openwebui docling

The content extraction engine I set in the documentation feature of OpenWebUI is docling. However, when handling files containing images, docling's powerful OCR recognition can indeed parse the images, but they are saved in base64 encoding. Moreover, this content is given to the large language model for processing with an "image" placeholder, which means that in reality, my model does not know what the images contain. What should I do?
The OpenWebUI settings are as follows：

The file with images is parsed as follows, and you can see that the image is an 'image'.