Adobe Research has made a groundbreaking breakthrough in video generation technology, effectively unlocking long-term memory in video world models. A team of researchers has successfully integrated State-Space Models (SSMs) to efficiently capture long-range dependencies in video data. This innovative approach combines the strengths of SSMs with dense local attention mechanisms, which work together to maintain coherence in generated video sequences. To achieve this milestone, the researchers employed a range of training strategies, including diffusion forcing and frame local attention. These techniques enable the model to learn complex patterns and relationships within video data, ultimately leading to more realistic and coherent video generation. By overcoming the long-standing challenge of long-term memory in video generation, Adobe Research has opened up new possibilities for applications such as video editing, animation, and virtual reality. The implications of this research are significant, as it has the potential to revolutionize the field of video generation. With the ability to effectively model long-term dependencies, video models can now generate more realistic and engaging content, such as complex scenes, characters, and storylines. This breakthrough is expected to have far-reaching consequences for industries such as entertainment, education, and advertising. Adobe Research's achievement is a testament to the power of innovative research and development in the field of artificial intelligence. By pushing the boundaries of what is possible with video generation, the team has paved the way for new and exciting applications that will continue to transform the way we interact with and experience video content.