Top tip: put any custom config variables (such as custom file naming) in the docker-compose file under “environment”.

But, I've since found out that my existing OCR'd PDFs have a janked-up OCR layer that Paperless-ng does NOT like - the text content is saved in a single column of characters. Not Paperless-ng's fault, just something to do with the way SwiftScan has saved the files.

So, after a LOT of hunting, I've eventually settled on [https://www.pdfshaper.com/|PDF Shaper Free] for Windows. The free version still allows exporting all images from a PDF. Then I convert all those images back into a fresh, clean PDF (no dirty OCR). This gets dumped in Paperless-ng and job's a good'un.

Top tip: experiment with the DPI setting for image exports to get the size/quality you want, as the DPI can be ignored in the import process.