Brands facing scrutiny over usage of AI-generated faces in creatives

Delhi: Recently, an anonymous email landed in the inbox of BestMediaInfo.com. Using Yatra.com's full-page ad in TOI as an example, the email highlighted the perils of generative AI in the marketing industry.

In the creative used for the ad, the AI-generated model's face had a striking resemblance to a real model whose image is available on the stock photography website, Images Bazaar. 

According to Images Bazaar, the image in question was not sold to Yatra.com.

The brand asserted that they sourced the image legally from Adobe. 

The anonymous mail sender pointed out the danger of using AI-generated faces for advertising. He wrote, “It’s illegal. The model could file a criminal/civil lawsuit against the client/agency for using her face without her permission. The photographer could sue the client/agency for a violation of copyright. And the stock photography company, Image Bazaar, could also take legal action against the client/agency. Therefore, we must think twice before using an AI-generated image for our campaigns.” 

When BestMediaInfo.com sent a query to Yatra.com on the misuse of the model’s face in the ad without her or Image Bazaar’s permission, it shared a statement: "At Yatra, we prioritise ethical practices and ensure that all our creative materials are sourced responsibly and with proper licensing." As confirmed by our creative agency, the image used in our print advertisement was sourced from Adobe Stock by the creative agency under their licencing terms with Adobe Stock.”

It further wrote in the statement, “We, along with our creative/media agencies, procure images through reputed platforms and utilise them within the bounds of the respective licencing agreements. Further, we are not privy to the process followed by these reputed platforms for procuring these images."

The anonymous sender of the email further remarked, "Today, the model's image was used in a Yatra.com ad. Tomorrow, any person's face could be inserted into a condom advertisement without their consent, or even exploited in any other manner."

Currently, AI is not programmed to request permission before gathering data, often disregarding the privacy policies of users. Yatra.com legally obtained the image from Adobe Stock, which was uploaded by a digital content contributor under the pseudonym "Prasannapix" on the platform. This AI-generated image was created using an Image Bazaar picture as a reference.

Who bears responsibility in this situation? Is it Adobe's fault for failing to monitor copyright infringements by its digital content contributors? Or is it the brand for neglecting to verify whether the final creative violates copyright, or perhaps its agency? 

This underscores the broader challenges associated with generative AI in advertising and emphasises the need for immediate actions to prevent its misuse in the marketing and advertising industry.

The issue of intellectual property is currently under litigation in the US. Last year, a collective of artists filed lawsuits against various generative AI platforms, such as Midjourney and Stable Diffusion, alleging the unauthorised use of their artworks to train the algorithms of these platforms without consent or compensation.

When BestMediaInfo.com consulted experts to address the creative challenges posed by GenAI in advertising, Home Credit’s CMO Ashish Tiwari remarked that ultimately, the responsibility falls on the brand. "It was a negligent oversight on the brand's part. They should have conducted a basic check to ensure the creative didn't infringe any copyrights."

Dhruv Sachdeva, Founder, Humour Me, a Delhi-based agency, said, “If Adobe has it in their stock image bank and Yatra.com licenced it from Adobe, then this is an Adobe issue. It is clearly a problem between Image Bazaar and Yatra.com.”

However, Deepak Verma, Chief Operating Officer of ImagesBazaar Group, emphasised that Yatra.com cannot evade this responsibility. He stated, "Ultimately, the brand acquired the image from a stock image platform and utilised the individual's face without their permission."

Verma informed BestMediaInfo.com that to prevent such issues, Images Bazaar prohibits its contributors from uploading AI-generated images, especially those containing faces.

He added, "Adobe is the sole platform that accommodates GenAI images. Conversely, Shutterstock, Getty Images, iStock, and Pixel do not support AI-generated images. Until regulations on AI are established, it will persist in infringing upon individuals' personal rights."

Verma also informed BestMediaInfo.com that they have already notified the model whose face is featured in the Yatra.com ad and are also actively engaging with the Adobe contributor. They have observed similar issues on the contributor's other accounts as well.

Abhik Santara, CEO of Atom Network, said, “Incorporating faces generated by GenAI can create challenges for brands. Some AI tools might be trained on images without the proper permissions in place. Using those generated faces, even unintentionally, could land a brand in hot water. The emergence of AI Image Generators has created a new frontier for copyright law, where the traditional principles are continuously being tested. This is not a one-off problem and needs a rectification of IPR laws at a worldwide level for the creative industry.”

According to Varun Khiatani, Strategy at Talented, the crux of the issue lies in the unknown origins of the data used to train these models. He commented, “Given the vast array of facial data available online, sourced from social media and other platforms, there's a real risk of generating and using a face that belongs to an actual person, potentially without their consent. This blurs the lines between digital creativity and privacy invasion, placing brands in a precarious position of unknowingly infringing upon individual rights.”

Tiwari mentioned that at Home Credit, they refrain from utilising generative AI to create images with faces due to this very reason, stating that "generative AI cannot currently produce original human faces."

