Meta tests generative AI ad tools

Meta has entered the generative AI battleground, announcing on Wednesday that it has begun testing AI-enabled text and image editing tools within its advertising toolkit.

The tech company said it has been working with a small group of advertisers over the past few months to trial tools that automate elements of advertising creation.

The tools, which Meta is incubating within an AI Sandbox, include a copy generator that creates multiple captions from one product description, an image background generator and “outcropping” editor that automatically fills in gaps when a creative asset is resized.

At a media event on Thursday (May 11), Meta said it was focused on developing capabilities “most helpful for businesses to achieve their goals,” but acknowledged that it is entering a fiercely competitive industry.

“Clearly, the overall generative AI space is developing incredibly quickly right now,” John Hegeman, VP of monetization said at the event. “We recognize that there are a number of tools that businesses are already using out there and will continue to use over time as well.”

Meta believes its “unique opportunity” is to integrate AI-enabled features into its products to allow advertisers to test iterations on the fly.

“[This] will allow advertisers to do a number of powerful things such as test tens of different types of backgrounds or text variations, which could allow them to further personalize their messages for different parts of their audience,” Hegeman said.

Meta stated that it will continue to test its trio of tools with a limited group of advertisers in order to “quickly gather feedback” before opening it up to a wider cohort in July. 

While AI is a useful tool to generate high volumes of creative assets, Meta said it wants to avoid flooding its platform with low-quality creative.

“That, too, is also why we're taking this iterative testing approach to make sure that the additional versions that we are able to generate with these tools are high-quality, perform well with certain audiences and are actually contributing to that overall performance,” Hegeman said.

Meta hopes to eventually add AI-generated video to its toolkit, but this will “take a bit longer to incorporate” due to the complexity of generating multiple frames, Hegeman said. Developers and designers have experimented with AI-generated commercials — and the results are rather frightening.

Generative AI has quickly become one of the most crowded technology industries following the explosive growth of OpenAI’s ChatGPT toward the end of last year.

Google released a slew of AI products during its I/O developer on Wednesday, including an image editor that boasts a sophisticated range of features, such as erasing objects and repositioning subjects within a frame. The search giant also announced a partnership with Adobe Firefly to power AI image creation within the Bard chatbot.

Amidst the proliferation of products, there are growing concerns about the ethics of AI and the adverse impacts it could have on businesses and society.

Creative industries are debating the legalities of using copyright-protected images and videos to train AI models. Meanwhile, as it becomes increasingly difficult to distinguish between real and manipulated media, there are growing calls for AI-generated creative to be clearly disclosed.

When prodded about these issues during the Thursday event, Meta’s responses were vague. 

Meta’s VP global business group Nicola Mendelsohn said it was “too early to say” when asked whether the company will add disclosures to AI-generated text or backgrounds.

Hegeman said Meta trained tools within its AI Sandbox on a “variety of sources” including “data from our products, public data, data that we license.”

Google more directly addressed one of the hot topics on Wednesday, when CEO Sundar Pichai pledged to embed techniques such as watermarking and metadata disclosures within its AI models to help users identify synthetically generated content.

Beyond image and text generators, Meta is also developing an AI agent that can interact with users in a more human-like manner than existing bots, which Hegeman said could be used to drive efficiencies within business messaging.

“We know there's a ton of interest in the potential to utilize AI agents for business messaging, for example, to help businesses with lead generation, sales or customer support. We suspect that, over time, tools like AI agents will be able to help more businesses use products like ‘click-to-message’ ads more effectively and with less cost,” he said.

“An example of how that can work is maybe a business can use those types of tools to help localize a chatbot for a new market with significantly less time and effort than it might have taken them otherwise,” he added.

Meta CEO Mark Zuckerberg told investors during the company’s Q1 earnings call last month he saw “an opportunity to introduce AI agents to billions of people in ways that will be useful and meaningful.”