Ouroboros-Diffusion: Exploring Consistent Content Generation in Tuning-free Long Video Diffusion

Authors: Jingyuan Chen, Fuchen Long, Jie An, Zhaofan Qiu, Ting Yao, Jiebo Luo, Tao Mei

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments of long video generation on the VBench benchmark demonstrate the superiority of our Ouroboros-Diffusion, particularly in terms of subject consistency, motion smoothness, and temporal consistency. ... Extensive experiments on VBench verify the effectiveness of our proposal in terms of both visual and motion quality.
Researcher Affiliation Collaboration 1 University of Rochester, Rochester, NY USA 2 Hi Dream.ai Inc.
Pseudocode No The paper describes methods like 'Coherent tail latent sampling', 'Subject-Aware Cross-Frame Attention (SACFA)', and 'Self-Recurrent Guidance' using mathematical formulations and descriptive text, but it does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any explicit statements about releasing code or a link to a code repository.
Open Datasets Yes We empirically verify the merit of our Ouroboros-Diffusion for both single-scene and multi-scene long video generation on the VBench (Huang et al. 2024) benchmark.
Dataset Splits Yes We sample 93 common prompts from VBench as the testing set for single-scene video generation. ... For each multi-prompt group, we generate 256 video frames for performance comparison.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or memory amounts used for running experiments.
Software Dependencies No The paper states: 'We implement our Ouroboros Diffusion on the text-to-video model Video Crafter2 (Chen et al. 2024a).' However, it does not provide specific version numbers for underlying software libraries (e.g., PyTorch, TensorFlow, CUDA).
Experiment Setup Yes The total number of time steps T in the DDIM sampler is set to 64, matching the queue length. The threshold for the low-pass filter in coherent tail latent sampling is set to 0.25. SACFA is applied only in the down-blocks and mid-block (with down-sampling factors of 2 and 4) of the spatial-temporal UNet empirically. The last 16 frames in the queue are involved in SACFA calculation. The self-recurrent guidance derived from the first 16 frames at the queue head applies to the last 16 frames at the tail. The parameter λ for updating the subject feature bank is set to 0.98.