Generate PowerPoint in Python (2026): python-pptx, an API — or both

There are three ways to generate PowerPoint files from Python in 2026: the python-pptx library (free, full control, you own every pixel), a structured rendering API like SlideForge (typed data in, validated consulting-grade .pptx out, $0.05/slide), or — new since mid-2026 — both at once: writing python-pptx code and running it in SlideForge's sandbox, where it gets themes, chart widgets, and output verification for free. This tutorial covers all three with working code.

1. python-pptx: The Standard Library

python-pptxis the de facto Python library for creating and modifying PowerPoint files. It's free, open source (MIT license), and gives you full programmatic control over every element in a .pptx file.

Install and create a basic slide

pip install python-pptx

from pptx import Presentation
from pptx.util import Inches, Pt
from pptx.dml.color import RGBColor

# Create presentation
prs = Presentation()
prs.slide_width = Inches(13.333)
prs.slide_height = Inches(7.5)

# Add a blank slide
slide = prs.slides.add_slide(prs.slide_layouts[6])

# Add a title
title = slide.shapes.add_textbox(Inches(0.5), Inches(0.3), Inches(12), Inches(0.6))
tf = title.text_frame
tf.text = "Q1 2026 Performance"
tf.paragraphs[0].font.size = Pt(28)
tf.paragraphs[0].font.bold = True

# Save
prs.save("basic_slide.pptx")

This works. But the output looks like it was made by a developer, not a designer. No color harmony, no spacing system, no visual hierarchy beyond font size.

Building a real slide: KPI Dashboard

Here's what it takes to build a consulting-quality KPI dashboard with python-pptx:

from pptx import Presentation
from pptx.util import Inches, Pt, Emu
from pptx.dml.color import RGBColor
from pptx.enum.text import PP_ALIGN

prs = Presentation()
prs.slide_width = Inches(13.333)
prs.slide_height = Inches(7.5)
slide = prs.slides.add_slide(prs.slide_layouts[6])

# Title
title = slide.shapes.add_textbox(
    Inches(0.5), Inches(0.3), Inches(12), Inches(0.6)
)
tf = title.text_frame
tf.text = "Q1 2026 Performance Summary"
tf.paragraphs[0].font.size = Pt(28)
tf.paragraphs[0].font.bold = True

# KPI metrics — repeat for each one
metrics = [
    ("$12.4M", "Revenue", "+18% YoY"),
    ("847", "New Clients", "+23%"),
    ("94.2%", "Retention Rate", ""),
    ("4.7", "NPS Score", "+0.3"),
]

for i, (value, label, trend) in enumerate(metrics):
    left = Inches(0.5 + i * 3.1)
    top = Inches(1.5)
    width = Inches(2.8)
    height = Inches(2.0)

    # Background box
    box = slide.shapes.add_shape(1, left, top, width, height)
    box.fill.solid()
    box.fill.fore_color.rgb = RGBColor(0xF5, 0xF5, 0xF5)
    box.line.fill.background()

    # Value text
    val_box = slide.shapes.add_textbox(left, Inches(1.7), width, Inches(0.8))
    val_tf = val_box.text_frame
    val_tf.text = value
    val_tf.paragraphs[0].font.size = Pt(36)
    val_tf.paragraphs[0].font.bold = True
    val_tf.paragraphs[0].alignment = PP_ALIGN.CENTER

    # Label text
    lbl_box = slide.shapes.add_textbox(left, Inches(2.5), width, Inches(0.4))
    lbl_tf = lbl_box.text_frame
    lbl_tf.text = label.upper()
    lbl_tf.paragraphs[0].font.size = Pt(9)
    lbl_tf.paragraphs[0].font.color.rgb = RGBColor(0x99, 0x99, 0x99)
    lbl_tf.paragraphs[0].alignment = PP_ALIGN.CENTER

    # ... and trend text, accent lines, spacing adjustments...

prs.save("kpi_dashboard.pptx")
# ~60 lines for a basic version. A polished version is 100+.

That's ~60 lines for a basic KPI dashboard. A polished version with accent lines, trend arrows, tinted backgrounds, and a takeaway banner runs 100+ lines. And this is one slide type — a consulting deck needs 10-15 different layouts.

The pain points

• Alignment is manual. Every coordinate is hardcoded in inches. Move one element and everything else needs adjusting.
• No design system. Colors, fonts, and spacing are ad-hoc. Consistency across slides requires discipline and shared constants.
• Limited chart support. No waterfall, treemap, or sunburst charts. Combo charts can corrupt files.
• No animations. The #1 requested feature (GitHub issue #1106) — still unsupported.
• Nothing checks the output. If your loop drops a metric or a textbox overflows the canvas, python-pptx saves the file happily. You find out in the meeting.
• Maintenance burden. Every new slide type is a multi-day project. Bug fixes in one template can break another.

