Producing a big movie in 3 minutes, the 'directorial era' of AI videos has arrived. I spent 2 hours last night, The newly released 'AI Movie Generation', also known as Dream Seedance 2.0, has been tested, Tell me some real feelings. 1. The real person texture has completed steps 0 to 90. The animation, light and shadow, and skin texture are basically equivalent to movie level production. You can see that Musk and his team have very realistic character textures. Activists can basically be laid off. I tried making over 10 short videos. There are virtual and real contrast scenes for testing. The themes include Labubu in the Bund battle against the Ice City of Miyuki, and Hayao Miyazaki's female lead shopping on Nanjing East Road. The integration between animation and the real world exceeds expectations. The most stunning features are the camera movement and storyboarding. For example, my prompt only wrote one sentence: 'The camera follows the character through the hallway to the riverbank.' It has planned its own set of camera language for following, pushing, pulling, and lifting, The resulting effect is simply like someone carrying a Stannikon in real photos. 4. Sound and picture synchronization. Listen carefully, AI actively matches the sound of lightning and thunder based on the picture. Of course, BGM Chinese dubbing is still a bit sluggish, and I have tried several times but none of them have been ideal. 5. The drawbacks are also very obvious. There is a 20% chance that the text in the screen will become garbled; The longest generation time is only 15 seconds, but it only takes about 3 minutes to generate. The overall feeling is that Seedance 2.0 wins in camera controllability, and you are experiencing the feeling of being a director. The strength of Chinese AI cannot be underestimated. But 15 seconds is currently the ceiling for all AI video models. When can AI videos truly transform from toys to tools with longer stable output time. This day won't be too far away.