Project page for Veda: Scalable Video Diffusion via Distilled Sparse Attention (ICML 2026).
A static, dependency-free site (HTML + CSS + vanilla JS) with interactive charts, drag-to-compare videos, an animated speed-up demo, and an interactive architecture diagram.
Just open index.html in a browser, or serve the folder:
python3 -m http.server 8000 # then visit http://localhost:8000Fonts load from Google Fonts, so an internet connection is needed for the exact typography.
Push this repository to GitHub, then in Settings → Pages choose
Deploy from a branch → main / root. index.html lives at the repo root, so the
site is served directly.
index.html # the page
logo.svg # brand mark
static/
css/ # base.css + theme-light.css
js/ # veda.js (interactions), charts.js (SVG charts), media.js (manifest)
images/ # the few raster figures still used (head-diversity, social preview)
videos/ # web-encoded clips: comparisons / gallery / race / analysis
posters/ # poster frames (WebP)
scripts/ # local build/media tooling (not needed to serve the site)
The video sections are manifest-driven via static/js/media.js
(window.__VEDA_MEDIA__), generated by the pipeline below.
Source material lives outside the repo (git-ignored). To replace clips:
- Drop files into
_assets/:_assets/comparisons/<name>/full.mp4+veda.mp4(before/after pairs)_assets/gallery/<name>.mp4(sample clips)- optional
label.txt/<name>.txtfor captions
- Run the pipeline (needs
ffmpeg+cwebp):
./scripts/process_media.shIt compresses everything, generates posters, and rewrites static/js/media.js.
scripts/build_arch.js regenerates the inline 3-panel architecture SVG from the
source vector if it changes.
Page © the authors. Built for the Veda project.