Skip to content
Back to skills

5 Image Insertion And Positioning

ASecurity

Sub-skill of python-docx: 5. Image Insertion and Positioning.

  • 17 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 9, 2026
developmentpythongo

Security analysis

A100/100

Scanned September 9, 2026

npx -y skills add vamseeachanta/workspace-hub --skill 5-image-insertion-and-positioning --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of 5 Image Insertion And Positioning?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for 5 Image Insertion And Positioning
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/vamseeachanta-5-image-insertion-and-positioning/badge)](https://www.skillsdirectory.com/skills/vamseeachanta-5-image-insertion-and-positioning)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: python-docx-5-image-insertion-and-positioning
description: 'Sub-skill of python-docx: 5. Image Insertion and Positioning.'
version: 1.0.0
category: data
type: reference
scripts_exempt: true
---

# 5. Image Insertion and Positioning

## 5. Image Insertion and Positioning


```python
"""
Insert and position images in Word documents.
"""
from docx import Document
from docx.shared import Inches, Pt, Cm, Emu
from docx.enum.text import WD_ALIGN_PARAGRAPH
from docx.oxml.ns import qn
from docx.oxml import OxmlElement
from pathlib import Path
from io import BytesIO
from typing import Optional, Tuple
import requests

def add_image_from_url(doc: Document, url: str, width: Optional[float] = None) -> None:
    """Add image from URL to document."""
    response = requests.get(url)
    image_stream = BytesIO(response.content)

    if width:
        doc.add_picture(image_stream, width=Inches(width))
    else:
        doc.add_picture(image_stream)


def add_image_with_caption(
    doc: Document,
    image_path: str,
    caption: str,
    width: float = 4.0,
    figure_num: int = 1
) -> None:
    """Add image with centered caption below."""
    # Add image
    para = doc.add_paragraph()
    para.alignment = WD_ALIGN_PARAGRAPH.CENTER
    run = para.add_run()
    run.add_picture(image_path, width=Inches(width))

    # Add caption
    caption_para = doc.add_paragraph()
    caption_para.alignment = WD_ALIGN_PARAGRAPH.CENTER
    caption_run = caption_para.add_run(f'Figure {figure_num}: {caption}')
    caption_run.italic = True
    caption_run.font.size = Pt(10)


def add_inline_image(paragraph, image_path: str, width: float = 1.0) -> None:
    """Add image inline with text."""
    run = paragraph.add_run()
    run.add_picture(image_path, width=Inches(width))


def create_document_with_images(output_path: str, sample_image_path: str) -> None:
    """Create document with various image placements."""
    doc = Document()

    doc.add_heading('Image Examples', level=0)

    # Check if sample image exists
    if not Path(sample_image_path).exists():
        # Create a placeholder message if no image
        doc.add_paragraph(
            'Note: Sample image not found. Replace with actual image path.'
        )
        doc.save(output_path)
        return

    # Simple centered image
    doc.add_heading('Centered Image', level=1)
    doc.add_paragraph('The image below is centered on the page:')

    img_para = doc.add_paragraph()
    img_para.alignment = WD_ALIGN_PARAGRAPH.CENTER
    run = img_para.add_run()
    run.add_picture(sample_image_path, width=Inches(4))

    # Image with caption
    doc.add_heading('Image with Caption', level=1)
    doc.add_paragraph('Images can include descriptive captions:')
    add_image_with_caption(
        doc,
        sample_image_path,
        'Sample chart showing quarterly results',
        width=4.0,
        figure_num=1
    )

    # Multiple images in a row (using table)
    doc.add_heading('Multiple Images Side by Side', level=1)
    doc.add_paragraph('Use tables to arrange images side by side:')

    # Create 1x3 table for images
    table = doc.add_table(rows=2, cols=3)

    for i in range(3):
        cell = table.cell(0, i)
        para = cell.paragraphs[0]
        para.alignment = WD_ALIGN_PARAGRAPH.CENTER
        run = para.add_run()
        run.add_picture(sample_image_path, width=Inches(1.8))

        # Add caption below each image
        caption_cell = table.cell(1, i)
        caption_cell.paragraphs[0].text = f'Image {i + 1}'
        caption_cell.paragraphs[0].alignment = WD_ALIGN_PARAGRAPH.CENTER

    # Inline image with text
    doc.add_heading('Inline Images', level=1)
    para = doc.add_paragraph('You can include small images ')
    run = para.add_run()
    run.add_picture(sample_image_path, width=Inches(0.5))
    para.add_run(' inline with your text for icons or small graphics.')

    doc.save(output_path)
    print(f"Document with images saved to {output_path}")


# Usage (provide path to actual image)
# create_document_with_images('image_document.docx', 'sample_chart.png')
```

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.