TECH NEWS – There are many instances where artificial intelligence can be used for harmful purposes, but in this case it is even more so, because what Google researchers have achieved is impressive, but scary to think about.

The researchers have developed the technology to make someone talk to you in the form of a video using a single still image. One image. This can be used to read someone’s input, for example, or to replace lip movement to make it more accurate in another language. But this is at the stage where someone’s identity can be stolen and false information can be spread by someone’s AI replica. In other words, deepfakes could reach a new, unprecedented level.

Google researchers call the technology Vlogger, which can create facial and body movements in video. It can also be used for video editing. In one short video clip, the presenter talks into the camera, but in another version the mouth is closed, and in yet another version the eyes are closed. The artificially open-eyed version is a nightmare, and we are certainly not alone in this, as the lack of blinking is unnatural.

It works by using two stages. According to its Github page, “1) a stochastic human-to-3d-motion diffusion model, and 2) a novel diffusion-based architecture that augments text-to-image models with both temporal and spatial controls. This approach enables the generation of high-quality, variable-length videos that are easily controlled by high-level representations of human faces and bodies.” While this sounds exciting, the technology is far from perfect.

Like the AI-generated videos, the mouth movements seem strange and sometimes even fascinating. Nothing is perfect at first (we’re not perfect ourselves), so Google will improve on some aspects later, but it’s still great to see what kind of technology the company’s researchers can create when they really don’t have much to work with.

Source: PCGamer, Github

2. Generation of Moving and Talking People

Here’s an example on talking face generation given just a single input image and a driving audio. pic.twitter.com/hd7HKDfYkP

— EyeingAI (@EyeingAI) March 18, 2024