科普讲解m0h @exploraX_2026-11-085:22

阿尔·卡彭纪录短片

播放
3.7万
点赞
549
评论
32
收藏
798

这部关于阿尔·卡彭的电影式纪录片以代码生成动画视频,使用了附带的旁白、照片和风格参考素材。

完整提示词2,636 字符

I want a cinematic motion video built entirely from code. No AI video generators and no stock footage. Every frame should be drawn with code (HTML canvas is fine) and rendered to an MP4.

INPUTS (attached):
- Script / voiceover: [attach audio file] -- line timings: [paste each line with its start time, or tell Claude to work them out from the audio]
- Photos of the real people/places/objects: [attached]
- Style references: [attached -- these show the look I want, match their feel]
- Format: [e.g. 21:9, 3840x1646, 30fps] -- length must match the voiceover exactly.

THE STYLE:
- The whole video is ONE continuous camera move over one big world: a map, a desk, a board, whatever fits the story. Nothing cuts and no slides change. The camera glides, pushes in on details, pulls back to travel, and dives in again.
- Motion must feel smooth and filmed. Every start and stop eases in and out. The camera never stops dead: during holds it keeps drifting slowly in the direction it's about to go. Add subtle handheld movement, motion blur on fast moves, film grain, a soft glow and a vignette.
- Real photos and documents drop onto the world as physical objects (prints, papers, clippings), placed where the story happens.

STORYTELLING AND READABILITY RULES (most important):
1. One idea per line of narration. Each line gets ONE hero on screen (a photo, a word or a document) and everything else dims back.
2. Every phrase appears on screen once, never repeated. If the narrator says it, either show it on a real document or type it on screen, never both.
3. No filler text. Body copy on documents should be soft grey lines that read as paper, not words people try to read.
4. Big, readable type: hero words very large, labels never tiny. If someone can't read it on a phone in one second, it's too small or there's too much.
5. Less is more. Leave out decorative clutter (random tape, stamps, memos, extra labels) that isn't part of the story.
6. Visual moments land exactly on the spoken word: a circle drawn as the word is said, a stamp slamming on the beat.

PROCESS:
- Before you build, write a beat table: time → narration → the one hero on screen → any on-screen text. Show it to me and ask me anything that's unclear.
- Take your time. Quality matters more than speed. Prioritise design, smooth movement and clarity over adding more stuff.
- Before the final render, check your own work: look at stills from every beat and frame sequences of the camera moves. Fix anything that's cluttered, hard to read, overlapping, or that jerks or stops abruptly.
- Deliver the final MP4 with my voiceover synced, in the exact format above.

出处:作者在原帖下的回复。提示词保留原文语言,未做改写。

需自备参考素材:原作除了文字提示词,还用了图片、视频、音频、文档或代码库作为输入,本站未收录这些素材。只复制提示词得不到同样的结果。

作者提到的工具

HTML Canvas

原帖:https://x.com/exploraX_/status/2119350502961889592

全部作品数据开源在 GitHub · awesome-opus-5.5-video。

模型归属以作者自述为准,本站未独立复现。播放量为 2026-11-10 的快照。作品版权归原作者,修改署名或下架请在 X 私信 @GoSailGlobal。

更多科普讲解案例