AI Deepfake Videos Can Target Anyone. Here's How to Detect, Trace & Stop One

Published by Ketan Ramshawala, CTO

AI deepfake videos no longer need a skilled editor. One clear photo of your face is enough. This guide breaks down a real case: how the fake was made, how it spread, how Strike traced it back to its source and how to detect and stop one aimed at you

How a Deepfake Video Circulates Before Anyone Notices

A deepfake video can sit online for months, gathering views in places you never visit, before it reaches anyone who knows you.

Recently, the AI-altered clip of actor Janhvi Kapoor that went viral in October 2026 shows exactly how that happens.

The fake was built from the official music video for "Chuttamalle", her song with Jr NTR from Devara. AI removed her clothing and left everything else real: her face, her co-star, the locations and the music. That realism is what made it believable at a glance.

The clip had been on an adult video site for four months before most people heard of it. Once it crossed into X, Telegram and private messages, it went viral within two days. By the time it was removed from X and a police case was opened in Hyderabad, copies had already been saved and forwarded.

Both actors spoke out and their words explain why this is abuse rather than gossip. Jr NTR wrote:

Source: Jr NTR | X Handle :https://x.com/tarak9999/status/2106292261843820830

Janhvi Kapoor's statement went further:

janhvi kapoor ig story

Why AI Deepfake Videos Will Keep Happening

AI deepfake videos will keep happening because they are cheap to make, easy to find and increasingly aimed at people who are not famous at all. Three recent studies show the scale:

8 million videos and counting : A Lancaster University study in the Journal of Online Trust and Safety estimates that sexual deepfake videos grew from 500,000 in 2023 to 8 million in 2025 and that 93% of the sites hosting them can be found through Google. Hindi film stars are among the most targeted, Forbes reported
Most targets are not celebrities : A June 2026 study from the CISPA Helmholtz Center for Information Security found that private individuals made up 55.8% of the people targeted in AI-made sexual imagery and 60.3% in videos. Earlier studies put that share below 5%.
Famous faces sell scams too : McAfee's 2025 research found that 90% of Indians have come across fake or AI-generated celebrity endorsements, with victims losing ₹34,500 on average

Janhvi Kapoor is far from the first public figure targeted. A deepfake of Rashmika Mandanna spread across India in 2023 and the main accused was later arrested. Explicit AI images of Taylor Swift flooded X in 2024 until the platform blocked searches for her name.

Fakes like these travel through the same four channels:

Adult video sites, where clips sit in themed folders for months with little notice.
Forums, which make and trade clips, stamp their watermark on them and return under new names when suspended.
Messaging apps, where full versions move privately out of reach of most reporting tools.
Search results, where a name can stay linked to a fake long after the original posts are gone.

Each channel feeds the reputation damage that follows. Fans and brands see the clip before they see the denial and every repost asks the person targeted to prove a negative.

What Other Threats Can AI Deepfakes Create?

Sexual fakes are the most visible use of deepfakes today, but they are not the only ones. The same tools that redrew a music video can copy your face and voice for almost any purpose.

These are the threats to prepare for

Live video call impersonation: A real-time deepfake of you joins a video call and asks a family member, manager or finance team to send money or share access
Voice clones: A short clip of interview audio can be enough to fake a phone call or voice note that sounds like you
Sextortion: Someone sends you a fabricated intimate video and demands money, threatening to share it with your contacts
Fake statements: A synthetic "leaked" apology, confession or rant surfaces at the worst possible moment, such as a film release, a brand deal or an election
Identity fraud: Your face and voice are used to pass video verification and open accounts or take loans in your name
Doubt about the real thing: Once fakes are common, genuine footage of you can be dismissed as AI, which makes it harder to prove what actually happened

Every one of these starts from photos, videos and audio of you that are already public. That is why finding misuse early matters as much as removing it.

