ideas

人需要获得海量信息然后才能在某些领域取得成就这个相互关系有强相关和弱相关的范畴

画大饼（永远兑现不了的承诺只兑现一点点）

自相矛盾（给一个有争议的话题或者让人群分化对立）

撒谎（一切都按照别人的愿望回答然而不去实现不去操作反向实现反向操作或者虚拟化电子化意识形态化）

video and audio needs to be analysised separately.

audio can be processed by chunks, splited tracks, while video can be itered frame by frame.

Predict emotion by means of graphics, motion, text, voice and swap context to achieve desirable effects.

Add random swap and funny pictures in addition to simple dictation.

You may group different kinds of content for specialized model training.

Syncing is hard. With database, it’s not.

Analyze the conversation with emotional/keyword frequency indicator, do some feature extraction.

Accelerate part of the video to be more expressive

Consider recording the media before real-time processing.

Scan the object via taobao streaming and make it dance.

Transplant lolita pictures to bilibili.

Share dialogs/info from soul/qq/wechat.

Repurpose a wide range of streaming platforms.

use ocr to filter out text info

find new title from danmaku or comments

我们都知道，视频主要由画面和音频组成。但还有一个元素同样包含了巨大的信息量 —— 字幕。结合现有的自然语言处理模型，我们便能实现对话抽取式的自动剪辑。PS: 软件图标是我妹妹Hannah设计滴~

学业之余我会陆续更新工具教程。