Google announces video generation AI 'Imagen Video'



While image generation AI such as “

Stable Diffusion ” has become a hot topic, video generation AI such as “ Make A Video ” and “ Phenaki ” are also appearing one after another. Newly, Google has announced 'Imagen Video,' which generates videos from natural language instructions such as 'a teddy bear washing dishes.'

Imagen Video
https://imagen.research.google/video/

In May 2022, Google announced AI `` Imagen '' that can automatically generate high-precision images from outlandish text.

AI system 'Imagen' that can automatically generate high-precision images even from outlandish text - GIGAZINE



And Google has released 'Imagen Video' that can generate a video of about 5 seconds instead of an image this time. You can see what kind of video is generated from the following.

Demo movie of Google's video generation AI 'Imagen Video' - YouTube


The video was generated by a so-called 'spell' text prompt, 'a teddy bear washing dishes'. The teddy bear's hands and the plate may bend like clay, but that gives the impression of clay animation. Also, the expression of running water is a point.



Imagen Video first processes the input text prompt with natural language processing AI '

T5 '. Next, it generates 16-frame video at 3 frames per second at 24 x 48 resolution based on ' Video Diffusion Models ' that generate video with a diffusion model . Then, upsampling this with the model 'Temporal Super-Resolution' and 'Spatial Super-Resolution', finally 1280 × 768 resolution and 24 frames per second generates 128 frames, or approximately 5.3 seconds of video.



Various other videos generated by Imagen Video are posted on Imagen Video's official website and SNS.







Related Posts:

in Software,   Video, Posted by log1l_ks