After initially disabling the capability, OpenAI today announced that customers with access to DALL-E 2 will be able to upload people’s faces for editing using its AI-powered image generation system. Previously, OpenAI only allowed users to work with and share photo-realistic faces, and prohibited uploading any photo that could depict a real person, including photos of prominent celebrities and public figures.
OpenAI claims improvements to its security system have enabled the face-editing feature by “minimizing the potential for harm” from deepfakes and attempts to create sexual, political and violent content. In an email to customers, the company wrote:
“Many of you have told us that you miss DALL-E coming up with your own outfits and hairstyles and editing the backgrounds of family photos. A reconstructive surgeon told us that he had used DALL-E to help his patients visualize the results. And filmmakers have told us they want to be able to edit images of scenes with people to speed up their creative processes… [We] developed new detection and response techniques to stop abuse.”
The change in policy is not necessarily opening the floodgates. OpenAI’s terms of service will continue to prohibit uploading photos of people without their consent or images that users have no rights to, although it’s not clear how consistent the company has been in enforcing those policies in the past.
In any case, it will be a real test of OpenAI’s filtering technology, which some customers have complained in the past of being overzealous and somewhat inaccurate. Deepfakes come in many flavors, from fake vacation photos to presidents of war-torn countries. Accounting for any emerging abuse will be a never-ending battle, in some cases involving very high stakes.
Undoubtedly, OpenAI — which is backed by Microsoft and leading VC firms, including Khosla Ventures — wants to avoid the controversy associated with Stability AI’s Stable Diffusion, an image-generating system available in an open source format without any restrictions. As TBEN recently wrote, it wasn’t long before Stable Diffusion — which can also edit facial images — was being used by some to create pornographic, non-consensual deepfakes of celebrities like Emma Watson.
So far, OpenAI has positioned itself as a brand-friendly, buttoned-up alternative to the rampant stability AI. And with the restrictions surrounding the new face editing feature for DALL-E 2, the company is maintaining the status quo.
DALL-E 2 will remain in the invite-only beta. In late August, OpenAI announced that more than a million people are using the service.