Omnihuman-1 can make realistic videos from just one image

TikTok's parent Bytedance has a new AI model, Omnihuman-1, which can creates lifelike videos from a single image and audio. Trained on 19,000 hours of footage, it combines images, audio, and body poses for ultra-realistic results, including controlling specific body parts and being able to create cartoons. Videos of Taylor Swift singing in Japanese and an Einstein TEDx talk made using Omnihuman have gone viral.