Generative AI and Content Credentials
2022 (DALL·E 2, Midjourney, Stable Diffusion); C2PA from 2021
Software that makes photo-realistic pictures from a text prompt blurred the line between photograph and image. Content Credentials try to record where a picture came from.
How it began
Diffusion models, trained on huge collections of images and captions, became public in 2022. Within months anyone could make convincing “photographs” of things that never happened. In parallel, Adobe, camera makers and news organisations had formed the Coalition for Content Provenance and Authenticity (C2PA).
How it works
A diffusion model starts from random noise and removes it step by step, guided by a text prompt, until an image forms. Content Credentials attach a signed record to a file of how it was made and edited, which can be checked later.
Why it mattered
It raised the old question of whether a photograph is evidence, now at scale. In 2023 Boris Eldagsen won a Sony World Photography Award with an AI image and turned it down to start a debate.
Limits and what replaced it
Credentials only help if they are kept and checked, and metadata is easily removed. How photography and generated imagery will be told apart is still open.
Today
AI generation and editing tools are built into phones and editors; cameras from Leica, Sony and Nikon can sign their images.