fix format

Author: RuilliFeng

.DS_Store

@@ -8,14 +8,11 @@
   <meta property="og:title" content="Incantation: Natural Language as the Action Interface for Multi-Entity Video World Models" />
   <meta property="og:description" content="Natural language beats closed action IDs as the action interface: 89% vs. 43% cross-entity transfer, 90% vs. 0% out-of-vocabulary prompt control, and 3-entity control from two-entity training." />
   <meta property="og:type" content="website" />
-  <meta property="og:image" content="static/images/figs/teaser.png" />
+  <meta property="og:image" content="static/images/figs/teaser.jpg" />
   <meta name="twitter:card" content="summary_large_image" />
   <meta name="citation_title" content="Incantation: Natural Language as the Action Interface for Multi-Entity Video World Models" />
   <meta name="citation_pdf_url" content="static/paper/incantation.pdf" />
 
-  <link rel="preconnect" href="https://fonts.googleapis.com" />
-  <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
-  <link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700;800&amp;family=JetBrains+Mono:wght@400;600;700&amp;family=Newsreader:opsz,wght@6..72,500;6..72,650&amp;display=swap" rel="stylesheet" />
   <link rel="stylesheet" href="static/css/main.css" />
   <link rel="icon" href="data:image/svg+xml;utf8,<svg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 64 64'><rect width='64' height='64' rx='12' fill='%23171614'/><path d='M15 41c13-25 21-25 34 0' fill='none' stroke='%238de0d5' stroke-width='7' stroke-linecap='round'/><circle cx='21' cy='24' r='5' fill='%23c45a1b'/><circle cx='43' cy='24' r='5' fill='%23f2efe7'/></svg>" />
 </head>
@@ -33,7 +30,7 @@
         <a href="#bibtex">Cite</a>
       </nav>
       <a class="team-chip" href="https://matrixteam-ai.github.io/" target="_blank" rel="noopener" title="Visit Matrix Team">
-        <img src="static/images/logos/official/matrix-white.svg" alt="Matrix Team" />
+        <img src="static/images/logos/official/matrix-white.svg" alt="Matrix Team" width="18" height="18" />
         <span class="text">Matrix Team</span>
       </a>
     </div>
@@ -53,7 +50,7 @@ <h1><span class="accent">Incantation</span></h1>
 
         <a class="presented-by" href="https://matrixteam-ai.github.io/" target="_blank" rel="noopener">
           <span class="pb-logo">
-            <img src="static/images/logos/official/matrix-white.svg" alt="" />
+            <img src="static/images/logos/official/matrix-white.svg" alt="" width="14" height="14" />
           </span>
           <span>Presented by <strong>Matrix Team</strong> — Neural Interactive Simulation</span>
         </a>
@@ -90,7 +87,7 @@ <h1><span class="accent">Incantation</span></h1>
 
         <div class="hero-teaser">
           <div class="figure-shell">
-            <img src="static/images/figs/teaser.png" alt="Incantation cross-entity transfer and multi-entity control teaser" />
+            <img src="static/images/figs/teaser.jpg" alt="Incantation cross-entity transfer and multi-entity control teaser" width="1659" height="713" decoding="async" fetchpriority="high" />
           </div>
           <p class="caption">
             <strong>Cross-entity action transfer and multi-entity control in Elden Ring.</strong>
@@ -199,7 +196,7 @@ <h3>Sliding cache without positional drift</h3>
             </article>
           </div>
           <div class="figure-shell dark-figure">
-            <img src="static/images/figs/workflow.png" alt="Incantation workflow with language-conditioned pretraining and Self-Forcing distillation" />
+            <img src="static/images/figs/workflow.jpg" alt="Incantation workflow with language-conditioned pretraining and Self-Forcing distillation" width="1314" height="806" loading="lazy" decoding="async" />
             <p class="caption">Training and streaming workflow: language-conditioned pretraining followed by ordinary-differential-equation-initialized Self-Forcing distillation.</p>
           </div>
         </div>
@@ -259,7 +256,7 @@ <h2>Attention that respects time</h2>
         <p class="section-sub">Each action prompt describes the current frame, so Incantation prevents that prompt from contaminating committed history frames.</p>
         <div class="asset-grid">
           <div class="figure-shell">
-            <img src="static/images/figs/masked-attention.png" alt="Decoupled text cross-attention restricted to the noisy target frame" />
+            <img src="static/images/figs/masked-attention.png" alt="Decoupled text cross-attention restricted to the noisy target frame" width="1175" height="419" loading="lazy" decoding="async" />
             <p class="caption">Text cross-attention is applied only to the noisy target frame; history frames keep bidirectional self-attention.</p>
           </div>
           <div class="text-panel">
@@ -394,7 +391,7 @@ <h3>Evidence: 3-entity control from 2-entity training</h3>
         </div>
 
