Skip to content

About

If you can prompt it, you can present it. Slides as single, offline HTML files: step-by-step reveals, LaTeX math and live interactive animations. Written by you or an AI and presented right in the browser.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

CapySlides

Presentations as single, offline HTML files in the ETH Zürich style: slides with step-by-step reveals, LaTeX formulas, reusable building blocks and interactive animations — opened by double-click, presented from the browser, submitted as one zip.

  • Self-contained: each presentation is a folder (deck + animations + images) that works offline and never depends on this repository.
  • Writes like HTML, presents like PowerPoint: steps per slide, slide navigator with live previews, remembered step per slide, animation overlays, automatic slide numbers, date and footer.
  • Hard to break: missing images or animations become labelled placeholders, and a ?dev mode plus tools/check.py list every problem with its slide and step.
  • AI-ready: AUTHORING.md and a CLAUDE.md in every project tell an AI assistant exactly how to write and check slides.

Showcase

Click through two real talks made with CapySlides, right in the browser, nothing to install: → / Space for the next step, D / A inside an animation.

EfficientNet: Rethinking Model Scaling for CNNs Decentralized Collective Learning for Self-managed Sharing Economies
An interactive animation slide of the EfficientNet talk The ring-city animation of the I-EPOS talk
Seminar talk, 21 slides, 5 of them interactive animations Seminar talk, 21 slides, 3 of them interactive animations

Their source is in showcase/ and shows how slides, steps, components and animations are written. They contain only material that may be published: own illustrations, freely licensed photos (credited on the slides) and charts redrawn from the papers' published numbers instead of the papers' figures.


Requirements

What Why
Chrome or Edge (any current version) Viewing and presenting; check.py / pack.py run it headless
Python 3.10+ (no packages needed) The tools in tools/
Internet Not needed (fonts and KaTeX are built into every deck)

Quick start

python tools/new.py MyTalk
  1. Create — new.py asks for title, subtitle, presenter, course, semester and (optionally) the names of animations, then creates presentations/MyTalk/.
  2. Write — open presentations/MyTalk/MyTalk_presentation.html in a text editor and edit the slides (see AUTHORING.md, or let an AI assistant do it — see below). Put images into images/.
  3. Look — open the same file in Chrome/Edge with ?dev at the end of the address (…/MyTalk_presentation.html?dev): problems are listed in ☰ Slides, all building blocks in ▦ Components.
  4. Check — python tools/image_sizes.py presentations/MyTalk/MyTalk_presentation.html after adding images, then python tools/check.py presentations/MyTalk (✓ no problems = ready).
  5. Present — open the file (without ?dev), press F11 for fullscreen, use → / ← (or a clicker).
  6. Submit — python tools/pack.py presentations/MyTalk writes presentations/MyTalk.zip (only the presentation files; verified to work on its own).

Another animation later: python tools/add_animation.py presentations/MyTalk "Model Scaling" creates the file and registers it in the deck.

Every command with all options: TOOLS.html (open it in the browser).

Presenting

Key / button Action
→ ↓ Space PageDown End next step, then next slide (clickers/laser pointers work)
← ↑ PageUp Home previous step / slide
D / A one step forward / back inside the animation that is visible (animation slide or open animation); the slides do not move
☰ Slides all slides with previews; click to jump (each slide remembers its step)
▶ Animation / ▶ Animations open the interactive animation(s); ◀ Presentation / ⇄ Switch to go back
Esc close a panel
F11 fullscreen — on your laptop the text then has exactly the sizes of the original deck

The slides are a fixed 1280 × 800 (16:10) canvas, scaled to the screen: they look identical on the laptop and on a mirrored beamer.

Working with an AI assistant

Open the project folder (presentations/MyTalk/) or this repository in Claude Code. The CLAUDE.md files tell the assistant where the guide is, which part of the file it may edit, and that it must run tools/check.py after every change. Useful requests:

  • "Create slides 3–8 for my talk on : motivation, method, results. Use the components."
  • "Add an image slide for images/results.png; reveal the red marker line on the second click."
  • "Turn this table into a data-table and highlight the rows of our model."
  • "Run check.py and fix every problem."
  • "Add a new animation panel that shows …" (in <name>_animations.html)

Updating to a newer template

The engine inside each deck has a version (engine/VERSION; the deck shows it in its DECK-ENGINE:BEGIN v… marker). After pulling a newer version of this repository:

python tools/update.py presentations/MyTalk

Only the generated engine blocks change (backups *.bak are written); your slides stay untouched. Within 1.x updates never change the meaning of existing attributes.

