ByteDance's unified multimodal architecture accepts text, image, audio, and video as inputs. Up to 15-second clips at 2K resolution with extreme character consistency and audio-visual beat matching. For creators, the multimodal input is the breakthrough: feed the model a reference image, a voice clip, and a text description simultaneously, and it synthesizes all three into coherent video. Then copyright disputes from major studios pause the global rollout, raising questions about the training data behind the magic.