The problem this solves
Images in PDFs without alternate text are invisible to screen readers. Missing alt text on Figure tags is the most frequently cited PDF accessibility failure – and it directly violates WCAG 2.1 Success Criterion 1.1.1 (Non-text Content) and PDF/UA (ISO 14289) requirements.
Writing descriptions manually for documents with dozens or hundreds of images is impractical at any scale.
What this action does
The Generate Alternate Text action uses BLIP – an open-source vision-language AI model developed by Salesforce Research and pre-trained with a Vision Transformer architecture to automatically analyze every image in a PDF and generate a descriptive alt text string for it. The action runs entirely on your local machine inside a Docker container: no API key, no usage costs, no document data sent to any external server.
It ships as two variants inside PDFix Desktop:
- Set Alternate Description – PDF embed: Automatically generates alt text for every Figure tag in the PDF and writes each description directly into the figure’s
Altattribute. The output is a fully remediated PDF ready for PDF/UA validation.
- Set Alternate Description – text file export: Generates alt text from individual image files and saves each description to a text file. Use this variant when you need to review or edit descriptions before embedding them, or when integrating with an external editorial workflow.
Action
| Free · Local | Set Alternate Description | Automatically generates alternate text for Figure tags using Salesforce BLIP large, embedding it into each figure’s Alt attribute |
| Free · Local | Set Alternate Description | Automatically generates alternate text from an image file using Salesforce BLIP large, saving the description to a text file |
Frequently Asked Questions
What is WCAG 1.1.1 and why does it apply to PDF images?
WCAG 2.1 Success Criterion 1.1.1 requires that all non-text content – including images, figures, and charts – has a text alternative that conveys the same information. In PDFs, this is implemented through the Alt attribute on Figure tags. Documents that fail this checkpoint cannot be considered accessible under WCAG, Section 508, or the European Accessibility Act.
What is Salesforce BLIP?
BLIP – Bootstrapping Language-Image Pretraining – is an open-source vision-language model developed by Salesforce Research. It is trained to understand and describe image content in natural language. The PDFix action uses the BLIP-large variant, which offers higher description accuracy than the base model.
Does this action require an internet connection or API key?
No. The BLIP model runs entirely on your local machine inside a Docker container. No image data is transmitted to Salesforce, PDFix, or any third-party service. This makes it suitable for use with confidential or regulated documents.
How accurate are the generated descriptions?
BLIP-large produces accurate descriptions for photographs, diagrams, and general figures. For highly specialized technical images – such as engineering schematics, medical scans, or complex charts – the generated descriptions should be reviewed by a human before finalizing. PDFix Desktop allows you to edit any generated alt text before saving.
How does this differ from the Generate Alternate Text (OpenAI) action?
Both actions generate alt text for PDF images, but they differ in cost and data handling. The Salesforce BLIP action is 100% free and runs entirely on your local machine with no data leaving your environment. The OpenAI action connects to the OpenAI API, requires an API key, and is billed per use — but can produce more nuanced descriptions for complex or context-dependent images.