How Strike Traced a Deepfake Video, Frame by Frame

Our deep fake detection software, Strike, found that the viral clip is the official music video with the actress's clothing removed by an AI video model while her face, her co-star and the soundtrack are real.

Every one of its 20 shots traces back to the original video and the clip was first uploaded four months before it went viral.

Here is how each question was answered, in plain terms.

The Case At A Glance
Posted : 30 May 2026, 11:41 IST, to an adult video-sharing site, by one account
Went viral : 2 to 3 October 2026, on X and Telegram
Source : All 20 shots located in the official T-Series Telugu music video
Verdict : Manipulated by a video-to-video AI model. Not a face swap. The face is genuine

Step 1: Where Did the Footage Come From?

From the official music video, published by T-Series Telugu in October 2024. Strike matched all 20 shots in the 49-second clip to it, the way you might match a torn photo to its negative by lining up the background, the pose and the light. In 16 shots, the original movement of hair, water and rock still shows through, because the real footage was driving the AI.

janhvi kapoor ai deepfake video analysis

Four of the matches. Left of each pair: the official video frame, shown whole. Right: the clip frame at the same moment, with opaque bars over intimate regions.

The audio confirmed it. The music lines up with the official song at a fixed point, 16.26 seconds in and never drifts. The first six seconds are also letterboxed and blurrier than the rest, a sign that two recordings of the same video were stitched together.

The two stitched segments: the letterboxed opening at 3 seconds (left) and full-frame footage at 30 seconds (right). Both trace to the same official video.

Step 2: What Kind of AI Made It?

A video-to-video AI model that anyone can download and run at home, paired with an add-on trained to remove clothing. The giveaway was on-screen text. A credit line and a YouTube Music badge came out as garbled letters, which only happens when a model redraws the entire frame, not just the body.

Enlarged four times: the on-screen credit at 13.5 seconds (top) and the YouTube Music badge at 16.5 seconds (bottom), redrawn by the AI as letter-like shapes.

Jewellery kept, clothing gone. Earrings, necklace, waist chain and sari pallu survived. In one shot the pallu hangs over a bare shoulder with nothing holding it up
Skin with too little detail. The torso carries 40% less fine texture than the face in the same frame
No flicker. The body moves in step with the real footage and hides correctly behind the co-star's hands, which frame-by-frame "undress" apps cannot manage

That rules out face-swap apps and makes mainstream video generators unlikely, since they block explicit output of real people. The exact model cannot be named from the file and Strike's report says so rather than guessing

Step 3: Is the Face Really Hers?

Yes. All 20 front-facing detections matched three licensed photographs of the actress, with no blending seam at the jaw or hairline.

Three Creative Commons photographs of the actress (left) and six face crops from the clip with their similarity scores.

This is the most important finding for deepfake detection. Most detectors look for tampering in the face. Because this face is real, a face-only detector would pass the clip as authentic. Proving it fake took source matching, identity checks and reading the marks the AI left across the rest of the frame.

Step 4: Who Posted It and When?

One account uploaded the clip to an adult video site at 11:41 IST on 30 May 2026, inside a folder of 27 AI-nudified clips of Indian actresses. The site's own search hides the page, so Strike reached it through the uploader's folder. It gathered about 16,000 views there before going viral.

Step 5: Will the Proof Stand Up?

Yes, because the file was handled as evidence from the moment it arrived

Fingerprinted on arrival with a SHA-256 hash, a digital fingerprint that changes if a single byte of the file changes
Never edited or uploaded. Nothing left the analysis machine and no online detector was used, since each upload would hand intimate imagery of a real person to another company
Logged and repeatable. A custody log records every step and scripts reproduce every number in the report
Redacted. Site names, web addresses and the uploader's account are held in a restricted pack for police and lawyers

How Can You Protect Yourself From AI Deepfake Videos?

