File size: 7,525 Bytes
4777430 a5e5746 4777430 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 | <!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<meta name="theme-color" content="#111827">
<meta name="description" content="See how sampled video frames expand into vision placeholders in MiniMax M3.1 Flash, using its pinned public processor. Built by alphaBell.">
<meta property="og:title" content="MiniMax Video Budget | alphaBell">
<meta property="og:description" content="A source-pinned visual calculator for MiniMax M3.1 Flash video placeholders.">
<meta property="og:image" content="https://huggingface.co/spaces/alphabell/MiniMax-Video-Budget/resolve/main/cover.png">
<title>MiniMax Video Budget | alphaBell</title>
<link rel="icon" href="favicon.svg">
<link rel="stylesheet" href="style.css">
<script src="budget.js" defer></script>
</head>
<body>
<header class="topbar"><a class="brand" href="https://alphabell.com" aria-label="alphaBell home"><span class="mark">α</span><span>alphaBell <i>/</i> video budget</span></a><a class="top-link" href="https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash" target="_blank" rel="noopener noreferrer">Publisher model ↗</a></header>
<main>
<section class="hero" aria-labelledby="title"><div class="hero-copy"><p class="eyebrow">Source-pinned model lens <span>01 / MiniMax M3.1 Flash</span></p><h1 id="title">One video.<br><em>Many visual slots.</em></h1><p class="lead">MiniMax M3.1 Flash groups sampled frames in pairs, then expands each group into visual placeholders. Enter the frame count and <strong>processed</strong> frame dimensions to see the exact arithmetic.</p><div class="hero-rule"><span class="pulse" aria-hidden="true"></span><span>Processor arithmetic, not an inference benchmark</span></div></div><div class="hero-visual" aria-hidden="true"><div class="filmstrip"><span>01</span><span>02</span><span>03</span><span>04</span></div><div class="connector">↓ two-frame groups</div><div class="pixel-grid"><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i><i></i></div><div class="visual-label">Frames → groups → placeholders</div></div></section>
<section class="workspace" aria-labelledby="tool-title"><div class="section-head"><p class="eyebrow">01 / Try a video shape</p><h2 id="tool-title">Count the visual placeholders</h2><p>Use the number of frames <em>after sampling</em> and the height and width <em>after resizing</em>. Dimensions must be multiples of 28 pixels.</p></div><div class="tool-grid"><form id="budget-form" novalidate><div class="input-grid"><label for="frames">Sampled frames <small>1 to 240</small><input id="frames" name="frames" type="number" min="1" max="240" step="1" value="16" inputmode="numeric"></label><label for="height">Processed height <small>pixels, step 28</small><input id="height" name="height" type="number" min="28" max="2016" step="28" value="672" inputmode="numeric"></label><label for="width">Processed width <small>pixels, step 28</small><input id="width" name="width" type="number" min="28" max="2016" step="28" value="672" inputmode="numeric"></label></div><p class="field-error" id="error" role="alert"></p><div class="presets" role="group" aria-label="Example frame counts"><span>Try:</span><button type="button" data-preset="1">1 frame</button><button type="button" data-preset="3">3 frames</button><button type="button" data-preset="16">16 frames</button></div><p class="field-note" id="range-warning" hidden>These dimensions are outside the processor's default video resize area of 3,136 to 602,112 pixels. Check the actual processed dimensions before using this result.</p></form><div id="result" class="result" aria-live="polite"><div class="result-label">Vision placeholders</div><div class="big-number" id="total">4,608</div><div class="result-equation"><span id="group-count">8</span> two-frame groups <b>×</b> <span id="per-group">576</span> per group</div><div class="result-facts"><div><span>Spatial patch grid</span><strong id="grid">48 × 48 patches, merged 2 × 2</strong></div><div><span>Odd-frame padding</span><strong id="padding">No frame is repeated.</strong></div></div><div class="actions"><button id="copy" type="button">Copy view link</button><button id="export" type="button">Download JSON</button></div><p id="status" class="action-status" role="status"></p></div></div></section>
<section class="timeline" aria-labelledby="timeline-title"><div class="section-head"><p class="eyebrow">02 / Temporal groups</p><h2 id="timeline-title">Where the frames go</h2><p>An odd final frame is repeated to complete the last pair. Each group receives the same placeholder count when processed dimensions are constant.</p></div><ol id="groups" class="groups" aria-label="Two-frame video groups"></ol><p id="group-overflow" class="group-overflow"></p></section>
<section class="explanation" aria-labelledby="math-title"><div class="section-head"><p class="eyebrow">03 / How it is calculated</p><h2 id="math-title">Three factors, one count</h2></div><div class="formula"><span>ceil(sampled frames ÷ 2)</span><b>×</b><span>(processed height ÷ 28)</span><b>×</b><span>(processed width ÷ 28)</span></div><div class="explain-grid"><article><strong>01</strong><h3>Pair the frames</h3><p>The processor pads an odd frame count by repeating its last frame, then groups frames in twos.</p></article><article><strong>02</strong><h3>Merge spatial patches</h3><p>Each 14-pixel patch grid is merged 2 by 2. One placeholder covers a 28 by 28 processed-pixel region.</p></article><article><strong>03</strong><h3>Count placeholders</h3><p>The text processor repeats a vision placeholder for every merged spatial location in every temporal group.</p></article></div><div class="scope"><strong>What this number covers</strong><p>This counts the repeated vision placeholders in the pinned MiniMax processor. It excludes timestamps, vision wrapper tokens, text, total prompt tokens, automatic frame sampling, resize decisions, model memory, speed and visual understanding. It does not require model weights or upload a video.</p></div></section>
<section class="sources" aria-labelledby="sources-title"><div class="section-head"><p class="eyebrow">Pinned publisher sources</p><h2 id="sources-title">Inspect the implementation</h2><p>All source links point to MiniMax M3.1 Flash revision <code>262ae6830934f3512211081b8a3f16425b93a239</code>.</p></div><div class="source-grid"><a href="https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash/blob/262ae6830934f3512211081b8a3f16425b93a239/video_processor.py#L122" target="_blank" rel="noopener noreferrer"><span>Video processor</span><strong>Frame padding and grids ↗</strong></a><a href="https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash/blob/262ae6830934f3512211081b8a3f16425b93a239/processing_minimax.py#L209" target="_blank" rel="noopener noreferrer"><span>Text processor</span><strong>Placeholder expansion ↗</strong></a><a href="https://huggingface.co/MiniMaxAI/MiniMax-M3.1-Flash/blob/262ae6830934f3512211081b8a3f16425b93a239/README.md" target="_blank" rel="noopener noreferrer"><span>Model card</span><strong>MiniMax M3.1 Flash ↗</strong></a></div></section>
</main>
<footer><span>Built by <a href="https://alphabell.com">alphaBell</a></span><span>Independent processor visualization. Model and processor by MiniMax.</span><a href="https://huggingface.co/alphabell" target="_blank" rel="noopener noreferrer">More lab projects ↗</a></footer>
</body>
</html>
|