On the 16th local time, Google's "Deep Thinking" company announced the launch of a new generation of artificial intelligence video generation model Veo 2, which greatly improves video resolution, image quality, and simulates real-world physical laws.
The company stated that, Veo 2 can create 4K resolution, 2-minute videos, significantly ahead of Sora Turbo, the latest video generation model released by OpenAI in the United States. According to the introduction, Veo 2 can generate videos based on text prompts or reference images, which can simulate the characteristics of motion, fluid mechanics, and light more realistically, achieve different lens and movie effects, and "meticulous" human expressions.
Collins, Vice President of Product at Deep Thinking, stated that in the coming months, the company will continue to iterate its products based on user feedback. He acknowledged that Veo 2 still has shortcomings, and the model's "coherence and consistency" in generating videos still need improvement. There is also room for improvement in generating complex details, fast and complex actions, and breaking through the limitations of reality.
The company also stated that to address the risk of deep forgery, it will use watermark technology to embed invisible markers into the video frames generated by Veo 2. (CCTV News)