Publishing a talk in the showcase

Everything inside presentations/ (and docs/) stays on your machine: only the folders' README is in git, so your talks are never pushed by accident. To publish one as an example on the website:

python tools/showcase.py presentations/MyTalk --preview results

It copies only the files the talk uses to showcase/MyTalk/ (checked offline first; nothing is published if there are problems), takes a screenshot of the chosen slide (--preview: slide id or number, remembered) and rewrites showcase/index.html. The showcase copy is public, so it may only contain material you may publish: edit it in place (e.g. replace paper figures) and refresh it with python tools/showcase.py showcase/MyTalk; your own project stays as it is (overwriting the copy from it again needs --replace). Commit and push showcase/; the workflow .github/workflows/showcase.yml then publishes it with GitHub Pages. One-time setup on GitHub: Settings → Pages → Source: GitHub Actions.

Troubleshooting

Symptom Cause and fix
Text looks smaller than in my old deck The browser window is not fullscreen (toolbars take height, the 16:10 slide shrinks). Press F11.
White box image "x.png" not found The file is not in images/ (or the path/name differs). Fix the path, then run image_sizes.py.
White box animation "x.html" not found The file is missing, its name differs from DECK.animations, or it lost its ANIM-CONTRACT block. Start animations from template/template_animations.html.
A formula is red KaTeX cannot parse it; check.py shows the error. For a literal dollar sign write \$.
Content appears one click late / early Steps are 0-based: data-step="0" = with the title; unmarked content comes last (AUTHORING.md §5).
Arrow keys move a slider instead of the slide Inside animations arrow keys go to the presentation; hold Ctrl or Shift to move only the slider.
check.py: "no working Chrome/Edge found" Install Chrome, or set DECK_TEST_BROWSER to the browser's .exe. A browser in the middle of an update can fail headless — the tools then try the next one.
The zip shows broken images Extract the zip before opening the deck (browsers can't read files inside a zip).
Something in the engine block was edited by mistake python tools/update.py <folder> rewrites the engine blocks from the repository (also at the same version); your slides stay untouched.

Repository map

Path Content
presentations/ Your presentation projects (one self-contained folder each; only its README is committed)
docs/ Your own reference material (only its README is committed)
showcase/ Published example talks + index.html, served by GitHub Pages (written by showcase.py)
template/ The generated starter deck and starter animation (new.py copies them)
tools/ new, add_animation, check, image_sizes, update, pack, showcase, build, fetch_vendor + tests
engine/ Source of the engine (JS/CSS), components, starter shells, vendored fonts/KaTeX
.github/workflows/ showcase.yml: deploys showcase/ to GitHub Pages
AUTHORING.md How to write slides (for people and AI)
TOOLS.html Every command and in-browser helper
SPEC.md Design, phases and the decision log
CLAUDE.md Rules for AI agents working in this repository

Credits

  • Default title-slide photo (engine/assets/eth_zurich_main_building.jpg, the ETH Zurich main building): Josip Ivanković on Unsplash, used under the Unsplash License. New decks show this credit on the title slide (DECK.cover.credit).
  • Showcase EfficientNet, resolution slide: photo by Henrique Ferreira on Unsplash (Unsplash License).
  • Fonts: Inter, Fira Code and DM fonts (SIL Open Font License 1.1); math: KaTeX (MIT). Their licence files are in engine/vendor/.

Development

python tools/build.py                     # regenerate template/ and AUTHORING.md from engine/
python -m unittest discover tools/tests   # build, engine (headless browser) and tool tests

Change engine behaviour only in engine/, bump engine/VERSION, record decisions in SPEC.md §20.

About

If you can prompt it, you can present it. Slides as single, offline HTML files: step-by-step reveals, LaTeX math and live interactive animations. Written by you or an AI and presented right in the browser.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Contributors

Languages