docling-project / docling-project/docling
granite_docling + docling: Images kept as images when output is not markdown
- Dominant language
- Python
- Stars
- 66.4k
- Forks
- 4.8k
- Avg merge
- 2d 21h
- Merged PRs (30d)
- 84
Description
### Bug
Demo of granite_docling can be instructed asking "convert chart to json" to produce a valid json of the data inside a chart.
Docling + granite-docling behaviour: if image is small (800x600), is just embedded as image into md and all other formats, if image is bigger (1600x1200) md is produced fine, all other formats (i.e.: --to json ) are embedding the picture. Larger images gets embedded in markdown too.
### Steps to reproduce
Download chart from WebGPU demo of granite-docling at: https://huggingface.co/spaces/ibm-granite/granite-docling-258M-WebGPU/blob/main/assets/chart.png
then convert it with:
docling --pipeline vlm --vlm-model granite_docling chart.png
result is only embedding the picture.
resize 2x and try the same command: result is 4k correct markdown
try to convert in json:
docling --pipeline vlm --vlm-model granite_docling --to json chart.png
result is only embedding the picture.
try the same command on resized image -> result is again only embedding the image.
### Docling version
2025-10-16 14:37:36,251 - INFO - Loading plugin 'docling_defaults'
2025-10-16 14:37:36,254 - INFO - Registered ocr engines: ['auto', 'easyocr', 'ocrmac', 'rapidocr', 'tesserocr', 'tesseract']
Docling version: 2.57.0
Docling Core version: 2.48.4
Docling IBM Models version: 3.9.1
Docling Parse version: 4.5.1
Python: cpython-312 (3.12.10)
Platform: Windows-11-10.0.26100-SP0
### Python version
Python 3.12.10
Contributor guide
Assessment
This issue has not been assessed yet.