Imagen Video is a text-to-video artificial intelligence mode that can produce videos at 24 frames per second. It's currently in a research phase, five months after it was mentioned that video synthesis models were being developed quickly.
Six months after the launch of Openai's DALLE-2 text-to-image generator, progress in the field of artificial intelligence has been heating up. Less than a week ago, Meta unveiled its text-to-video tool, Make-A-Video
In addition to generating videos based on the work of famous painters, Imagen Video can also generate 3D rotating objects while preserving object structure, and render text in a variety of animation styles. General-purpose video models can help decrease the difficulty of high-quality content generation, according to the company.
The key to Imagen Video's abilities is acade of seven diffusion models that transform the initial text prompt into a low-resolution video. The final video is a little over five minutes in length.
AdvertisementThere are a variety of video examples on the website, from the mundane to the more amazing. There are obvious artifacts, but they show more detail than previous text-to- image models.
Today, another text-to-video model was officially introduced. It's called Phenaki and it can create long videos. With the number of papers on arXiv growing rapidly, it's difficult for some researchers to keep up with the latest developments.
The LAION-400M image-text dataset contains 14 million video-text pairs and 60 million image-text pairs. It can contain sexually explicit and violent content, as well as social stereotypes and cultural biases, even though it has been trained on problematic data. The firm is worried that the tool may be used to create fake, harmful or explicit content.
"We have decided not to release the imagen video model or its source code until these concerns are mitigated."
All Rights Reserved © 2024
