ai21.net
语言

I used Opus 5.5 to generate 50 podcast animations for extr…

演示视频

I used Opus 5.5 to generate 50 podcast animations for extr… — 1

提示词

I used Opus 5.5 to generate 50 podcast animations for extremely hard to visualize concepts like Jev for Legal.  Absolutely blown away. 

We're about to see an explosion of visual podcasts, this entire medium is about to shift 

Sharing my prompt below.

【作者补充制作说明】
Go through the entire [episode] cut and add full-screen animations and graphics wherever they help. Be aggressive and over-inclusive: 35–50 moments, at least 2 per chapter. The goal is to fill empty talking-head time, explain ideas, and create visual hooks.

Style. Pick an aesthetic that's genuinely interesting and very, very cool, built in JavaScript with HyperFrames and assembled in Tesseract. Inspiration (all JS, no image assets):

https://x.com/Voxyz_ai/status/2102531681450119426: paper, typewriter-tape labels, hand-drawn marks
https://x.com/donaldjewkes/status/2102801274173587569: halftone, manga and pixel print, kinetic data type

Stay inside the LegalQuants brand (read brands/legalquants/): charcoal, white and black only, Arial, and the supplied lockup top-right, static. Write a shared style bible and a component kit first, so every animation looks like the same show.

Moments I definitely want:

[LIST]

Rules

Whenever anything goes full-screen (animation, screen share or still), one of us is in a corner picture-in-picture, bottom-right, following whoever is talking.
No punch-ins or zooming on the cameras. Keep each speaker relatively centered.
Copy must be verbatim or faithful to what's said. Never invent facts, numbers or quotes.
Sync each animation's beats to the spoken words.
Keep each graphic inside one uncut stretch of audio, and leave a few seconds of faces between graphics.

Process

An Opus 5.5 agent at extra-high effort finds every timestamp where an animation could fill space and hook the viewer, with a concept for each.
Opus 5.5 at medium effort builds each animation.
Opus 5.5 at extra-high effort judges each one once on brand consistency with the others, visual interest, relevance to the narrative and creative judgment.
Only cues the judge considers egregiously bad go back for a fix. No blanket redo loops.
Assemble the full cut with PiP baked into every full-screen visual, and give me a folder of the individual animations.

中文译文

我用 Opus 5.5 为极难可视化的概念(例如 Jev for Legal)生成了 50 个播客动画。彻底被震撼到了。

我们即将看到视觉播客的爆发,整个媒介即将转变

下面分享我的提示词。

【作者补充制作说明】
通览整个 [episode] 剪辑,在任何有帮助的地方加入全屏动画和图形。要激进、宁多勿缺:35–50 个时刻,每章至少 2 个。目标是填满空白的说话人头时间、解释想法,并制造视觉钩子。

风格。选择一种真正有趣、非常非常酷的美学,用 JavaScript 配合 HyperFrames 构建,并在 Tesseract 中组装。灵感(全部用 JS,无图像素材):

https://x.com/Voxyz_ai/status/2102531681450119426:纸张、打字机胶带标签、手绘标记
https://x.com/donaldjewkes/status/2102801274173587569:半调网点、漫画与像素印刷、动态数据字体

保持在 LegalQuants 品牌规范内(读取 brands/legalquants/):仅炭灰、白和黑,Arial,以及右上角提供的组合标识,保持静态。先写一份共享的风格圣经和一套组件库,让每个动画看起来都像同一档节目。

我确定想要的时刻:

[LIST]

规则

每当任何内容进入全屏(动画、屏幕共享或静帧),我们其中一人就在角落画中画出现,位置在右下角,跟随正在说话的人。
不要对摄像机做推近或变焦。让每位说话者保持相对居中。
文案必须逐字忠实于所说的话。绝不虚构事实、数字或引述。
将每个动画的节拍与所说的话同步。
让每个图形保持在一段未剪辑的音频之内,并在图形之间留出几秒的人脸画面。

流程

一个 Opus 5.5 智能体以超高努力找出每一个可以用动画填充空间并钩住观众的时点,并为每个时点给出一个概念。
Opus 5.5 以中等努力构建每个动画。
Opus 5.5 以超高努力对每个动画评判一次,评判标准是与其他动画的品牌一致性、视觉趣味、与叙事的关联以及创意判断。
只有评委认为糟糕到极点的动画片段才退回修改。不做全盘重做的循环。
组装完整剪辑,把画中画烘焙进每一个全屏视觉,并给我一个包含各个独立动画的文件夹。

溯源

原作者
stokebuilder
发布日期
2026-09-26(由推文 ID 反解)
原始出处
查看原帖 ↗
收录来源
yihui-dev/awesome-opus5-5-videos · stokebuilder-356793 · MIT