On May 8th, Meta, the parent company of Facebook and Instagram, announced that it is expanding its generative AI advertising products with new tools that can automatically create diverse images and overlay text on them.
During a press conference, company executives stated that this new tool will be introduced in a testing phase without watermarks, unlike the images generated by Meta's AI assistant for regular users. Meta had previously emphasized that adding watermarks is a crucial security feature.
John Hegeman, Meta's Head of Commercial Business, mentioned that the company is also exploring ways to label the generated ads. Guidelines on this labeling process will be announced globally when the new tool is rolled out, expected around the end of this year.
Meta is currently investing billions of dollars in developing and supporting generative AI models, aiming to persuade advertisers that by automating the creative aspect of ad campaigns, they can achieve greater returns.
In February of this year, Google, a digital advertising giant, unveiled a similar plan for AI ad tools. Google stated that ads created using their AI tools will be marked with the SynthID watermark technology developed by their AI research lab, DeepMind.
With Meta's image generation tool, advertisers can upload product photos and create various related images. For example, they can adjust the placement of products or demonstrate how they can be used in different settings.
The new tool also includes text generation capabilities for image titles and key product selling points, allowing text to be directly added to the automatically generated images.
Meta announced that in the coming months, they will introduce a feature that enables advertisers to customize various images by inputting text prompts.
Advertisers are increasingly embracing AI ad tools that can automatically tailor content for different audience segments. However, there is more hesitation towards newer generative AI tools. Some brands are cautious about how tech companies use their uploaded images to improve models, fearing that their brand logos or other intellectual property may appear in images generated by others.