He also told BestMediaInfo.com that a lot depends on how one enables co-pilots to effectively perform their duties. “They must receive training and have real-time access to data that provides context and is trusted to be complete and accurate,” Tiwari said. 

Verma seconded Tiwari’s thoughts and said that any responsible brand won’t use generative AI images with faces until it gets direct rights from the person from whose face AI took reference. “No brand would like to get into litigation of any sort,” he added.

For Sachdeva of Humour Me, it’s more of a licencing issue. “The image is owned by the photographer. It’s the stock images platform’s problem to not license it for usage from the photographer who clicked the original image in the first place.” 

Verma also commented that Adobe has really big issues related to AI-generated images. “Even if one wants to get the image removed from the platform, it will take at least 90 days to happen,” he opined. 

In the past, Adobe found itself embroiled in controversy over AI-generated stock images depicting a Gaza explosion during the Israel-Hamas war, which were available on Adobe Stock. Multiple publishers continued to use these images without acknowledging their AI-generated nature. At that time, the tech company publicly stated that it clearly labels images created using GenAI. Additionally, several artists have previously complained about AI copyright infringement on Adobe Stock.

Emmanuel Upputuru, Founder and Creative Chairman, EFGH, a Gurugram-based creative agency, said, “The toothpaste is out of the tube. The use of AI will only increase exponentially. And so will the cases surrounding AI.”

Discussing such scenarios where generative AI generates faces resembling real individuals without their consent, Upputuru suggested that one crucial piece of evidence that might need to be presented is the prompt used for the AI. “That can reveal what the intent was—at least to an extent,” he said. 

Santara listed a few checkpoints that brands must adhere to when taking GenAI route in making creatives:

• Double-Checking Permissions: Even with a reputable vendor, it's wise not to take anything for granted. We should meticulously review model licenses to ensure they cover the specific ways the generated faces will be used.

• Human Expertise is Key: A rigorous human review process is essential. For style or prompt descriptions, we must use suggestions that are not general. Also, in the prompts, one must avoid mentioning already published works or characters that are protected by copyright. And not upload existing reference photos in our prompts.

• We should use AI to create works that are sufficiently different from existing works to avoid copyright infringement. This can be done by using different datasets, unique prompts, and parameters to create unique pieces.

• Another important hack should be to utilize tools for reverse image search to verify outcomes. e.g., Google Lens or Google Reverse Image.

• Get a human designer to alter AI-generated images before it used for commercial use.

• Lastly, always cite the AI model that is used.

Khiatani suggested two ways in which such a challenge can be curbed until we have regulation in place: “First, fine-tuning or training the GenAI models with their own dataset of facial data provides more control over the generated faces, ensuring a level of originality and compliance. This approach, which is surprisingly accessible, allows brands to tailor the AI's output to their specific needs while minimising legal risks. Secondly, a thorough manual review of every generated element is essential, despite being more rudimentary. This step is crucial not just for faces but for identifying potentially problematic or unsafe elements that could harm the brand's image.”

This isn't the first instance of brands encountering trouble for utilising GenAI-generated creatives featuring faces resembling individuals without their consent. For example, the ed-tech brand upGrad used the likeness of Google CEO Sundar Pichai in an advertisement. Similarly, Zomato, a food delivery platform, depicted footballer Lionel Messi, actor Leonardo DiCaprio, and Tesla CEO Elon Musk in one of its videos.

AI is like nuclear technology and has massive implications for human lives, emphasised Sachdeva. Therefore, the world has come together to figure out regulations surrounding it. He added, “Using somebody's face is the least of the problems because these things are just going to continue to happen more and more till there isn't enough regulation around this space.”

Last year, the Advertising Standards Council of India (ASCI) and Khaitan & Co., a full-service law firm, jointly released a whitepaper on generative artificial intelligence (AI), highlighting the opportunities, risks, and legal considerations around its use in advertising. 

Some of the best practices outlined in the whitepaper to mitigate risks associated with generative AI were: 

1. Review the AI platform’s terms of use and licensing compliance to reduce liability risks. 
2. Obtain the necessary authorisations and licenses for uploaded materials, including copyrighted and trademarked content. 
3. Avoid prohibited input prompts and carefully check the output for prohibited content before commercial use. 
4. Mitigate liability risks by implementing robust content review processes, establishing guidelines, and including AI disclaimers in marketing materials. 
5. Safeguard confidential information and ensure data privacy by enforcing nondisclosure agreements and implementing robust security measures. 
6. Upskill human labour in editorial oversight and compliance to avoid employee displacement.

After discussing the challenges of AI in marketing, Sachdeva believes that if AI is causing the problem, it can also offer solutions. He states, "AI will facilitate rapid recognition of global infringements of intellectual property." If there's any IP or trademark infringement, it will be promptly detected."

He also mentioned, "Blockchain presents another avenue to prevent infringements. Though it's not currently in place, it's inevitable. If there's an infringement overnight, anyone using it will be notified promptly, revealing the extent of the infringement upon their personal brands."