How to detect Deepfakes: a complete guide to recognizing and protecting yourself against fake videos and audios

In recent years, we've seen how artificial intelligence has facilitated the creation of ultra-realistic content, better known as deepfakes . These manipulated videos, images, and audio recordings are posing a real challenge to how we identify the veracity of information online. It's becoming increasingly difficult to distinguish reality from fiction, and the ability to falsify faces, voices, or even entire situations has given rise to new forms of disinformation, digital fraud, and identity theft.

The rise of this technology, facilitated by the availability of increasingly sophisticated tools, compels everyone to exercise extreme caution before accepting any audiovisual or audio content as authentic. Learning to detect deepfakes is essential for protecting ourselves both personally and professionally, and also for curbing the spread of fake news, scams, and media manipulation.

What are deepfakes and how do they work?

The term deepfake comes from the combination of "deep learning" and "fake," and refers to content falsified using advanced deep learning or artificial intelligence techniques. In practice, it involves images, videos, or audio manipulated by algorithms capable of replicating human faces, voices, and expressions with extreme realism . The trick lies in the use of neural networks—such as GANs (Generative Adversarial Networks)—and the ability of these systems to analyze thousands of data points to then generate new, hyperrealistic images that perfectly mimic a real person.

While originally used for special effects in film, today there are publicly accessible applications that allow any user to play with this technology and fake videos, images and audio on a large scale, and this has multiplied the risk of fraud, disinformation campaigns and identity theft worldwide.

The impact and risks of deepfakes

According to recent data, the number of deepfakes on the internet is increasing exponentially every year . For example, the World Economic Forum estimated a 900% annual growth in this type of content by 2023. The danger of deepfakes is so real that, according to specialized reports, attacks involving impersonation and video manipulation increased by up to 450% in Latin America alone during 2023. Their use has already triggered problems in journalism, politics, finance, entertainment, and public safety.

The main risks include:

• Defamation or reputational damage to individuals and companies through credible setups.
• Use of fake videos for bypass biometric authentications and impersonate others.
• Realization of fraud on digital platforms and extortion.
• Growing spread of fake news and social and political manipulationwhich can destabilize markets and international relations.
• Risk of theft or transfer of money using fake audio and video, as has happened with romance scams and online fraud.

Identifying a deepfake requires skills that combine technical analysis and critical thinking, as the sophistication of these creations continues to improve.

Signs and tricks to detect deepfakes

Advances in deepfake technology are making these videos increasingly convincing, but even the best still show subtle flaws or anomalies that can help us uncover the deception. Here are the key signs to look out for:

1. Unnatural eyes and blinking

One of the most frequent errors is that the number of blinks or their rhythm is unnatural . Some sequences appear forced or show very few blinks, something humans usually do regularly. This detail, although it is being refined, remains an important clue.

2. Strange facial and body movements

Look for discrepancies between facial and body movements, or for stiff, unnatural expressions that are out of sync with the voice . If the face moves in a way that's not coordinated with the body, it could be a deepfake. There may also be lip-sync errors.

3. Inconsistent lighting and shadows

Deepfakes often feature fake, blurry, or unrealistic shadows . If the light hits the face unnaturally or doesn't affect the background and skin equally, be suspicious. The face's color may also differ from the neck or the rest of the body.

4. Blurred facial features, hair, and ears

The hairline and facial contours often show imperfections: uneven edges, blurriness, or fuzzy areas. Hair is especially difficult for algorithms to process, as are features like earrings, glasses, or the junction between the chin and neck.

5. Problems in the mouth and teeth

When the person speaks, blurred or unrealistic teeth may appear, or the mouth may have an odd shape or be completely out of sync with the audio . The technology still struggles to realistically represent the tongue, teeth, and oral cavity.

6. Audio and voice desynchronization

Manipulated audio , especially in voice deepfakes, can sound robotic, with artificial pauses or unnatural intonation. The coordination between lip movement and speech may be impaired , making it easier to detect.

7. Skin that is too smooth or has distortions

Some deepfakes show skin without pores, wrinkles, strange moles, or distorted skin when moving . Unnatural colors or excessive brightness in the eyes can also betray AI manipulation.

8. Duration and context of the videos

The most sophisticated fakes usually require many hours of work and therefore tend to be short . The longer the video, the more likely it is that errors will accumulate.

9. Consistency with public information

If a controversial video of a public figure appears only on an unreliable platform, or if no other media outlet picks up the information, be suspicious. Compare its content with other sources to see if it aligns with the person's biography, history, or stances.

