Image being None

[gustavotrapp]

I am trying to extract the images from my PDF and save them so I can use a model to process them separately, but all of my PictureItem have None as the image

self_ref='#/pictures/0' parent=RefItem(cref='#/body') children=[] content_layer=<ContentLayer.BODY: 'body'> meta=None label=<DocItemLabel.PICTURE: 'picture'> prov=[ProvenanceItem(page_no=1, bbox=BoundingBox(l=435.3965759277344, t=1031.965051651001, r=657.4712524414062, b=715.9979858398438, coord_origin=<CoordOrigin.BOTTOMLEFT: 'BOTTOMLEFT'>), charspan=(0, 0))] source=[] comments=[] captions=[] references=[] footnotes=[] image=None annotations=[]
self_ref='#/pictures/1' parent=RefItem(cref='#/body') children=[] content_layer=<ContentLayer.BODY: 'body'> meta=None label=<DocItemLabel.PICTURE: 'picture'> prov=[ProvenanceItem(page_no=1, bbox=BoundingBox(l=433.48516845703125, t=171.64434814453125, r=655.8321533203125, b=6.304443359375, coord_origin=<CoordOrigin.BOTTOMLEFT: 'BOTTOMLEFT'>), charspan=(0, 0))] source=[] comments=[] captions=[] references=[] footnotes=[] image=None annotations=[]

[dosubot[bot]]

The image field on PictureItem is None because you need to set generate_picture_images=True in your PdfPipelineOptions. By default this is False, so the pipeline doesn't crop individual picture images from pages.

If you're using the Python SDK directly, configure it like this:

from docling.datamodel.pipeline_options import PdfPipelineOptions
from docling.document_converter import DocumentConverter, PdfFormatOption

pipeline_options = PdfPipelineOptions(
    generate_picture_images=True,  # enables per-picture image cropping
    images_scale=2.0,              # optional: higher resolution
)

converter = DocumentConverter(
    format_options={
        InputFormat.PDF: PdfFormatOption(pipeline_options=pipeline_options)
    }
)

If you're using docling-serve, note that there's a bug where image_export_mode: "embedded" does not enable generate_picture_images — only "referenced" does [1]. As a workaround, use "referenced" mode, which will populate the image field with a URI and metadata. You'd then need to crop/encode the images client-side if you need the actual bytes inline.

To reply, just mention @dosu.

---

Docs are dead. Just use Dosu.