In 2026, AI can generate as much as two minutes of steady, high-definition video from a single textual content immediate. Instruments like OpenAI’s Sora 2, Runway’s Gen-4, and Kuaishou’s Kling 3.0 not simply copy visible patterns. They now simulate gravity, fluid dynamics, and the way mild and shadow work together in the actual world.
The hole between captured footage and algorithmically generated video is closing quick.
In Nigeria and throughout Africa, this has created a significant issue. Throughout the 2023 and 2024 election cycles, artificial audio and video clips have been used to take advantage of ethnic and regional tensions and manipulate voter sentiment.
Simply earlier than voting started in 2023, a deepfake audio clip of presidential candidate Atiku Abubakar discussing election rigging circulated on-line, in keeping with The Cable, a Nigerian on-line newspaper, timed to land when emotional stress was at its peak and significant considering at its lowest.
The rip-off panorama has shifted, too. The outdated “Nigerian Prince” e mail trick has been changed by high-fidelity deepfake movies that includes recognisable faces. Enterprise government Ibukun Awosika, well being influencer Aproko Physician, and WTO Director-Common Ngozi Okonjo-Iweala have all had their likenesses used to advertise fraudulent funding schemes and pretend medical cures.
The Nation, a Nigerian newspaper, reported that in 2025, singer Ayra Starr and actress Kehinde Bankole have been targets of coordinated AI-generated harassment utilizing fabricated footage.
The instruments to create these fakes are actually accessible to anybody. Meaning the accountability of recognizing them falls on you. This information breaks down precisely what to search for.
Visible indicators to search for within the video
Regardless of how superior fashionable AI has turn into, producing video stays computationally costly, and refined failures are frequent. The trick is understanding what to search for.
1. Watch the eyes
The human eye is likely one of the hardest issues for AI to simulate. In an actual particular person, blinking occurs each 2 to 10 seconds and is all the time accompanied by small muscle actions across the eye and a short refocus of the iris because it reopens.
AI-generated faces typically blink in robotic patterns, both too rhythmically or far too not often, generally solely as soon as each 30 to 60 seconds. Past blinking, the eyes are inclined to lack “saccadic” motion, the tiny, fast shifts the eyes make when scanning a room or following a dialog. The result’s a hard and fast, lifeless gaze that feels disconnected from regardless of the particular person is supposedly saying or reacting to.
A deepfake video of Ngozi Okonjo-Iweala was flagged by analysts, partly as a result of her gaze remained unnaturally regular all through, with out adjusting to the lighting within the supposed interview room.
2. Test the pores and skin
Present AI fashions like Sora 2 can render 4K video, however they typically oversmooth pores and skin. The result’s a face that appears waxy, overly polished, or oddly clear. Human pores and skin has pores, tremendous hairs, variations in pigmentation, and a slight, rhythmic color change because of blood flowing via capillaries. These particulars are costly to render, and AI tends to skip them.
Pay shut consideration when the particular person smiles or speaks. On an actual face, the pores and skin across the eyes and mouth wrinkles and strikes. On an AI face, it typically stays suspiciously nonetheless. The absence of these small crinkles across the eyes throughout a real smile (generally known as Duchenne markers) is a powerful indicator that the face is artificial.
In deepfakes of Ibukun Awosika, the pores and skin round her eyes remained static even whereas she was supposedly delivering an lively promotional speech. That degree of stillness is physiologically unattainable.
3. Watch the mouth and lip sync
Talking entails an intricate coordination of the jaw, tongue, and lips. AI approximates this fairly than really simulating it, and the result’s a mouth that appears like it’s sliding throughout the face fairly than being hooked up to a jaw and cranium.
Lip sync breaks down most clearly on plosive feels like “p,” “b,” and “m,” which require the lips to completely press collectively earlier than releasing. There’s typically a slight delay between the audio and the visible. Additionally, take a look at the inside of the mouth. In lots of AI movies, enamel seem as a single white block fairly than particular person constructions with pure gaps and imperfections.
