Mixed text-and-image document RAG data preparation¶
Use this template to build an intelligent parsing and knowledge-extraction workflow for mixed text-and-image documents. It produces multimodal knowledge data for retrieval-augmented generation (RAG) applications based on text and images. The workflow identifies text and image sections, then organizes them into structured segments for document retrieval, summarization, and knowledge-base use.
Template details¶
Select View details in the template list to open the template details page. It shows example processing results and the workflow topology.
Use the template¶
Select Mixed text-and-image document RAG data preparation in the template list, then select Use template from the list or details page to create a data-processing task and its workflow.
The system includes sample data for initial testing.
Create the target location yourself.
Adjust parser, segmenting, and other workflow-operator configuration as needed.
Select Create and start and wait for the workflow to finish.
View processing results¶
Open Catalog, find the target location selected by the workflow, and select the preview button to the right of the file.
Export to Dify¶
Open Data connections > Connectors and create a Dify connector. For configuration details, see Connectors.
Under Data export, choose Export to knowledge base > Dify.
Configure the export, select the processed JSON file, and start the export task. The system synchronizes text-and-image segments to the target knowledge base.
Wait for the task status to become Completed, then verify the result in Dify.
Build a RAG application¶
Configure a model provider and API key on the settings page.
Select an appropriate AI model.
Create an application in Dify Studio and associate the imported knowledge base.
Select Preview to test mixed text-and-image question answering.
