The field of video generation is rapidly advancing, with a focus on improving the efficiency and quality of long video generation. Recent developments have centered on addressing the challenges of autoregressive models, such as error accumulation and limited context understanding. Researchers are exploring new architectures and techniques, including frame-level autoregressive designs, spatiotemporal cubes, and joint denoising schemes, to enable real-time and interactive long video generation. These innovations have led to significant improvements in video quality, temporal coherence, and generation speed. Noteworthy papers in this area include: LongLive, which presents a causal, frame-level AR design for real-time and interactive long video generation, and Autoregressive Video Generation beyond Next Frames Prediction, which introduces a unified framework that supports a spectrum of prediction units, including spatiotemporal cubes. Rolling Forcing is also notable for its novel video generation technique that enables streaming long videos with minimal error accumulation. Arbitrary Generative Video Interpolation and Pack and Force Your Memory also present significant contributions to the field, with the former enabling efficient interpolation at any timestamp and of any length, and the latter introducing a learnable context-retrieval mechanism and an efficient single-step approximating strategy for long-form video generation. Self-Forcing++ is another notable work that proposes a simple yet effective approach to mitigate quality degradation in long-horizon video generation.