slides-to-images

A library that converts a PowerPoint deck into an animation-playback video plus per-slide static frames — for embedding in another codebase, not for running standalone. Two independent capabilities:

• Timing analysis (build_recording_plan, compute_deck_timeline) — pure Python, parses the deck's OOXML <p:timing>/<p:transition> trees to compute per-slide/per-click pacing. No external dependencies. See TIMING.md for the full model.
• Conversion (convert) — drives a real, disposable LibreOffice Impress instance (via the docker-libreoffice-impress container image) to actually play the deck, recording video throughout and taking a live screenshot the moment each click settles. See IMPRESS_PROTOCOL.md for how that's driven end to end.

Install

pip install slides-to-images

The PyPI distribution name and the importable package name differ: pip install slides-to-images, but import slides2images.

For local development from a source checkout instead: pip install -e . (or .[test]/.[dev] — see pyproject.toml).

convert() additionally needs a reachable Docker daemon, with the image pulled: docker pull registry.gitlab.com/riphixel/docker-libreoffice-impress:latest.

Usage

Timing analysis only (no Docker required)

from pathlib import Path
from slides2images import build_recording_plan

plan = build_recording_plan(Path("deck.pptx"), click_delay_ms=500)
for slide in plan:
    print(
        f"slide {slide['index']}: "
        f"{len(slide['click_gaps_ms'])} click(s), "
        f"holds {slide['post_click_hold_ms']}ms after the last one"
    )

compute_deck_timeline() returns the same data uncollapsed — every animation node's start/end/duration, trigger, and target shape — useful for debugging a deck's timing rather than just driving a recorder with it.

Full conversion to video + static frames

from pathlib import Path
from slides2images import convert

result = convert(Path("deck.pptx"), Path("out/"), width=1920, height=1080)

print(result.video_path)  # out/recording.mp4
for image_path in result.image_paths:  # out/images/frame-0-0.png, frame-0-1.png, frame-1-0.png, ...
    print(image_path)

convert() raises slides2images.RecordingError for anything that stops a conversion short — an unsupported file extension, a deck with zero slides, or a failure in the underlying Docker/LibreOffice pipeline.

run.py in the repo root is this same example as a working CLI:

python3 run.py deck.pptx   # writes deck/recording.mp4 + deck/images/*.png

Supported input files

SUPPORTED_EXTENSIONS (also enforced by convert()):

• Office Open XML Presentations and Templates — .pptx, .ppsx, .ppmx, .potx, .pomx (both strict and transitional OpenXML).
• Legacy Microsoft PowerPoint 97/2000/XP/2003 — .ppt, .pps, .ppm, .pot, .pom. Converted to .pptx internally before anything else runs (python-pptx, which the timing model uses, only reliably opens the modern OOXML zip structure).

RESOLUTIONS is a small convenience dict of common (width, height) pairs ("vga", "720p", "hd", "4k") if you'd rather not hardcode pixels.

See also

• TIMING.md — how animation and transition timing is computed, including the corner cases found by validating against real PowerPoint-rendered reference videos.
• IMPRESS_PROTOCOL.md — how convert() actually drives a live Impress instance end to end: UNO connection, auto-advance disabling, the per-slide control loop, container lifecycle.
• features.md — end-user-visible feature list.
• AGENTS.md — guidelines for working on this codebase.

Links

• PyPI: https://pypi.org/project/slides-to-images/
• Source, issues: https://gitlab.com/riphixel/slides-to-images