PPTX Studio

View, edit and save PowerPoint decks without leaving Obsidian.

yut0takagi.github.io/obsidian-pptx-studio — what it does, in English and Japanese.

Obsidian natively opens Markdown, images, audio, video, PDF and canvas — .pptx is not on that list, so a deck in your vault is invisible in the file explorer and cannot be linked to. This plugin registers the extension and renders slides directly from the OOXML, with no external converter and nothing to install alongside it.

Install

Settings → Community plugins → Browse, search for PPTX Studio, then install and enable it.

To run a build that has not been released yet, point BRAT at yut0takagi/obsidian-pptx-studio, or do it by hand: download main.js, manifest.json and styles.css from the latest release into <vault>/.obsidian/plugins/pptx-studio/.

Features

• Deck view — click a .pptx in the file explorer to open it. Thumbnail rail, ↑/↓ (or k/j) to page, zoom, speaker notes.
• Edit text in place — double-click a text box, type, Cmd/Ctrl+S to save. Only the slide XML you touched is rewritten; every other part of the file is repacked byte for byte, so animations, transitions, embedded fonts and any formatting this plugin does not model survive untouched.
• Edit like PowerPoint — a tabbed ribbon (Home, Insert, Format, Design, View, and a Table tab that appears when one is selected) over a slide canvas with multi-select, marquee selection, drag, resize, rotate, snapping guides, clipboard, z-order, align and distribute, group and ungroup, and a right-click menu.
• Format — font and size, bold/italic/underline/strikethrough, text colour, alignment, bullets and numbering, list level, line spacing, vertical anchor, shape fill and outline. With text selected the change applies to exactly those characters; with only the shape selected it applies to all of its text.
• Insert — text boxes, twelve shape presets, lines, pictures from the vault, and tables.
• Tables — click a cell to select it, shift-click to extend, then insert and delete rows and columns, merge and split cells, and fill them.
• Slides — add from any layout, duplicate, delete, and reorder by dragging a thumbnail. Set a slide's background colour.
• Rulers and guides — drag a guide out of a ruler, drag it to move, drag it off the slide to remove it. Shapes snap to guides. They are stored in the deck the way PowerPoint stores them, so they appear there too.
• Crop pictures — the source shows through dimmed while you drag the crop handles, and "reset" restores the whole image.
• Selection pane — every shape on the slide in stacking order: click to select, rename, hide, restack. For the shapes direct manipulation cannot reach.
• Also — hyperlinks, format painter, change a shape's preset, rotate and flip by 90°, and numeric position, size and rotation fields.
• Find and replace — across the whole deck, not just the slide on screen. Every hit is listed with the words either side of it and the slide it is on, and clicking one goes there. A phrase split across runs by a change of formatting is still found, and the replacement keeps the first run's formatting.
• Paste a picture — an image on the system clipboard drops straight onto the slide. Where the clipboard offers the same picture several ways, the lossless one is taken.
• Transitions and animations — a transition per slide, and fade, wipe or fly in as an entrance on any shape, starting on click, with the previous effect or after it. The order can be rearranged. An effect written by PowerPoint that this plugin does not model is listed but left alone rather than rewritten.
• Embed slides in notes — ![[deck.pptx]] embeds the deck, ![[deck.pptx#3]] pins a single slide.
• Extract to Markdown — turn a deck into an outline note so its content is searchable, linkable and shows up in backlinks.
• Export — a slide as PNG at 1×, 2× or 3×, the whole deck as a PDF, or straight to the print dialog. The PDF is written directly rather than through a converter, so it needs nothing installed either.

Language

The interface follows Obsidian's own display language, so a Japanese vault gets a Japanese ribbon with nothing to configure. Settings → PPTX Studio → Language can pin it to English or Japanese instead.

Getting around

The ribbon collapses with the chevron at the right of the tab strip, and remembers whether it was collapsed. In a narrow pane it drops its labels and group titles rather than scrolling sideways. Selecting a table brings up the Table tab automatically and deselecting it goes back to where you were.

A small toolbar follows the selection with the things you reach for constantly — fill, outline, bold, arrange, duplicate, delete, and crop on a picture. Nothing lives only there; it is a shortcut to what the ribbon already has.

Clicking in the ribbon never takes focus away from the slide, so the arrow keys keep working immediately afterwards.

Seeing the XML

The side pane (View → Selection) lists every shape on the slide, and underneath it shows what the file actually says about the selected one: its p:cNvPr id, the element it is (p:sp, p:pic, p:graphicFrame), the part it lives in, its placeholder type, and its position and size as both pixels on screen and the raw EMU stored in a:off / a:ext. A shape with no a:xfrm of its own says so, rather than quietly reporting the frame it inherited from the layout.

The frame, the rotation and the name are fields rather than labels: type over one and press Enter. A length takes pixels, or a number with a unit — emu, pt, in, cm, mm — so a value read out of another file can be pasted in as it stands. Everything a field writes goes through the same commands the ribbon uses, so it undoes with Cmd/Ctrl+Z like any other edit. The shape id, the part and the element stay read-only.

Below that is the shape's XML, indented and scrollable, with a button that copies it.

Drag the pane's left edge to resize it, and the line between the shape list and the properties to change how the height is shared. Either section folds away by clicking its title, and whichever is left open takes the whole column. Widths, heights and folds are all remembered; double-clicking a divider puts it back.

Holding a box, and typing in it

A box is in one of two states, and its border says which. A solid border means you are holding the box: dragging moves it, the handles resize it, the arrow keys nudge it, and typing replaces its text. A dashed border means you are inside the text: there is a caret, and the keys go to it.

