Google presents two artificial intelligence text-to-video systems

0 30


Google presents two artificial intelligence text-to-video systems

Recently, Meta shared a new AI model that makes text prompts turn into short silent videos. Google also had a similar idea and recently released two new artificial intelligence (AI) text-to-video systems. The first is called Imagen Video and focuses on image quality. While the second is called Phenaki and focuses more on creating longer clips.

For security reasons, and more specifically so that the systems are not used for malicious purposes, Google has decided to keep the two systems private. This is also what Meta has done with its AI Make-A-Video. In addition, to restrict these malicious uses, researchers at Google explained that they experimented with filters to intercept NSFW prompts and produce videos but they did not clearly say if it really worked. According to them, ” several significant safety and ethical challenges remain “.


Google

Imagen Video

There is an old text-to-image system called Google Imagen. The high-quality model, Imagen Video, takes its name from this old system. Thus, Imagen Video relies on advanced techniques from Google Imagen, new components to the pipeline to transform static images into fluid motion.

Regarding quality, according to Google researchers, this first model produces 16 frames of 3fps footage at 24×48 resolution. This low-res content is then passed through various AI super-resolution models, which increase this output to 128 frames of 24fps footage at 1280×768 resolution. Meta’s Make-A-Video template is 768×768.

Phenaki

Phenaki is different from the first text-in-video model, Imagen Video. It deals with creating longer videos that follow detailed prompt instructions. That is to say, you include a long descriptive text and Phenaki will take care of generating a long video for you. Researchers say the length of videos is limitless.

These researchers state that although the quality of videos generated by Phenaki is not yet indistinguishable from actual videos, reaching this bar for a specific set of samples is within the realm of possibility even today “. Only, it can be dangerous if Phenaki is used for the purpose of producing videos of someone without their knowledge and consent.

source

Leave A Reply

Your email address will not be published.