/* MMDiff project page — top sections: Nav, Hero, Abstract */ const DS = window.MMDiffDesignSystem_07b833; const { Button, Badge, Kicker, StatBlock, SpectrumBar, Card } = DS; const I = window.Icons; const D = window.MMD; const iconFor = { paper: I.paper, arxiv: I.arxiv, github: I.github, data: I.data, hf: I.hf }; function NavBar() { const [scrolled, setScrolled] = React.useState(false); React.useEffect(() => { const onScroll = () => setScrolled(window.scrollY > 24); window.addEventListener('scroll', onScroll, { passive: true }); return () => window.removeEventListener('scroll', onScroll); }, []); const go = (id) => (e) => { e.preventDefault(); const el = document.getElementById(id); if (el) window.scrollTo({ top: el.offsetTop - 64, behavior: 'smooth' }); }; return (

MMDiff

); } function Hero() { return (

{D.meta.venue}

Extending diffusion transformers for multi-modal generation

{D.meta.authors.map((a, i) => ( {a.name} ^{a.sup} {i < D.meta.authors.length - 1 ? · : null} ))}

¹{D.meta.affiliation}

{D.meta.links.map((l) => { const dis = !!l.soon || (l.gated && !D.meta.released); return ( ); })}

{D.meta.released ? 'Paper \u0026 arXiv links available after release' : 'Links available after release'}

{/* Teaser figure on the lab surface */}

MMDiff teaser: generated images with their decoded saliency, segmentation and depth — MMDiff gets more out of one frozen image generator: in a single pass it creates an image and reads several detailed maps, perfectly aligned to it.

); } function Abstract() { return (

The idea

{D.idea.lead}

{D.idea.body}

{D.stats.map((s) => )}

); } Object.assign(window, { NavBar, Hero, Abstract });