DALL-E image generation AI can now edit real faces


Mathilde Rochefort

September 21, 2022 at 12:40 p.m.

3

Faces generated by DALL-E © © OpenAI

Faces generated by DALL-E © OpenAI

DALL-E, OpenAI’s artificial intelligence capable of generating images from text, can now edit visuals containing human faces. Until then, this feature was prohibited for fear of malicious use.

AI users only had the ability to work and share photorealistic faces. They could not upload any photo that might represent a real person, including photos of celebrities and public figures.

DALL-E has improved in security

In an email sent to its users, the company explained that it had made great progress in terms of security, now allowing it to accept working with the faces of real people. In particular, it has improved its filters to remove images that contain ” sexual, political and violent content “.

DALL-E users will thus be able to upload a person’s photo and generate variants of this image, or modify specific characteristics of the latter, such as a person’s clothing or hairstyle. This should make many people in the creative industry happy. Deploying this feature will, however, put a strain on OpenAI’s filtering, which some customers have complained about in the past because of its inaccuracy.

Art-generating AIs are controversial

This decision shows the way in which the manufacturers of image generators and their users try to manage the potential drawbacks of this technology, which can very quickly be exploited by malicious people through the creation of deepfakes notably.

For example, the Stable Diffusion platform, available in open-source without any restrictions, was used to create deepfakes non-consensual pornography from celebrities like Emma Watson. Stable Diffusion, like the new feature in OpenAI, allows editing of human faces. However, OpenAI is much more picky about the malicious uses of its technologies, especially since it is a company supported by very big names in the tech sector such as Microsoft, who absolutely do not want to be associated with such dark matters.

In addition to the dangers they represent, art-generating AIs are targeted by artists who consider that these technologies, capable of creating a work in an instant from a few words, do art and their work a disservice.

Sources: The Verge, TechCrunch



Source link -99