Prepare Document Knowledge Base with Image Index¶

Add page and document-image indexes to text knowledge base preparation. Use it when retrieval also needs to use images.

Prepare your resources¶

Prepare documents containing text and images, a text knowledge base, image-index settings, and an output volume. Start with one representative document to check parsing and indexing.

The text embedding model processes text; the image embedding model processes images. Check that image indexing is enabled and that its table, model, dimensions, preprocessing version, and distance metric match. Captions alone do not establish an image index.

Use the template¶

On the workflow creation page, start from a template, search for Prepare Document Knowledge Base with Image Index, and load it onto the canvas.

Complete the required fields in the parameter form.

Setting	Purpose
Source files	Documents or folders to process
Parsing tier	Controls parsing speed, quality, and cost
Knowledge base	Select an existing knowledge base or create one
Output location	Catalog location for parsed result files

Before running, check Image Embedding and confirm that the image model matches the index dimensions.

Save the parameters, then start a manual run. The system saves the current workflow before submitting execution.

Complete processing flow¶

Arrows show execution order; configured bindings supply each step’s input.

Prepare Document Knowledge Base with Image Index

Adjust text and image indexes¶

Setting	How to use it
Parsing tier	Standard by default; use Enhanced for scans or complex layouts; see General Document Parsing
Text chunks	Template defaults: 512 characters with overlap 50; check parsed text before adjusting
Three-level text index	Enabled by default, with section size 5
Image index	Confirm it is enabled and retains page and object image references
Model and dimensions	Match image model output to the image index; do not reuse text dimensions without checking

The template embeds actual images and allows files with no indexable images. no_indexable_images means no image vectors were written for that file; disabled means indexing was disabled. Neither indicates successful image indexing.

Inspect the completed run¶

Check file counts and image sources in parsed records.
Inspect text written and image page_rows, visual_object_rows, and file_statuses.
Check indexed_source_file_ids for sources actually indexed. all_source_file_ids only lists sources considered.
Inspect saved parsed files and both text and image lineage results.
Validate text and image retrieval separately with known content.

The template builds the text index before the visual manifest and image index, then saves parsed documents and registers relationships. Manifest and file-saving steps use documents before chunking; text indexing uses chunks.

Example: an illustrated equipment manual¶

Use a manual containing a parameter table and an equipment diagram. Keep default chunking, verify both model configurations, and run. Compare text results with the table, then inspect image records and their source against the diagram.

If images are missing, inspect readable image references in parsed output before checking manifests and indexing. Successful text retrieval alone does not verify the image path.

Common problems¶

Symptom	Check and action
Text retrieval works but images do not	Check image enablement, sources, and per-file status
Image dimensions mismatch	Compare model output with index settings
Some files have no image rows	Distinguish no indexable images from failures in `file_statuses`
Rerunning changes the index	The template uses `OVERWRITE` for images; check the destination and existing results first
Final registration fails	Inspect source-to-derived-image mappings before repeating steps that already wrote data

Last updated on Sep 15, 2026