Animate one LLM request as code-drawn frames—no After Effects

ماذا تثبّت أو تنشئ

Prompt pins goal, 1920×1080 ~18s 30fps, hand-drawn style, architecture TOKENIZE→SCHEDULE→PREFILL→DECODE→STREAM with KV as FILL ONCE then READ+APPEND loop, full overview always visible, renderAt(t) with no clocks/random.

النص التوجيهي

1. Goal: animate one LLM request through a serving pipeline so a viewer gets the path in one glance.
2. Canvas: 1920×1080, landscape, ~18s, 30fps, H.264.
3. Style: hand-drawn — cream paper, ink outlines, marker washes, monospace labels.
4. Architecture: TOKENIZE → SCHEDULE → PREFILL → DECODE → STREAM → RESPONSE. KV CACHE is NOT a linear step — Prefill FILL ONCE into KV; Decode ⇄ KV (READ + APPEND) until stop.
5. Always show the full overview. Pulse the active node; move packets along every edge (main chain + KV loop). Animate GPU util + KV occupancy with the request.
6. Render contract: every frame is a pure function of t (renderAt(t)); no clocks/random; pin stage timings to frame numbers; then export.

التشغيل الأول

1. 1Read https://x.com/ElioJackee/status/2109253928370221199 (credit @ElioJackee).
2. 2Paste the author's full prompt into a new bot.
3. 3Let it generate the renderAt(t) code and export H.264.
4. 4Watch once for timing glitches; ask for frame-pinned fixes.
5. 5Credit the author if you publish the clip.

ما الذي تعدّله

Pin stage timings to frame numbers so the render is reproducible.

المصدر

WonderThoughts · AI @ElioJackee · المنشور الأصلي

غير تابع لـ xAI. لم نكتب هذا النص التوجيهي. تبقى شروط القوالب التابعة لطرف ثالث سارية.