Search papers, labs, and topics across Lattice.
This paper introduces Loopy, a novel framework for generating high-quality looping videos by leveraging an anchored position embedding shifting strategy within the DiT model. By identifying the optimal attention layer that serves as a positional anchor, the authors enhance temporal control, transforming the model's perception of video generation from linear to circular. Experimental results show that Loopy significantly improves both temporal consistency and visual fidelity in generated looping videos, making it applicable for various multimedia contexts.
Anchoring positional embeddings in video generation can drastically enhance the quality of looping videos, achieving unprecedented temporal coherence and visual fidelity.
Looping videos are essential for practical applications such as web graphics, game development, and social media. However, existing approaches typically fail to generate high-quality looping videos due to the neglect of how video generation models perceive temporal order and how this relates to the looping behavior. In this work, we are the first to reveal that position embedding at different attention layers within DiT exhibits varying levels of positional control, with the most pronounced layer acting as an anchor. We formulate this anchored layer as the reference point of the looping video, offering strong contextual priors for the remaining layers to facilitate the generation of seamless and coherent video content. Based on this insight, we propose an anchored position embedding shifting strategy that applies layer-specific shift lengths according to each layer's temporal control effect, effectively transforming DiT's temporal perception from a straight line to a circle. Leveraging this strategy, we develop a general framework, Loopy, for high-quality looping video generation, supporting both RGB and RGBA videos, while also enabling advanced AIGC features such as identity control and style transfer. Experiments demonstrate that our approach significantly improves temporal consistency and visual fidelity in generated looping videos. The released model is available on our website: https://donghaotian123.github.io/Loopy.