Double-click, Enter or F2 goes from holding to typing; Esc or Cmd/Ctrl +Enter goes back, keeping what was typed. While you are typing, the border and the handles still belong to the box — dragging one leaves the text and moves or resizes the box in the same gesture — and so does a press on any other shape, which simply selects it.

Keyboard

How rendering works

The deck is unzipped in memory and each slide is walked into a small model, then rendered as absolutely-positioned DOM at the deck's native pixel size and scaled with a CSS transform — so zooming is free and text stays crisp and selectable.

The parser covers what real decks actually use: theme colour schemes and the lumMod/lumOff/tint/shade transforms, placeholder inheritance from slide to layout to master, list styles and auto-numbered bullets, gradients, images with cropping, tables with merged cells, groups, connectors, and SmartArt (via the plain-shapes fallback drawing PowerPoint stores alongside it).

Charts are drawn as SVG — column, bar, stacked and percent-stacked, line, area, pie, doughnut and scatter — from the values every chart part caches alongside its embedded workbook. That cache is what the deck was showing when it was saved, so no spreadsheet has to be opened to plot it. Series take their colours from the deck's theme accents unless the chart names its own, and a missing value stays a gap in the line rather than becoming a zero.

Extracting a deck to Markdown turns each chart into its own data table, since the numbers behind a chart are what someone would want to search for later.

Editing and safety

Text and shapes inherited from a layout or master are drawn but not editable — changing them would silently rewrite every slide built on that template.

Animations are written as the tree PowerPoint itself emits, because a timing tree that merely satisfies the schema is one PowerPoint refuses to open. A sequence holding an effect this plugin cannot write back is read and can be removed, but is never rebuilt — reordering it would mean dropping the effect it does not understand.

Editing a run keeps the runs around it intact, so typing inside a bold word leaves it bold. Only restructuring a paragraph — merging runs, deleting across them — collapses it onto the formatting of its first run.

Moving a placeholder that inherited its position writes a new a:xfrm onto the slide, which is what PowerPoint does too.

Undo and redo work on whole package parts rather than on individual elements. An edit that adds a part — inserting a picture, adding a slide — cannot be expressed as a change to an element, so recording parts is what lets one mechanism cover every command and guarantees undo lands on a state the parser has already accepted. The test suite asserts exactly that: after running all sixteen editor commands and undoing them, every part is byte-identical to where it started.

The first time you save a deck, a one-off copy is written beside it as <name>.pptx.bak. Binary files do not go through Obsidian's file recovery, so that copy is your undo of last resort.

If the file changed on disk since you opened it, the save is refused rather than overwriting someone else's work.

Why it stays responsive

An edit re-derives only the slide it touched. DeckEditor reports which package parts a command changed, and a change confined to one slide's part and its relationships rebuilds that slide alone; only a change to the presentation, a layout or a master rebuilds the deck. On a 27-slide deck that is 1.7ms instead of 61ms, and one slide re-rendering instead of all of them.

Rendered slides are cached per index and only the edited one is dropped. Thumbnails render lazily as they scroll into view, and refresh individually. The ribbon, selection pane, floating toolbar and status bar all refresh together in one animation frame rather than each reacting to every event.

                      whole deck    one slide    render one slide
21 slides, 702 shapes    30.0ms        0.58ms          4.5ms
27 slides, 233 shapes    60.6ms        1.67ms          6.1ms

npm run smoke -- --bench prints these for any deck (set PPTX_SMOKE_BENCH=1).

Development

npm install
npm run dev      # watch build
npm run build    # typecheck + production bundle
npm run lint     # Obsidian plugin rules over src
npm test         # unit tests for the parser's pure pieces
npm run smoke -- ~/Downloads/*.pptx   # parse + save round trip against real decks

Linting covers src only. That is the code that ships into Obsidian, and the rules worth enforcing — no Node built-ins, createEl over innerHTML — are about that runtime. The tests and build scripts run under Node, where the same rules only produce noise, and tsc already typechecks them. Errors fail the build; warnings are left visible rather than blocking.

The unit tests cover the layers the rest of the parser is built on — the XML helpers, EMU-to-pixel frames and preset geometry, and theme and colour resolution — where a mistake changes what a slide looks like without breaking anything loudly enough for a round trip to notice.

The smoke test runs the parser and the renderer under Node with jsdom, then drives the same code paths the UI does: it edits a run in the rendered DOM, deletes a paragraph, moves and resizes shapes, inserts text boxes, shapes, tables and pictures, groups and ungroups, reorders, aligns, formats, adds, duplicates, reorders and deletes slides — then undoes the lot and checks the package came back byte-for-byte.

Releasing

npm version patch --no-git-tag-version   # or minor / major

That moves package.json, and the version script carries manifest.json and versions.json along with it, so the three cannot drift apart. Commit the bump on a release/x.y.z branch and open a pull request.

Merging it publishes the release. The workflow reads the version out of manifest.json, notices it has not been released, checks that all three files agree, rebuilds, runs the unit and smoke tests again — the merge commit is not necessarily one CI has already seen — and attaches main.js, manifest.json and styles.css to a release named after the version. The tag is created by the release itself, against the commit that was built. There is nothing left to do by hand, and no window in which a tag exists without a release behind it.

Every other push to main ends at a job that finds the current version already released, and stops.

Pushing a tag still works, for releasing a commit that is not the tip of main:

git tag 0.1.5 <commit> && git push origin 0.1.5

Tags are the bare version with no v, because Obsidian looks for a release whose name matches the manifest.

Each of those files is attested, so a download can be checked against the workflow run that produced it:

gh attestation verify main.js --repo yut0takagi/obsidian-pptx-studio