Optimizing Image Library for Document Agents

This article explains how the Templafy Document Agent searches for images and how to organize your Templafy image library for the best results.

Overview

The Universal Document Agent only uses images from your Templafy library. It will never pull random stock photos from the internet, which means you stay in full control of your visual brand.

For the agent to find and use the right images, your library needs to be properly organized.

How the agent searches for images

When choosing an image for a placeholder, the Document Agent combines two types of search.

Vector search helps the system understand what an image represents and matches it to your content based on meaning, not just keywords. This means it can surface relevant images even when the exact words are not present in the file metadata. Because vector search carries more weight, uploading clear, relevant, on-brand visuals matters more than spending hours tagging every file.

Text search looks at the file metadata: the image name, description, and tags. While it plays a supporting role, well-written metadata still helps the agent search your library more effectively.

How to optimize your image library

Taking time to organize your library improves the quality of the agent's output. Below are the steps we recommend.

1. Upload enough images. If your library is too small, the agent will not have many options to choose from. A larger, more varied collection gives the agent more to work with.
2. Upload high-quality images with clear meaning. Images that are too abstract or blurry are harder for the agent to interpret. The clearer the image, the better the agent can match it to your content.
3. Name your images clearly. Avoid generic file names. Use descriptive names that reflect what is actually in the image.
4. Tag your images. Tags help the agent search your library more effectively. For example, an image of a handshake might be tagged with "partnership," "agreement," or "collaboration."
5. Use the description field. Add context that a file name and tags alone cannot capture, like the aesthetic, the mood, or the setting. The more relevant information you add, the better the agent can match images to content.