A helpful take a look at is to mute the audio and watch the mouth actions alone. If the jaw motion seems rhythmic however doesn’t match the muscular effort you’d anticipate from the phrases being spoken, the video is probably going pretend.
4. Take a look at the background
AI turbines usually focus their processing energy on the principle topic and let the background slide. Excessive-quality fakes can look convincing at first look, however nearer inspection typically reveals issues.
Search for individuals within the background with lacking fingers or limbs, furnishings that seems warped or distorted, and textual content on indicators or screens that’s gibberish or retains shifting between frames. Within the “Baphomet Ebook Membership” AI photographs that circulated in South Africa, the youngsters appeared convincing, however the bookshelves behind them contained books that appeared to soften into each other.
Additionally, watch how water, hearth, or smoke behaves within the background. Fluid dynamics are notoriously tough for AI. Liquids typically transfer in slow-motion or lack the chaotic, random splash patterns of precise physics.
5. Check head rotation and hand occlusion
Most deepfake fashions are educated on front-facing photographs. When an artificial face turns towards a full facet profile, the rendering typically breaks down. Widespread artefacts embody the ear blurring or shifting place, the jawline separating from the neck, or glasses showing to soften into the pores and skin.
AI additionally struggles with occlusion, which happens when one object passes in entrance of one other. If an individual waves their hand throughout their face in a deepfake video, the face will typically flicker or momentarily lose its alignment, briefly revealing the underlying supply picture.
This has turn into a normal verification step for security-aware organisations: ask the caller to slowly wave their hand throughout their face and watch what occurs to the picture beneath.
6. Test shadows and perspective
AI understands the world as a set of pixel relationships, not as a three-dimensional house ruled by geometry and physics. This results in particular failures you’ll be able to take a look at for.
In any actual video, all parallel strains in a scene (rooflines, desk edges, hallway partitions) should converge at a single vanishing level on the horizon. AI fashions typically get this flawed, creating a number of conflicting vanishing factors in a single body.
Shadows are one other giveaway. Search for topics casting shadows in a single course whereas a close-by object casts its shadow in a unique course. In a deepfake video of South African President Cyril Ramaphosa, the lighting on his face stayed constant whilst he supposedly moved in entrance of a window, which isn’t how mild works.
Your eyes can solely catch a lot. These instruments go deeper, analysing pixel knowledge, organic indicators, and file metadata to calculate the probability {that a} video is artificial.
| Device | What it detects | Finest for | Key limitation |
| Actuality Defender | Multi-model scoring throughout audio, video, picture, and textual content | Actual-time screening of uploads and stay calls | Enterprise pricing, not constructed for particular person shoppers |
| Intel FakeCatcher | Organic blood-flow indicators within the pores and skin (rPPG) | Excessive-assurance verification of human topics | Requires high-quality, uncompressed video |
| Microsoft Video Authenticator | Pixel-level artifacts, frame-by-frame evaluation | Newsroom triage and investigative work | Much less efficient on closely compressed clips |
| Deepware Scanner | Temporal inconsistencies between frames, face-swaps | Catching AI faces layered over actual actors | Much less efficient on full AI era vs swaps |
| InVID WeVerify | Video fragmentation for keyframe reverse looking, recognized fakes database | Journalists and fact-checkers | Depends on present debunked content material in its database |
| UncovAI | C2PA metadata, GAN signatures, WhatsApp verification bot | Common shoppers, particularly in Africa | Belief Rating is probabilistic, not definitive |
| CloudSEK | Risk intelligence and infrastructure mapping | Monitoring the origin of coordinated pretend campaigns | Platform-based, not a client device |
| Pindrop Pulse | Acoustic and behavioural voice evaluation | Name centres and banking fraud prevention | Voice-focused, not video-native |
The heart beat take a look at: Why biology beats pixels
Some of the dependable detection strategies in 2026 is analysing physiological indicators that AI merely doesn’t simulate. As a human coronary heart beats, it causes refined rhythmic modifications in pores and skin color as blood strikes via the capillaries. Programs like Intel’s FakeCatcher use distant photoplethysmography (rPPG) to detect these indicators.