2. SlideForge API: Typed Data In, Validated Slide Out

SlideForge renders the same KPI dashboard from a typed request: pick a form from the catalog (150+ layouts — KPI boards, waterfalls, Gantt plans, org charts, funnels…), put your numbers in the documented fields, and the engine binds them verbatim — your values appear exactly as sent, deterministically, with no LLM in the loop:

import requests

API = "https://api.slideforge.dev"
KEY = {"Authorization": "Bearer sf_live_YOUR_KEY"}

resp = requests.post(f"{API}/v1/render/intent", headers=KEY, json={
    "form": "kpi_metrics",
    "headline": "Q1 2026: all four KPIs trending up",
    "data": {
        "metrics": [
            {"value": "$12.4M", "label": "Revenue", "trend": "+18% YoY"},
            {"value": "847", "label": "New Clients", "trend": "+23%"},
            {"value": "94.2%", "label": "Retention Rate"},
            {"value": "4.7", "label": "NPS Score", "trend": "+0.3"},
        ],
    },
    "takeaway": "On track for the $50M full-year target",
})
result = resp.json()
print(result["status"])    # "complete"
print(result["fidelity"])  # "verbatim" — every value bound exactly as sent
job_id = result["job_id"]

# Download with the same Bearer key
pptx = requests.get(f"{API}/v1/jobs/{job_id}/pptx", headers=KEY)
open("kpi_dashboard.pptx", "wb").write(pptx.content)

Rendering completes in well under a second at $0.05/slide, and the output is a native .pptx with editable shapes — open it in PowerPoint and change anything.

The part python-pptx can't give you is the honesty layer: every response declares what happened. fidelity: verbatim means every supplied value rendered exactly. If your waterfall's deltas don't sum to the declared total, you don't get a wrong slide — you get status: completed_with_errorswith the exact discrepancy ("start+deltas = 49, declared end = 70"), and you aren't billed. A render that isn't usable is free.

For content that doesn't arrive as typed data yet, send a prose brief and let the router pick the form:

resp = requests.post(f"{API}/v1/render/auto", headers=KEY, json={
    "brief": "Revenue bridge Q4 to Q1: start $10.2M. New clients +$2.1M, "
             "upsell +$1.3M, churn -$0.8M, FX -$0.2M.",
})
# Routed to the waterfall form, numbers bound verbatim, <1.5s, $0.05

3. Both: Your python-pptx Code, Run in the SlideForge Sandbox

If you already write python-pptx — or your layout doesn't exist in any catalog — you don't have to choose. POST /v1/render/code executes your build() function in a sandbox where it gets what raw python-pptx lacks: a theme, pre-built chart widgets, chrome (title/takeaway furniture), a geometry linter, and the same output verification as the structured path.

code = '''
def build(prs, slide):
    # Chrome (title, takeaway band) is already rendered — draw the body.
    # add_widget draws SlideForge's own theme-correct parts:
    add_widget(slide, "kpi_metrics", x=0.6, y=1.4, w=12.1, h=4.6, content={
        "metrics": [
            {"value": "$12.4M", "label": "Revenue", "trend": "+18% YoY"},
            {"value": "847", "label": "New Clients", "trend": "+23%"},
            {"value": "94.2%", "label": "Retention Rate"},
            {"value": "4.7", "label": "NPS Score", "trend": "+0.3"},
        ],
    }, theme=THEME)
'''

resp = requests.post(f"{API}/v1/render/code", headers=KEY, json={
    "code": code,
    "headline": "Q1 2026: all four KPIs trending up",
    "takeaway": "On track for the $50M full-year target",
})
# Your code, their theme + widgets + linter. $0.05, ~2s, no LLM involved.

The widget catalog covers the engine's own chart primitives — waterfall, Gantt, funnel, org chart and the rest — so "custom layout" stops meaning "hand-place every rectangle." Static analysis catches overlapping and off-canvas shapes before you open the file, and geometry that can't work (negative sizes, impossible constraints) is rejected with the exact numbers instead of silently producing a broken slide.

4. When to Use Which

Use python-pptx alone if you need full control in an air-gapped environment, or have zero budget and unlimited time.

Use the intent API if your data is structured and the deck must be right unattended — reports, agents, SaaS export. Typed fields in, verbatim slide out, validated.

Use code mode if you think in python-pptx but want the theme, the chart widgets, and something checking the output — or your layout is genuinely outside any catalog.

They're also compatible in the other direction: every SlideForge output is a standard .pptx that python-pptx can open and post-process.

Getting Started

Sign up free (60 free slides, no card) and follow the quickstart guide to render your first slide. Browse the form catalog and per-form schemas in the template gallery, or read the detailed comparison of SlideForge vs python-pptx.