Can Docling Be Integrated with Google Cloud Vision for OCR?

[pradeept]

While researching online, I came across the package docling-google-ocr. However, when I tried to install it, it conflicted with the version of the Docling parser I’m using. To make it work, I would need to downgrade Docling, which would cause me to lose several features.

Is there another way to use Google Vision for OCR without downgrading Docling?

Thank you.

[dosubot[bot]]

Hi @pradeept! I'm Dosu and I’m helping the docling team.

There's no built-in Google Cloud Vision integration in Docling, but you don't need to rely on the third-party docling-google-ocr package (or downgrade Docling) — you can write your own OCR plugin using Docling's plugin system.

Docling uses a pluggy-based plugin architecture that lets you register custom OCR engines. The steps are:

1. Create an options class extending OcrOptions with a unique kind discriminator [1].

2. Create an OCR model class extending BaseOcrModel and implement the __call__() method to process page images and return TextCell results [2].

3. Register your plugin via a setuptools entrypoint in your pyproject.toml [3]:

   [project.entry-points."docling"]
   google_vision_ocr = "your_package.module"

4. Use it with allow_external_plugins=True:

   pipeline_options = PdfPipelineOptions(
       do_ocr=True,
       allow_external_plugins=True,
       ocr_options=YourGoogleVisionOcrOptions(lang=["en"]),
   )

The docling-surya plugin is a good reference implementation to follow — it's a third-party OCR plugin that follows exactly this pattern [4]. Your Google Vision plugin would work similarly, but instead of running a local model, you'd call the Google Cloud Vision API in the __call__() method and map the results to Docling's TextCell format [5].

One note: there's a reported issue where the --allow-external-plugins CLI flag may not work properly [6], so using the Python API as shown above is the more reliable approach for now.

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.