As a result of generative AI creates pixels based mostly on visible patterns fairly than organic programs, an artificial face will lack this pulse. Even Sora 2 at 4K decision fails this take a look at as a result of it’s predicting the subsequent almost definitely association of colored pixels, not rendering a residing human being.
The limitation is important, although: this sign disappears in low-resolution or closely compressed video. It is not going to work on a WhatsApp clip that has been forwarded a number of occasions.
The C2PA normal: Proving actuality as an alternative of detecting fakes
The business is shifting from detection towards provenance, which suggests proving {that a} video is actual fairly than figuring out whether or not it’s pretend. The C2PA normal, adopted by Adobe, Sony, and Leica, embeds a cryptographic signature right into a file in the meanwhile of seize. Each edit made to the video after that time is recorded in a tamper-evident path.
The issue is that platforms like X (previously Twitter) typically strip metadata to avoid wasting bandwidth. This removes the C2PA manifest completely, leaving the video unverified even when it was initially genuine. This creates what is named a “provenance hole,” which scammers actively exploit by sharing movies on platforms that strip metadata.
What to do if you find yourself uncertain
Instruments assist, however an important factor is your personal verification course of. Organisations like Africa Test and Dubawa emphasise that recognizing a pretend is a means of connecting dots, not a single click on.
Step 1: Test your emotional response first
AI-generated misinformation is constructed to set off sturdy feelings. If a video makes you’re feeling sudden anger towards a political opponent, pleasure about an funding alternative, or concern about one thing taking place to somebody you respect, deal with that response as a warning signal fairly than affirmation.
Deepfake content material is often timed to maximise emotional influence and minimise vital considering. The pretend Atiku audio within the 2023 Nigerian election dropped simply hours earlier than polling started for precisely this cause.
Step 2: Examine the supply and path
Ask the place the video truly got here from. Test whether or not the account posting has a verifiable historical past and an actual identification. A real video of a major occasion (a presidential deal with, a significant accident, a celeb look) will all the time generate secondary protection from impartial sources.
If a video claims to point out one thing world-altering and no credible outlet is reporting it, that alone is a powerful indicator of fabrication.
Step 3: Test the file metadata
Use a device like ExifTool or the “Get Information” perform in your Mac or PC to look at the file’s hidden info. AI-generated movies typically include timestamps or software program tags that contradict the story of when and the place the video was supposedly shot.
A video claiming to be stay footage from Lagos however carrying a creation date from three days earlier is an instantaneous purple flag.
Step 4: Run a keyframe reverse search
As a result of movies are made up of particular person frames, you’ll be able to deal with a single body as {a photograph}. Utilizing the InVID WeVerify extension, extract a body and run it via Google Pictures or TinEye.
This typically reveals that the video is definitely older footage from a unique nation or context that has been altered. A extensively circulated shallow pretend that includes Bola Tinubu was debunked precisely this fashion, by discovering the unique, unedited footage of the identical occasion on YouTube.
Step 5: Confirm via a second channel
For something involving monetary directions or requests from an individual in authority, all the time verify via a separate, trusted channel. If a video of a CFO orders a wire switch, name the CFO’s recognized workplace quantity straight or ship a message via a verified platform to substantiate.
This out-of-band verification is at the moment the one foolproof defence in opposition to real-time deepfake impersonation.
Elevate your perspective with NextTech Information, the place innovation meets perception.
Uncover the most recent breakthroughs, get unique updates, and join with a world community of future-focused thinkers.
Unlock tomorrow’s tendencies at this time: learn extra, subscribe to our publication, and turn into a part of the NextTech group at NextTech-news.com
