visible. KweliAI cannot access private accounts, password-protected content, or posts that require logging in to see.

Visual signs a video might be fake — and why they're unreliable

There are visual signals that older deepfakes and AI videos used to exhibit. These are worth knowing, but they are no longer reliable for modern AI-generated content:

Unnatural blinking — early deepfakes blinked too rarely or too mechanically. Modern deepfakes trained on more video data have largely fixed this.
Blurring at facial edges — where the replaced face meets the original neck or hairline. High-quality deepfakes blend these transitions seamlessly.
Inconsistent lighting on the face — when the light direction on the swapped face doesn't match the scene. Better deepfake tools now correct for this automatically.
The "uncanny valley" feeling — a subtle wrongness that's hard to articulate. This is real and worth trusting as a signal, but it's not consistent enough to be a reliable detection method.
Audio-visual desync — lip movements that slightly lag behind speech. This persists in some deepfake voice clones but is harder to spot in compressed video.

None of these visual tells are reliable for 2025-era AI video tools. Sora, Runway Gen-3, and Kling have all demonstrated the ability to produce video that passes casual visual inspection by most people, including media professionals. Model-based detection is the only consistent method.

Checking political and celebrity deepfakes

Some of the most consequential fake videos circulating online involve politicians, public officials, and celebrities. These are used in election misinformation campaigns, financial fraud (fake endorsements and investment scams), and non-consensual intimate content.

If you see a video of a politician saying something shocking, a celebrity endorsing a product they've never mentioned, or a public figure appearing in a compromising situation, these are the highest-priority cases for verification.

Paste the post link from wherever you saw the video — X, YouTube, Facebook, TikTok — into KweliAI. The deepfake model analyses every frame of the video for AI manipulation signals. If the verdict is Deepfake or AI-generated with high confidence, do not share the video and report it to the platform.

"A video of a public figure saying something you've never heard them say before, posted by an account you don't recognise, going viral — this is the highest-risk scenario for a deepfake. Verify before sharing."

For political content during election periods, supplement KweliAI's technical verdict with source verification — check whether the original clip can be found on the official accounts of the person involved, whether established news organisations have covered it, and whether fact-checking organisations have reviewed it.

Questions about fake and AI-generated videos

Someone sent me a video of myself I didn't make — what is this?
This is likely a sextortion scam using a deepfake. Do not pay any money. Do not respond. Screenshot the message and report it to the platform and to your local police. In many countries, creating and distributing non-consensual intimate deepfakes is now a criminal offence. Contact the Cyber Civil Rights Initiative (cybercivilrights.org) for support specific to this situation.

How do I check a video I received in a WhatsApp or Telegram message?
KweliAI works from public social media post URLs. If the video was sent directly to you in a private message and has not been posted publicly, KweliAI cannot scan it in its current form. Ask the sender for the original public post link, or check if the video appears publicly somewhere using reverse video search tools like Google Video Search.

Can KweliAI check live video?
Not currently. KweliAI analyses static posts and uploaded video content. Live stream verification is a more complex technical challenge that current consumer-facing tools, including KweliAI, do not yet support.

The video is on X and I can't scan it — why?
Make sure you're pasting the direct link to the specific post, not to the user's profile page or to a search result. The URL should contain the post ID (a long number after /status/ for X posts). If the post has been deleted or the account is protected, KweliAI will not be able to access it.

Is KweliAI's video detection as accurate as image detection?
Video detection is technically more challenging than image detection because social media platforms compress and re-encode video, which can reduce the strength of AI generation artifacts. KweliAI's models are specifically trained on social-media-quality video for this reason. The confidence rating on each result reflects the certainty level — a medium-confidence result on video warrants scanning additional content from the same source.