The aim of the VR-funded project "The Folk high School as Mirrors of Society" is to utilize annual reports of Swedish folk high schools as heuristic tools for understanding how these schools have presented and legitimized themselves, and to investigate broader societal changes that took place as Sweden modernized (1868-2018). With a history spanning 150 years, Swedish folk high schools have produced substantial printed materials, now gradually converted to a large digital text corpus through digitization. Recently visual large language models have been released that promise to solve a couple of challenges we have faced when converting scanned documents to searchable and analyzable data. Perhaps most promising is DeepSeek-OCR. This model can be used to perform high quality OCR, parse the layout of documents allowing the extraction of images and image captions from the documents, as well as providing textual descriptions of the images extracted. We have adapted the DeepSeek-OCR GitHub code to fit our purposes (https://github.com/balljok/DeepSeek-OCR) allowing us to run the model on our dataset in three passes. The first pass performs OCR and extract images. The second pass uses DeepSeek-OCR to create and embed descriptions of the images in the image file metadata. The third pass applies another LLM to perform translation of the image descriptions.