No one can stop someone else from making a fake. What you can control is how fast you find it, how well you prove it and how quickly it comes down

Watch where your face appears. Run regular reverse image and face searches and set alerts on your name and handles
Don't trust face checks alone. Use deepfake detection that inspects the whole frame, since the newest fakes leave the face untouched.
Keep your originals. Your own copies of photos and videos make it far quicker to prove what was altered
Save evidence before you report. Capture links, screenshots and timestamps before anything is taken down
Report through official routes
1. India: platforms must act within 2 hours on complaints about intimate or morphed imagery under the IT Amendment Rules 2026. You can file at cybercrime.gov.in or call 1930
2. UAE: altering or sharing someone's image without consent is a crime under the UAE cybercrime law (Federal Decree-Law No. 34 of 2021), with prison terms and fines of up to AED 500,000. You can report through the Dubai Police eCrime platform (ecrime.ae), Abu Dhabi Police's Aman service (8002626) or the Ministry of Interior's My Safe Society app
3. US: the TAKE IT DOWN Act gives platforms 48 hours to act
Don't engage or reshare. Every share, even an angry one, spreads the clip further

Doing this by hand means checking dozens of sites and filing a different form for each one. That is the work Strike takes on.

Detect. Trace. Strike. How Strike Stops a Deepfake Before It Spreads

Strike by Synctric is an AI deepfake detection tool for images and video. It finds where your face appears online, proves what has been manipulated and files removal requests into each platform's own reporting forms, with a trained reviewer approving every case first.

1. Detect

Deepfake detection on every image and video

Two independent AI systems, one for forensic reasoning and one for scoring AI generation, return a single verdict (authentic, likely AI or AI-generated) with a confidence score

Continuous AI face monitoring

From one verified photo, Strike runs an hourly reverse image search sweep across FaceCheck.id and PimEyes and groups every confirmed match by the site hosting it.

Social listening on eight platforms

Google, Bing, X, Reddit, LinkedIn, YouTube, Facebook and Instagram are watched for your name and handles, with harassment, impersonation and non-consensual intimate imagery (NCII) flagged

2. Trace

When a fake surfaces, Strike traces where the footage came from, what made it and who posted it and builds an evidence pack for police and lawyers

3. Strike

Reviewed takedowns, filed for you

After a reviewer approves, Strike files platform forms, abuse notices and search requests in one pass, reaching official NCII forms rather than stopping at an abuse inbox

Remove photos from Google and Bing

Strike asks both search engines to de-index the offending site under their NCII policies; the search engines decide whether to act.

Re-upload protection via Hash Bank

Every removed image is fingerprinted and shared with StopNCII, whose participating platforms, including Meta, TikTok and Reddit, block matching uploads

Live case status

Platform replies are read and sorted automatically, with a follow-up after 48 hours of silence

You stay in control throughout- nothing is sent without human approval, consent is given per request and your evidence is visible only to you and Strike support.

“Protecting your reputation online starts with knowing where your video appears”

FAQs

Can someone make a deepfake video from your Reels or YouTube videos?

Yes. Public videos give AI tools exactly what they need: your face from many angles, your expressions, your movement and sometimes your voice. The more video of you that's online, the more convincing a fake can be. Keep track of where your videos are posted and keep your original files so you can prove what's real.

Should you respond publicly to a deepfake video of you?

A short, calm statement that the video is fake and is being removed usually works best. Don't describe the clip, link to it or name where it's posted, because that sends people looking for it. Let the statement close the conversation while removals happen in the background.

Can Strike check a video someone sends you?

Yes. Strike's deepfake detection checks both images and videos and gives you a clear verdict, authentic, likely AI or AI-generated, with a confidence score. It's a quick way to know what you're dealing with before you decide how to respond.

Can Strike help if a deepfake video of you has already gone viral?

Yes. Strike finds where the video is posted, traces where it came from and files removal requests with each platform once a reviewer approves the case. It asks search engines to drop the pages hosting it, blocks known copies from being uploaded again on participating platforms and keeps watching for new uploads, so you can focus on your response instead of chasing links.