         <div class="figure-shell" style="margin-top:34px;">
-          <img src="static/images/figs/baseline-comparison.png" alt="Qualitative comparison against Seedance, Kling, LongLive, and Incantation" />
+          <img src="static/images/figs/baseline-comparison.jpg" alt="Qualitative comparison against Seedance, Kling, LongLive, and Incantation" width="1373" height="800" loading="lazy" decoding="async" />
           <p class="caption">Qualitative comparison on Elden Ring. Strong video generators preserve visual fidelity, but Incantation is the method designed for per-frame player-boss action control.</p>
         </div>
 
@@ -435,11 +432,11 @@ <h2>Cross-world replication</h2>
         <p class="section-sub">The same architecture and recipe are applied to Elden Ring and the visually unrelated King of Fighters world, changing only action-vocabulary slots. Real-time streaming is an enabling system property, not the core interface claim.</p>
         <div class="asset-grid">
           <div class="figure-shell">
-            <img src="static/images/figs/margit-rollout.png" alt="Elden Ring long-horizon generated rollout" />
+            <img src="static/images/figs/margit-rollout.jpg" alt="Elden Ring long-horizon generated rollout" width="788" height="810" loading="lazy" decoding="async" />
             <p class="caption">Elden Ring rollout sampled from a continuous generated session.</p>
           </div>
           <div class="figure-shell">
-            <img src="static/images/figs/kof-rollout.png" alt="King of Fighters generated rollout under the same architecture" />
+            <img src="static/images/figs/kof-rollout.jpg" alt="King of Fighters generated rollout under the same architecture" width="788" height="723" loading="lazy" decoding="async" />
             <p class="caption">King of Fighters rollout under the same architecture and training recipe.</p>
           </div>
         </div>
@@ -536,7 +533,7 @@ <h3>0.25 s</h3>
 
         <div class="asset-grid" style="margin-top:30px;">
           <div class="figure-shell">
-            <img src="static/images/figs/annotator.png" alt="Local annotation interface for Incantation" />
+            <img src="static/images/figs/annotator.jpg" alt="Local annotation interface for Incantation" width="1756" height="960" loading="lazy" decoding="async" />
             <p class="caption">Blinded action-control accuracy interface. Annotators see the generated clip and per-entity target action, but not whether it came from natural language or the Action-ID baseline.</p>
           </div>
           <div class="text-panel">

pages/.DS_Store

@@ -18,9 +18,9 @@
   --max: 1120px;
   --radius: 8px;
   --shadow: 0 18px 55px rgba(30, 23, 12, 0.12);
-  --sans: "Inter", -apple-system, BlinkMacSystemFont, "Segoe UI", Arial, sans-serif;
-  --serif: "Newsreader", Georgia, "Times New Roman", serif;
-  --mono: "JetBrains Mono", ui-monospace, SFMono-Regular, Menlo, monospace;
+  --sans: -apple-system, BlinkMacSystemFont, "Segoe UI", Arial, sans-serif;
+  --serif: Georgia, "Times New Roman", serif;
+  --mono: ui-monospace, SFMono-Regular, Menlo, Consolas, monospace;
 }
 
 * { box-sizing: border-box; }