Gan.ai

gan_ai_logo

Creating Hyper-Realistic Deepfakes for HDFC Bank’s Viral Anti-Fraud Campaign

Deepfakes for Fraud Awareness banner

HDFC Bank’s “End of Scam Sale” campaign was a groundbreaking initiative that aimed to raise awareness about online fraud and the potential misuse of deepfake technology. The campaign revolved around creating a fictitious fashion brand called “Lulumelon” and using hyper-realistic deepfake videos of popular Bollywood actress and dancer Nora Fatehi to promote unbelievable discounts and too-good-to-be-true offers.

 

The campaign was meticulously orchestrated to initially deceive consumers by leveraging their optimism bias and fear of missing out on lucrative deals. However, as unsuspecting shoppers visited the “Lulumelon” website, they were met with a surprising twist – a reveal video that exposed the deception and educated them about the dangers of falling for such scams.

About Gan.ai

Gan.ai was founded in 2021 by Suvrat Bhooshan, to lead the field in AI-powered video personalization. Their unique generative AI models enable the personalization of video content according to individual viewer data and preferences.

 

Committed deeply to ethical and responsible AI, Gan.ai cooperates closely with its customers to ensure that any data used for a campaign is collected knowingly from the actor and is used solely for the extent of the campaign.

About HDFC Bank

HDFC Bank is one of the largest private sector banks in India, known for its innovative approach to banking and customer service. The bank has consistently prioritized customer education and awareness, particularly in the realm of financial literacy and fraud prevention.

 

With the “End of Scam Sale” campaign, HDFC Bank demonstrated its commitment to staying ahead of the curve by addressing the emerging threat of deep fake-enabled fraud. By leveraging cutting-edge technology and creative storytelling, the bank aimed to empower its customers with the knowledge and vigilance required to navigate the increasingly complex digital landscape.

About FCB Kinnect

FCB Kinnect is a leading integrated marketing agency that has consistently delivered innovative and impactful campaigns for its clients. As the creative force behind the “End of Scam Sale” campaign, FCB Kinnect played a crucial role in conceptualizing and executing the groundbreaking initiative.

 

The agency’s out-of-the-box thinking and willingness to push boundaries enabled the creation of a campaign that not only captured attention but also delivered a powerful message engagingly and memorably.

The Execution

HDFC Bank collaborated with their creative agency FCB Kinnect to conceptualize and execute the campaign. The team created multiple social media handles and website domains for the fictitious “Lulumelon” brand, mimicking the characteristics of fraudulent schemes to add authenticity to the deception.

Gan.ai's Role

Gan.ai played a crucial role in the campaign by creating hyper-realistic deepfake videos of Nora Fatehi. Our involvement began with an extensive informed consent phase involving all stakeholders, especially Nora Fatehi. This phase outlined the use of her deepfake and the data collection of images and videos which were curated and meticulously processed for the deepfake.

 

We used our proprietary face swap technology to train generative AI models. These models accurately mapped Fatehi’s facial features, expressions, and movements, allowing us to create a realistic deepfake of Nora’s face onto Vigil Aunty.

 

Gan.ai also replicated Nora Fatehi’s voice using advanced audio processing and synthesis techniques. We analyzed her vocal characteristics and trained dedicated audio models to produce synthetic speech that matched her voice. The synthetic audio was then seamlessly synchronized with the deepfake videos, resulting in a natural and believable outcome.

The Process

  1. Data Collection and Preparation: Gan.ai curated a dataset of Nora Fatehi’s images and videos, ensuring diversity in expressions, angles, lighting conditions, and backgrounds. The data was meticulously processed and annotated, labeling facial landmarks, expressions, and other relevant features.

  2. Model Training and Fine-Tuning: Using our proprietary face swap model on the prepared dataset to accurately map Fatehi’s facial features, expressions, and movements.

  3. Audio Cloning and Lip-Syncing: Gan.ai cloned Fatehi’s voice by analyzing her vocal characteristics and training dedicated audio models to generate synthetic speech. The synthetic audio was then synchronized with the deepfake videos using advanced lip-syncing algorithms.

  4. Post-Processing and Compositing: Gan.ai‘s team meticulously refined and enhanced the deepfake videos through post-processing techniques, including adjusting lighting, color grading, and compositing the footage into desired backgrounds and scenarios.

The Result

The final, hyper-realistic deepfake videos of Nora Fatehi were seamlessly integrated into the “Lulumelon” brand’s promotional materials, including social media posts, website content, and the eventual reveal video. The convincing nature of the deepfakes initially sparked confusion and curiosity among viewers, effectively setting the stage for the campaign’s powerful message about the dangers of online fraud and the importance of staying vigilant.

 

The campaign went viral, generating widespread media coverage and discussions around the ethical implications of deepfake technology. It not only raised awareness about the potential misuse of AI in enabling scams but also highlighted the need for society to stay informed and proactive in combating such threats.

 

Gan.ai‘s cutting-edge deepfake technology was instrumental in bringing the campaign’s vision to life, showcasing the potential of AI while also emphasizing the need for responsible development and deployment of these powerful technologies.

By Tejas Shahasane

I'm a content marketer fueled by a fascination with AI and the opportunities it unlocks. You can catch me saying "lights out and away we go!" almost every weekend.

Leave a comment

Your email address will not be published. Required fields are marked *