According to TechSpot , ByteDance - the Chinese technology corporation behind TikTok, has just announced OmniHuman-1 - a deepfake system capable of recreating human images with incredible realism. Videos created by OmniHuman-1 can fool ordinary viewers, blurring the line between real and fake in digital content.
Many of the videos created by OmniHuman-1 include Albert Einstein, Jensen Huang, and many others, demonstrating how incredibly realistic deepfakes can be.
PHOTO: OMNIHUMAN-1 SCREENSHOT
Unlike previous deepfake models, OmniHuman-1 only needs a photo and a sound clip to create a complete video. Users can adjust the aspect ratio, body composition, or even edit the movements in the existing video. This AI can change hand movements and gestures with high precision, making the product look more natural. However, some details are still imperfect, such as the unnatural position of holding a wine glass or the strange hand rotation in a video simulating Albert Einstein's lecture.
The model was trained on 18,700 hours of video, using an “omni-conditions” approach to learn from a variety of data sources, including text, audio, and body movements.
Despite its technological advances, deepfake technology also poses serious risks. In recent years, deepfakes have been used to manipulate information, commit financial fraud, and attack personal reputations. During the 2024 election, many deepfake videos were used to mislead voters. Last year, financial fraud involving deepfakes resulted in billions of dollars being stolen, including one case in which a fraudster impersonated actor Brad Pitt to trick a woman into transferring $850,000.
Concerned about these risks, many AI ethics experts have called for deepfake regulation. Several US states have enacted laws against harmful deepfakes, but there is no comprehensive federal framework. California has proposed a bill that would allow courts to force deepfakes to be removed and punish those who distribute them, but the legislative process has encountered many obstacles.
At present, ByteDance has not announced plans to release OmniHuman-1, but a research paper on this model has been published, showing the potential and risks of the growing deepfake technology.
Source: https://thanhnien.vn/cong-cu-ai-deepfake-moi-cua-bytedance-gay-nhieu-lo-ngai-18525020521585617.htm
Comment (0)