Skip to main content

​
Overview

The PPTX backend (MsPowerpointDocumentBackend) parses Microsoft PowerPoint presentations (.pptx files) and converts them directly to DoclingDocument format. Each slide becomes a page with extracted content including text, tables, and images.

​
Features

  • Slide-by-page conversion - Each slide becomes a document page
  • Text extraction - Titles, subtitles, body text, and notes
  • List detection - Bullet points and numbered lists with hierarchy
  • Table extraction - Tables with cell spans and structure
  • Image extraction - Embedded pictures and shapes
  • Notes preservation - Speaker notes as furniture content
  • Grouped shapes - Handles grouped shape content
  • Placeholder detection - Identifies title, subtitle, and body placeholders

​
Usage

​
Basic Conversion

from docling.document_converter import DocumentConverter

converter = DocumentConverter()
result = converter.convert("presentation.pptx")

# Access converted document
doc = result.document
print(f"Slides: {len(doc.pages)}")

for page in doc.pages:
    print(f"\nSlide {page.page_no}:")
    # Iterate items on this slide
    for item, _ in doc.iterate_items(page_no=page.page_no):
        print(f"  {item.label}: {getattr(item, 'text', '')}")

​
With Format Options

from docling.document_converter import DocumentConverter, PptxFormatOption

converter = DocumentConverter(
    format_options={
        PptxFormatOption: PptxFormatOption(
            # PPTX backend has no specific options currently
        )
    }
)

result = converter.convert("presentation.pptx")

​
Slide Structure

Each slide is organized as a chapter group: 
from docling_core.types.doc import GroupLabel

for group, _ in doc.iterate_items():
    if isinstance(group, GroupItem) and group.label == GroupLabel.CHAPTER:
        print(f"Slide: {group.name}")
        # All slide content is nested under this group

​
Supported Elements

​
Text Content

Slide titles and subtitles are automatically detected based on placeholder types:
from docling_core.types.doc import DocItemLabel

for item, _ in doc.iterate_items():
    if item.label == DocItemLabel.TITLE:
        print(f"Title: {item.text}")
    elif item.label == DocItemLabel.SECTION_HEADER:
        print(f"Subtitle: {item.text}")
Regular text content from slide bodies:
for item, _ in doc.iterate_items():
    if item.label == DocItemLabel.PARAGRAPH:
        print(f"Text: {item.text}")
Speaker notes are extracted as furniture-layer content:
from docling_core.types.doc.document import ContentLayer

for item, _ in doc.iterate_items():
    if item.content_layer == ContentLayer.FURNITURE:
        print(f"Note: {item.text}")

​
Lists

PowerPoint lists are detected and preserved:
  • Bullet lists - Unordered list items with bullet markers
  • Numbered lists - Ordered lists with automatic numbering
  • Multi-level lists - Nested list hierarchies based on indentation
from docling_core.types.doc import ListItem

for item, _ in doc.iterate_items():
    if isinstance(item, ListItem):
        indent = "  " * item.level
        print(f"{indent}{item.marker} {item.text}")

​
List Detection Algorithm

The backend uses PowerPoint’s paragraph properties to determine list items:
  1. Checks direct paragraph properties (<a:pPr>)
  2. Falls back to shape-level list styles (<a:lstStyle>)
  3. Checks layout placeholder styles
  4. Uses slide master text styles
Bullet markers:
  • <a:buChar> - Character bullets (•, ○, ■, etc.)
  • <a:buAutoNum> - Automatic numbering
  • <a:buBlip> - Picture bullets
  • <a:buNone> - Explicitly no bullet

​
Tables

Complete table extraction with structure: 
for table, _ in doc.iterate_items():
    if isinstance(table, TableItem):
        print(f"Table: {table.data.num_rows} x {table.data.num_cols}")
        
        for cell in table.data.table_cells:
            print(f"  ({cell.start_row_offset_idx},{cell.start_col_offset_idx}): {cell.text}")
            print(f"    Span: {cell.row_span}x{cell.col_span}")
            print(f"    Header: col={cell.column_header}, row={cell.row_header}")
Features:
  • Cell content and formatting
  • Merged cells (rowSpan, gridSpan)
  • Header row/column detection
  • Empty cell handling

