# 구글, 텍스트-to-비디오 AI "Imagen Video" 공개

> Clean Markdown view of GeekNews topic #7555. Use the original source for factual precision when an external source URL is present.

## Metadata

- GeekNews HTML: [https://news.hada.io/topic?id=7555](https://news.hada.io/topic?id=7555)
- GeekNews Markdown: [https://news.hada.io/topic/7555.md](https://news.hada.io/topic/7555.md)
- Type: news
- Author: [xguru](https://news.hada.io/@xguru)
- Published: 2022-10-07T10:52:01+09:00
- Updated: 2022-10-07T10:52:01+09:00
- Original source: [imagen.research.google](https://imagen.research.google/video/)
- Points: 9
- Comments: 1

## Topic Body

- Video Diffusion Model로 텍스트를 입력받아 동영상을 만드는 "Text-conditional Video Generation System"  
- 텍스트에서 저해상도 비디오(24x48 픽셀, 16프레임, 3fps) 영상을 만들고, 디퓨전 모델을 7개 중첩(cascade)시켜서 업스케일 하는 것이 특징  
- 최종 출력은 1280x768 24fps . 5.3초 길이의 비디오를 생성 가능  
- 논문 : [Imagen Video : High Definition Video Generation with Diffusion Models](https://imagen.research.google/video/paper.pdf)

## Comments



### Comment 12641

- Author: xguru
- Created: 2022-10-07T10:53:02+09:00
- Points: 1

[Imagen - 구글의 text-to-image diffusion model](https://news.hada.io/topic?id=6609)  
[Imagen-pytorch - 구글 Imagen을 Pytorch로 구현](https://news.hada.io/topic?id=6627)  
[Make-A-Video : 텍스트로 비디오를 생성하는 AI](https://news.hada.io/topic?id=7531)