When generative AI tools evolved to the point where they could create convincing deepfakes – synthetic images, video and audio designed to mislead users – businesses with an eye on tech evolution quickly recognized the threat they posed.
By now, businesses across all industries in digital understand deepfakes’ risk to their reputation, and to their ability to retain customers and attract the best partners. And as AI advances, the challenges of detecting and sounding the alarm on deepfakes are only increasing.
Deepfake detection calls for an AI-powered approach. At the same time, an educated, informed and engaged employee base should learn about the telltale signs of AI-manipulated media, and augment deepfake detection tech with human observational skills.
Here are five ways to identify deepfakes – a crucial step in protecting your brand, preserving trust across your market, and bolstering your overall bottom line.
1. Look for Inconsistent or Unnatural Physical Features
Generative AI can produce strikingly realistic images and video, but it still has difficulty consistently producing nuanced details. Close inspection can reveal elements that humans can recognize as inauthentic. For example:
- Unnatural eye movement. AI often produces images and videos of eyes that don’t seem to blink enough, or at a “normal” cadence. Sometimes the eyes seem to track or focus in unnatural ways.
- Asymmetrical facial features. Look at details like ears, jawlines and teeth. AI can struggle with symmetry and facial proportions.
- Unnatural hands. Infamously, generative AI has a tendency to produce images of hands and other extremities with unrealistic proportions or angles, or too many or too few fingers.
Zooming in on images and video, pausing video to look closely at stills, and conducting frame-by-frame reviews can help identify these subtle, yet common indicators of a deepfake.
2. Listen for Audio Anomalies
Despite AI advancements in creating and manipulating audio, deepfake tech still has shortcomings. Some common signs of voice cloning in deepfake video and audio include:
- Poor audio/mouth syncs. Audio and mouth or facial movement lag or seem to clash. Pronunciation and articulation feel unnatural.
- Robotic tone. AI-generated voices may sometimes seem flat and mechanical. The pace of speaking may feel stilted, and the tone may sound inexpressive and unemotional.
- Audio artifacts. Does the audio sound consistent – or do you hear warbles, sudden silences, static, or inconsistent background sounds? These are all signs that the audio may be manipulated and edited together.
For detecting a deepfake, it’s important to review the audio track and visual syncs. This is especially critical when the images, video and audio address sensitive themes.
3. Inspect Backgrounds and Understand Context
Deepfake tech has a tendency to produce blurred or unnatural backgrounds. Here are some common effects of AI-generated backgrounds:
- Flickering or “melting.” Between the subject of the video or image and their surroundings, the edges and borders may seem to blur or produce ripple effects.
- Inconsistent lighting. The lighting and shadows may not align with the subject, their movements and their background. Lighting may seem to change abruptly.
- Contextual clues. A deepfaked subject may appear in an unexpected or uncharacteristic setting or location.
In addition to closely inspecting these elements, teams also need to verify whether previously published materials show the subject in the setting where they’re depicted.
4. Compare and Crowdsource for Credibility
Every business with strong partnerships in the marketplace and loyal customers or audiences has the opportunity to leverage those human connections. Put these advantages to work in these ways:
- Comparing sources. Inspect questionable video, images and audio, matching it against archival footage, earlier interviews and the subject’s prior public statements.
- Crowdsource insights (and eyeballs). Engage your business network and communities – including, when relevant, your internal teams, clients, partners or audiences – to spot inconsistencies and abnormalities.
- Check timelines and context. Verify the location and timing of the event, cross-checking against confirmed dates, times and actions to flag unrealistic, impossible or factually wrong contexts.
Deepfake detection technology and manual inspection, taken together, provide the most comprehensive protection.
5. Automate the Deepfake Detection Process
Businesses should always provide a manual review process, but automated AI detection methods will provide valuable efficiency and accuracy. These tools include:
- Metadata analysis. Businesses can get assistance from tools that scan content metadata for any suspicious patterns in edit history, file origins, media creation dates and more.
- Authenticity verification tools. Using AI tools to detect AI content drastically increases a business’s ability to locate and flag data inconsistencies, digital artifacts that evade the human eye, and other indicators of deepfakes.
- Reverse image and video search. A reverse search may turn up images and videos in different settings or contexts than the deepfaked content depicts. It can also help detect face-swapped media – a common tactic for deepfakers in fast-moving media cycles and viral moments.
Businesses should look to tech solutions their teams can use and incorporate in their workflow to detect deepfakes.
Copyleaks Helps Businesses Defeat Deepfakes
Among deepfake detection solutions, Copyleaks’ AI Deepfake Detector stands out in its ability to efficiently analyze images, video and audio. Using innovative machine learning models, Copyleaks identifies signs of AI manipulation – far beyond the common telltale signs. Continuous development and innovation enables Copyleaks to understand and keep pace with the evolution of deepfake technology and tactics.
Taking advantage of deepfake detection technology protects the integrity of a business’s communications and credibility in the marketplace, and reduces financial and legal risk brought about by deepfake and cheapfake attacks. In an environment where authenticity is not a given, ensuring authenticity is a competitive advantage.