​
Images

Extracts embedded pictures: 
for item, _ in doc.iterate_items():
    if isinstance(item, PictureItem):
        print(f"Image found on slide {item.prov[0].page_no}")
        
        # Access image data
        img = item.image.pil_image
        img.save(f"slide_image_{item.self_ref}.png")
Supported:
  • Inline pictures
  • Picture shapes
  • Image formats: JPEG, PNG, BMP, etc.
  • DPI information preserved

​
Slide Layout

Slide dimensions and layout information: 
for page in doc.pages:
    print(f"Slide {page.page_no}:")
    print(f"  Size: {page.size.width} x {page.size.height}")
    # Size in EMU (English Metric Units) by default

​
Provenance Information

All extracted items include provenance with position on slide: 
for item, _ in doc.iterate_items():
    if item.prov:
        prov = item.prov[0]
        print(f"Item on slide {prov.page_no}:")
        print(f"  BBox: {prov.bbox}")
        print(f"  Text: {getattr(item, 'text', '')}")
Bounding boxes use slide dimensions as coordinate system.

​
Grouped Shapes

Handles PowerPoint shape groups: 
# Grouped shapes are automatically processed recursively
# Content from all shapes in a group is extracted
The backend recursively processes:
  • Shape groups
  • Nested groups
  • Individual shapes within groups

​
Advanced Features

​
Placeholder Types

Automatic detection of PowerPoint placeholders:
  • PP_PLACEHOLDER.TITLE - Slide title
  • PP_PLACEHOLDER.CENTER_TITLE - Centered title
  • PP_PLACEHOLDER.SUBTITLE - Subtitle
  • PP_PLACEHOLDER.BODY - Content placeholder
  • PP_PLACEHOLDER.OBJECT - Object placeholder

​
Line Breaks

Line breaks in PowerPoint text are converted to spaces for better text flow: 
# PowerPoint: "Line 1\nLine 2"
# Extracted: "Line 1 Line 2"

​
Empty Slides

Slides without content still create page entries: 
for page in doc.pages:
    # Every slide creates a page, even if empty
    print(f"Slide {page.page_no} exists")

​
Performance

  • Speed: Fast declarative conversion (no ML models)
  • Memory: Low memory footprint
  • Concurrency: Thread-safe per document instance
import concurrent.futures

def convert_pptx(file_path):
    converter = DocumentConverter()
    return converter.convert(file_path)

# Process multiple presentations in parallel
files = ["pres1.pptx", "pres2.pptx", "pres3.pptx"]
with concurrent.futures.ThreadPoolExecutor() as executor:
    results = list(executor.map(convert_pptx, files))

​
Limitations

Known Limitations:
  • Animations: Animation sequences not preserved
  • Transitions: Slide transitions not captured
  • Embedded Media: Videos and audio not extracted
  • Charts: Chart data not extracted (renders as image)
  • SmartArt: SmartArt graphics may not render correctly
  • Master Slides: Template information not fully preserved

​
Troubleshooting

Cause: Complex list style inheritanceCheck: Verify list formatting in PowerPoint sourceNote: Backend checks multiple levels of style inheritance
Cause: Custom start numbers or broken numberingSolution: Backend respects start attribute on ordered lists
Check: Verify slides have speaker notes in PowerPoint
# Notes appear as FURNITURE content layer
for item, _ in doc.iterate_items():
    if item.content_layer == ContentLayer.FURNITURE:
        print(f"Note: {item.text}")
Note: Images are extracted at original embedded resolutionWorkaround: Use higher resolution images in source presentation

​
Export Formats

After conversion, export to various formats: 
result = converter.convert("presentation.pptx")
doc = result.document

# Export to Markdown (slide-by-slide)
markdown = doc.export_to_markdown()

# Export to JSON
json_doc = doc.model_dump_json()

# Export to plain text
text = doc.export_to_text()

​
Use Cases

Content Extraction

Extract text and tables from presentations for analysis or archival

Slide Summarization

Convert presentations to text format for LLM processing and summarization

Training Materials

Extract course content from educational presentations

Documentation

Convert technical presentations to structured documentation

​
See Also

Build docs developers (and LLMs) love

Get started for freeTalk to us