TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

CogVideo: Large-Scale Pretraining for Text-to-Video Generation via Transformers

9 点作者 lnyan大约 3 年前

1 comment

true_religion大约 3 年前
It’s an interesting demo, but I think the tech generates each frame individually only using previous frames and the text as guidance.<p>This means that every face in the clips is of a similar but not quite exactly the same person. For example the eye color of the subject can change, hair can become fuller, and in one case the subject became noticeably better looking as facial symmetry improved over the course of the generation.