OCR in Nextcloud - giving up after a month

Hi everyone,

I wanted to share my experience trying to set up OCR in Nextcloud over the past month. This isn’t a call for help - I’m done trying. Just feedback that I hope might be useful for the Nextcloud team or others considering OCR functionality.

I’ve spent a month attempting various approaches:

  • Full Text Search with Tesseract OCR app

  • OCRmyPDF integration

  • Paperless-ngx bridges with Nextcloud

  • Various custom configurations and workarounds

For context, I’m comfortable managing my own server - Docker environments, PostgreSQL databases, reverse proxies, etc. I’ve successfully set up complex self-hosted systems before. This isn’t about lacking technical skills.

Yet nothing has worked reliably for OCR. Things that should be straightforward turned into endless debugging sessions. Components that claim to work together simply don’t. Manual tests succeed while automated processes fail silently.

After a month of troubleshooting, reading documentation, checking logs, adjusting configurations, and rebuilding indexes, I’m calling it. The time investment just isn’t worth it anymore.

I love Nextcloud for collaboration, file sharing, and team workflows. But the OCR functionality feels fundamentally broken or at least severely under-documented. It’s genuinely disappointing because everything else works so well.

Just wanted to put this out there. Maybe it’ll help set expectations for others, or maybe it’ll reach someone who can improve the situation.

Thanks for reading.

1 Like

I agree that FTS could use up-to-date community facing documentation.

It may be better on the enterprise side - not sure.

The most straightforward community FTS deployment is likely via AIO, since it already comes with FTS (optionally) pre-configured.

Might be worth logging a reminder issue at the Nextcloud Documentation repo: GitHub - nextcloud/documentation: 📘 Nextcloud documentation

I have to admit Nextcloud is not good with “foreign” or “non-core” tasks. File management, sharing and Office (Collabora integration :wink: ) work well, but other things like Photos app, Paperless-ngx integration are really basic and unready. Mail app becomes somewhat useable after years

In my setup I switched to Joplin long time ago for managing notes, I’m planning so start using Immich or Photoprism for pictures and most likely Paperless-ngx is better for structured documents management.

Would be really great if there would be less “half-ready” apps in favor of solid 3rd party integrations.

3 Likes

Speaking of Nextcloud Mail, Roundcube has been under Nextcloud’s umbrella for quite some time now. Why not just use that and adapt the design a little more to Nextcloud? It might not be all that easy to integrate it seamlessly into Nextcloud’s features, but at least the core of Roundcube, the email program itself, has been rock solid for years and years, even with very large mailboxes.

:100: Agree. After all, neither M365 nor Google Workspace have a built-in DMS or photo management solution. I think, integrations with well-established products such as PaperlessNGX or Immich, which specialise in these things, would indeed be more useful.

1 Like

Regarding OCR: Fulltextsearch tesseract is “broken” since years. The good thing is: you don’t need it.

workflow_ocr does a great job and provides good defaults and the possibility for custom adjustments if required. Processed pdfs are automatically picked up by fulltextsearch. Thus, no need for fulltextsearch tesseract.

Workflow OCR - Apps - App Store - Nextcloud

2 Likes

Tried Workflow OCR but could’nt make it work neither :rofl:

So ended, modifying all my doc names to make them searchable !

Hi,

I would personally recommend the memories app: GitHub - pulsejet/memories: Fast, modern and advanced photo management suite. Runs as a Nextcloud app.

That’s what I’m using. However, it still relies on Nextcloud’s auto upload and Recognize for object and face recognition. In those categories, Immich is simply better.

Here’s an example. I set up two test instances. Immich and Nextcloud with Recognize and uploading the same few hundred photos to both instances. Immich recognised most of the faces within just a few minutes, whereas Recognize stubbornly insisted that there were too few faces to build a cluster or something like that..

Another reason, and in my opinion a more important one, is that the more apps you install on Nextcloud, the greater the likelihood of problems arising.

Despite this, I do like Memories quite a bit, and I think the Android app is actually quite good as well.

Just my two pennies worth on OCR. I use OCRmyPDF a lot and have it running locally, but it is always a little hit and miss depending on the format of the text in the document. I have played a lot with selecting different options in Tesseract to get better results but with little success on problem documents. I have an old copy of Finereader on an old W10 box and often drop back to that to get a clean result, but only for useful stuff, and those I spend the time spellchecking. Tesseract is not a good base to convert a scanned PDF into a ‘word processor’ document which Finereader is a lot more capable of doing.

All that said, on the whole having thrown a lot of magazines through the scanner and added search to them, they work well once then fed into FTS options. I can find the mags containing a particular target easily enough direct from Dolphin and the same capability should work well enough in Nextcloud?

2 Likes

I also use ocrmypdf for pdf files and textcleaner with tesseract for picture files. Works like a charm for me.

Yes, but full-text search is another one of those things (like Recognize) that, at best, only generates error messages in the log and, at worst, simply doesn’t work for a few weeks if you’ve updated to a new Nextcloud version too early. In the end, it doesn’t matter which OCR backend you use, because if the Fuill Text search app itself doesn’t work, it won’t work, even if you only upload PDFs where the text is already available. :wink:

And honestly. Now you could say, don’t always update right away, and I say that myself when others here have problems. But keeping track of a multidude apps and backends, whether they are all compatible with the next version, and how they might affect each other, is a full-time job. At the end of the day, it’s easier to use dedicated services that (more or less) just work on their own, instead of trying to squeeze everything into Nextcloud.

But just to be clear, I’m talking about full-text search, which is something you would expect from a modern file storage/sharing service today, i.e it should be part of the core functionality, not an app, in my humble opinion. OCR, however, is another topic altogether. :wink:

For Google this is possibly true. M365 is SharePoint at its core, which was developed as a DMS in the beginning. I deal with it professionally. You can manage tens of thousands of documents, have automated multilingual metadata, document workflows, ocr and machine learning models, identifying metadata and so on.