Meta to expand AI image generation offerings for ads

Meta Platforms (META.O), the parent company of Facebook and Instagram, said that it was expanding its suite of generative AI ads products to offer tools that can automatically create variations of images and overlay text atop them.

The tool will launch in test form without the watermarks the social media company is applying to all images generated by its user-facing Meta AI assistant, which it has touted as a key safety feature, executives said at a press conference.

John Hegeman, Meta’s head of monetization, said the company was still working through how labelling would work for ads and would share guidelines by the time the tool rolls out globally, likely around the end of this year.

The announcement comes as Meta is pumping billions of dollars into building and supporting its generative AI models, while trying to convince advertisers they can get more bang for their buck by automating the work involved in creative aspects of their campaigns.

Fellow digital ads giant Google announced a similar expansion of AI ads tools in February. Google said ads created using its tools would be labelled using the SynthID watermarking technology developed by its AI research lab, DeepMind.

With Meta’s image generation tool, advertisers will be able to upload images of its products and generate other versions of those images, for example by adjusting the products’ orientation or showing people using them in different settings.

It is also expanding its text-generation offerings for headlines and key selling points, while adding the capability to overlay text onto generated images directly.

The company will add an option in the coming months for advertisers to input text prompts that can be used to tailor the image variations, it said.

As with earlier generative AI tools, advertisers running campaigns in regulated industries like politics will be barred from using the products.

Advertisers have rushed to embrace AI ads tools that automate the placement of their campaigns in front of different groups of users, but generally have been more hesitant about newer generative AI tools.

Some brands have expressed reservations about how tech companies will use any images they upload to improve the models, wary that their logos or other intellectual property could wind up in others’ generated images.