10. Forced emotions and appeal to sentimentality

Much of the content that is heavily misrepresented aims to generate intense emotional reactions (anger, fear, surprise) to encourage virality. If you notice that a message seems designed to manipulate your emotions, be alert: misinformation spreads more effectively the stronger the emotional connection it creates.

Technological tools to identify deepfakes

While it's true that some clues are detectable by sight, for more delicate and professional uses, specialized tools are recommended . These tools leverage both artificial intelligence and data analysis, and are grouped into several categories:

1. AI-based tools for detecting deepfakes

There are solutions that use advanced machine learning to compare facial patterns, movements, or inconsistencies that are inaudible or invisible to the human eye. Here are some leading examples and their main functionalities:

• SentinelA cloud-based service that enables real-time detection by analyzing facial landmarks, blinks, and temporal consistency. It can analyze videos, images, and audio, generating a visual report of the affected areas.
• SensibilityA widely used platform for identity monitoring and protection in organizations, governments, and businesses. It analyzes large volumes of visual information and detects face swaps, manipulated audio, and proof of life. Especially useful in sectors with a high risk of identity theft.
• Oz LivenessCertified to the ISO 30107 standard, it acts as a defense against spoofing attacks through an SDK that is easy to integrate into mobile and web applications.
• WeVerifyAn ideal tool for those who validate viral content on social media, such as journalists, activists, or moderators. It uses blockchain to maintain a public database of known forgeries.
• HyperVergeAn advanced solution that combines biometric verification, facial recognition, and anti-fraud analysis across multiple international sectors. It has received global recognition and certifications.
• Intel's FakeCatcherReal-time detector with a 96% accuracy rate. It analyzes video looking for the subtle trace of blood flow in the pixels, impossible to perfectly simulate with current AI.
• Microsoft Video AI Authenticator: Free tool that gives a confidence score indicating the probability of manipulation in videos or images, detecting grayscale alterations undetectable to humans.
• Deepware: Allows you to analyze uploaded videos to check if they have been manipulated with AI techniques, in a simple and accessible way for all users.
• Phoneme-Viseme MismatchFocused on detecting discrepancies between the sounds heard (phonemes) and what is seen in lip movements (visemes). If there is an inconsistency, it is probably a deepfake.
• DuckDuckGooseThe focus is on protecting businesses and users from cyber, reputational, and financial threats. It provides clear explanations of manipulations and is especially valuable in corporate environments.

2. Metadata analysis tools

Analyzing a file's metadata (creation date, edit date, location, etc.) can help uncover clues about its authenticity. Tools like MediaInfo or ExifTool allow you to inspect these details hidden within multimedia files.

3. Tools that detect anomalies in the content

Other solutions such as FaceForensics or Audio Fingerprinting systems track strange patterns in facial movements, voice or elements of the video itself, looking for statistically improbable inconsistencies in genuine productions.

Practical tips and best practices to protect yourself from deepfakes

In addition to technological tools, combating deepfakes also requires human judgment and common sense . Here are some essential recommendations:

• Always verify the origin of the image, video or audio, especially if it comes from social networks or instant messaging.
• Distrust content designed to evoke strong emotions such as anger, fear, or surprise. The greater the emotional charge, the more likely it is an attempt at manipulation.
• Join the reverse image search using specialized search engines to check if the photo appears in other contexts or with other dates.
• Do not share videos or images if you have not been able to verify their authenticityespecially if they affect the reputation of a person, organization, or may have public consequences.
• Stay up to date on new trends and techniques in artificial intelligence and detection toolsTechnology is advancing rapidly, and the struggle is constant.

Future perspectives and the fight against deepfakes

The deepfake business is not going to lose momentum: it's projected to be a multi-billion dollar market, with increasingly sophisticated attacks. Legitimate technology developers, companies, and governments are taking action, collaborating to strengthen digital security and AI ethics.

There are promising projects and free resources available for those who wish to thoroughly analyze the veracity of multimedia content. While no solution is perfect, the combination of human analysis and technological tools will be key to meeting this challenge.

The increasing sophistication of deepfakes poses a constant challenge to the trust and integrity of online information. Detecting them requires a critical mindset, the use of detection technologies, and a commitment to not participating in the disinformation chain. The more aware we are of these risks, the less likely we are to fall for digital deceptions and the more effective our defense will be against the threats that are already present and will undoubtedly become even greater in the near future.

How to detect deepfake videos on the internet: methods, signs and tools