ByteDance's OmniHuman-1 may be the most realistic deepfake algorithm yet

ByteDance's OmniHuman-1 produces almost lifelike deepfakes, raising ethical concerns.

: ByteDance introduced OmniHuman-1, an AI producing realistic deepfake videos using a single photo and audio clip. Despite minor imperfections, its results surpass previous deepfake models. OmniHuman-1 is trained with 18,700 hours of diverse data, optimizing learning efficiency. The technology's potential for misuse prompts calls for stricter regulations.

ByteDance has launched OmniHuman-1, an AI system that generates deepfake videos nearly indistinguishable from reality, relying on just a single photo and audio to create them. Users can modify aspects of the videos, such as the aspect ratio and body movements, with results being incredibly lifelike, yet not entirely flawless.

OmniHuman-1 was developed using 18,700 hours of video data, employing a unique 'omni-conditions' approach that allows it to learn from multiple sources like text prompts, audio, and body poses. This method leads to more efficient data usage compared to older deepfake models.

The powerful potential of OmniHuman-1 raises concerns about misuse for misinformation, fraud, and other harmful activities, as seen in past incidents and scams. Despite calls from AI ethics experts for regulations, laws on malicious deepfakes vary across states, with California's potential legislation still pending.

deepfakeByteDanceAI technology