The news: ByteDance's OmniHuman-1 model converts photos into full-body realistic videos with speech, gestures, and movement. Video-generation breakthroughs could transform digital media and marketing but also raise concerns about the proliferation of deepfakes and misinformation, per VentureBeat.
Examples include a video of Taylor Swift singing the “Naruto” theme in Japanese, a fake TED Talk, and a deepfaked lecture by Albert Einstein. OmniHuman-1 is not yet publicly available.
The future of AI content creation: OmniHuman-1 was built on 18,700 hours of training data and can theoretically generate videos of unlimited length, constrained only by available memory.