Firefox 132.0.1 Adds Mozilla AI for Image Descriptions

Mozilla has released the corrective update Firefox 132.0.1, addressing issues related to playing videos on certain sites with Widevine L3 DRM protection, such as Nintendo.com. The new version also fixes an error that affected the default theme design after updating Firefox.

Additionally, Mozilla has urged users and developers to participate in testing a machine learning model called Distilvit, designed to automatically generate text descriptions of images. This model, introduced in Firefox 130, is currently being tested in the PDF viewer to create annotations and will later be used to describe image contents for visually impaired individuals using screen readers. Users are requested to provide feedback on the generated descriptions to improve accuracy, especially for complex images.

The Distilvit model for text generation occupies 180 MB on disk, encompasses 182M parameters, and utilizes the Vision Transformer (ViT) model for image analysis and the Distilgpt2 model for generating text. The model is trained on publicly available data through the Odo dataset, and its distribution aligns with the openness standards set by the Open Source Initiative. The code for the project is shared under the Apache 2.0 license on GitHub.

/Reports, release notes, official announcements.