⏰ Generate Figure and Formula Alt Text, MathML, Table Summaries, and Annotation Descriptions Automatically with PDFix Desktop

You use your own OpenAI account and API key. OpenAI usage is not included in the PDFix license. PDFix simply connects to your OpenAI API key, so:

• You need your own OpenAI subscription/account.
• You can limit and monitor your own usage and costs.
• All billing for OpenAI runs through your own OpenAI account, not PDFix.

Processing time depends on which method you use:

Template-based methods: These run purely on the machine using PDFix’s own layout rules. They are very fast – typically, you can process many pages per second and large batches of files efficiently.

Local AI models: Speed depends on the performance of your machine and the specific model. BLIP, as integrated in PDFix Desktop, is quite fast and provides good results, but it’s still naturally slower than pure template processing because each image is actually sent to and processed by the AI model.

Cloud AI models: Speed depends on your internet connection, the chosen model, and the AI’s response time.

Right now, Docker is the primary way to integrate External Actions (including AI models like Docling) with PDFix Desktop.

Why Docker:

• You only need to get approval for one application (Docker) instead of many separate tools.
• It simplifies the installation of complex AI tools that have many dependencies.
• It lets you run multiple different AI models on the same machine without managing conflicting dependencies manually.

You can freely modify or merge AI-generated templates with your own.
The AI template is a standard JSON file that you can edit to:

• Add or adjust element properties (for example, marking an element as a heading)
• Modify tagging rules
• Apply heuristics or logic to improve structure recognition

You can do this manually or programmatically – for instance, by adding custom properties such as alternate text, role mappings, or structural adjustments to enhance accessibility and accuracy.

Yes. This is exactly what the local AI models are for. You can use offline AI actions, such as BLIP (for image descriptions) or Paddle (for math formulas and other tasks) and other local models available as External Actions. These run completely locally on your machine, even without an internet connection, so:

• No document data or images are sent to the cloud.
• You keep all content on-premise, which is ideal for confidential or regulated data.

If you choose to use OpenAI or other cloud-based AI, your data will be sent to that provider’s servers.

Yes – this is already possible. The AI models and External Actions demonstrated (via Docker) can also be run as command-line processes. Docker is used as a CLI tool, not a continuously running service. You can design a workflow where:

1. You run PDFix CLI to process PDFs
2. You run Docker-based AI actions as additional command-line steps
3. You chain these commands in scripts/batch jobs to automate full pipelines

If you need help setting this up, PDFix can provide examples, resources, and guidance on integrating AI actions with your PDFix CLI workflows.

At the moment, there is no built-in, ready-to-use feature in PDFix Desktop that automatically detects repeated identical images, and reuses a single AI result for all occurrences. However, this scenario can be handled with a custom solution:

• A separate application or script that identifies duplicate images
• That custom tool then calls the AI once per unique image and applies the same alt text to all duplicates
• In some workflows, if repeated images are purely decorative, you might choose to artifact the duplicates and describe only one instance – but that depends on accessibility requirements and document semantics.

So: no automatic duplicate-detection/reuse yet, but it’s technically feasible via a custom integration.

Yes, this is fully supported and is often the recommended approach.

Example workflow:

1. First pass – template/basic actions
   • Use template-based functions like Set Alternate Description or Set Actual Text to automatically set alt text for images you can clearly define with rules and target specific images by size, position, page number, or other template rules.
2. AI actions
   • After the first pass, some images still won’t have alt text. Then you run an AI action (e.g., BLIP, OpenAI) on the remaining images.
   • Let AI generate descriptions only where templates didn’t already set alt text.
   • You can also bundle this into a single custom workflow/action in PDFix:
     • Action step 1: Set Alternate Description (template-based).
     • Action step 2: Set Alternate Description (OpenAI) or another AI model.

This combined workflow can then be run on a single document, or in batch mode across many PDFs, so the mixed strategy is fully automated.