Intelligent Assistant for the Contextualized Description of Images in News

Image description in journalistic environments is essential both for providing context to the photographs accompanying news stories and for ensuring the accessibility of digital content for people with visual impairments. However, the creation of captions and, in particular, alternative text is often constrained by production deadlines and newsroom workflows.
To address this need, the project "Intelligent Assistance for Context-Aware Photo Description in News Media", developed by 3Cat and Eurecat within the framework of CIDAI, explored the use of artificial intelligence models to automatically generate alternative text and photo captions from both news images and their accompanying textual context.
Throughout the project, several experiments were conducted using real news articles and images to analyse the impact of contextual information, person identification, translation quality, and potential biases introduced by model instructions. The evaluation carried out with media professionals showed a general preference for texts generated using the news context, as well as higher ratings for photo captions than for alternative text. Nevertheless, human supervision remains essential to ensure editorial quality.
The project's outcomes were subsequently transferred to 3Cat's production environment through integration into its newsroom workflows. The resulting functionality enables journalists to generate automatic proposals that can be reviewed and edited before publication, improving productivity, strengthening accessibility, and maintaining journalists as the central decision-makers in the editorial process.