Images created by AI abound, but identifying them will soon be easier


With the proliferation of deepfakes and generative AI, it is increasingly difficult to separate real images from false information. Qualcomm has partnered with Truepic to implement C2PA certification within the Snapdragon ecosystem. This certification makes it possible to identify the original photos and to inform of the modifications made to them, in particular with generative AI.

qualcomm truepic c2pa

False information abounds on the Internet and social networks. Their resurgence has a considerable impact on reader confidence. Especially since it is increasingly easier to create and modify an image thanks to generative artificial intelligence. Whether it be Midjourney, Slab or Stable Diffusion, many models allow you to create a truer-than-life image from simple written parameters. Photoshop even goes so far as to integrate generative AI to democratize image creation.

Read also – OpenAI creates a tool for detecting images generated by its AI, not enough to make deepfakes disappear

The risk of being invaded by non-legitimate images, generated by artificial intelligence, is all the greater as these models arrive in smartphones. We first noticed this with the Pixel 8 and Pixel 8 Pro, which allow the user to create a personalized wallpaper using keywords. And we will see it in the coming weeks or months thanks to smartphones equipped with Qualcomm’s Snapdragon 8 Gen 3, an SoC compatible with many models, including generative AI specialized in imaging like Stable Diffusion.

qualcomm truepic c2paqualcomm truepic c2pa
Stable Diffusion runs on smartphones running Snapdragon 8 Gen 3

Image generative AI is coming to smartphones

A smartphone equipped with a Snapdragon 8 Gen 3 and Stable Diffusion is capable of creating an image from keywords in less than a second. Aware that this poses an ethical problem, Qualcomm announced a partnership with Truepic, a company specializing in “digital imaging security”. This means that Truepic technology makes it possible to certify that an image is legitimate and that another has been modified, whether manually, with an image editor, or automatically, using artificial intelligence.

qualcomm truepic c2paqualcomm truepic c2pa

Truepic’s technology is based on a standardized certificate created by C2PA (Coalition for Content Protection and Authenticity). In this grouping, you find in particular Microsoft, Adobe, Intel, arm, as well as a large part of the players linked to photography, music and video. The idea of ​​C2PA is to create a certificate that adds new metadata to content, including additional EXIF ​​for images (created or generated) and photos (original or modified).

C2PA certificate generates authenticity and modification data

In these EXIF, you find classic information (camera, location, date and time, original resolution). But you will find other less classic ones: the name of the creator, as well as all the modifications that have been made to it, including those with artificial intelligence. If a filter has been applied, this will be indicated. If a zoom-out was used, this will also be indicated. The goal is simple: when you open a C2PA certified photo, you can consult its authenticityits origin and the modifications that have been made to it, by clicking on an icon appearing on the screen.

qualcomm truepic c2paqualcomm truepic c2pa

Qualcomm’s goal is to participate in the deployment of this standardized certificate which will improve trust between those who produce content, whatever its nature, and those who consume this content. Is this enough today? No. Because the compatible ecosystem is limited. Indeed, the number of compatible software and operating systems does not cover all content creators and consumers. Result: as soon as you leave the ecosystem, the certificate and modification data are unavailable, or even lost. But this partnership is certainly going in the right direction.



Source link -101