Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..
WorldWeaver: Generating Long-Horizon Video Worlds via Rich Perception
Authors: Zhiheng Liu, Xueqing Deng, Shoufa Chen, Angtian Wang, Qiushan Guo, Mingfei Han, Zeyue Xue, Mengzhao Chen, Ping Luo, Linjie Yang
NeurIPS 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experiments on both diffusion-based and rectified flow-based models demonstrate the effectiveness of World Weaver in reducing temporal drift and improving the fidelity of generated videos. |
| Researcher Affiliation | Collaboration | 1The University of Hong Kong, 2Byte Dance Seed |
| Pseudocode | No | The paper describes the methodology in prose and mathematical equations but does not contain any explicit pseudocode or algorithm blocks. |
| Open Source Code | Yes | Answer: [Yes] Justification: We release the codes and in supplementary materials |
| Open Datasets | Yes | To comprehensively evaluate the effectiveness and generalizability of our proposed method, we conduct experiments using two publicly available foundation models: Wan2.1-1.3B [63] (flow-based)... and Cog Video X-2B [72] (diffusion-based), trained on the in-the-wild DROID robotic manipulation dataset [42]... |
| Dataset Splits | No | The paper describes the datasets used and how they were trained, but does not explicitly provide details about training/test/validation dataset splits, percentages, or sample counts. |
| Hardware Specification | Yes | All training is conducted on 32 NVIDIA A100 GPUs with a learning rate of 1e-4, using the Adam W optimizer [50], with a per-GPU batch size of 4. |
| Software Dependencies | No | The paper mentions the use of 'Adam W optimizer [50]' but does not provide specific version numbers for any other software libraries, frameworks, or programming languages used. |
| Experiment Setup | Yes | All training is conducted on 32 NVIDIA A100 GPUs with a learning rate of 1e-4, using the Adam W optimizer [50], with a per-GPU batch size of 4. The Wan2.1-1.3B model is trained for 50K iterations while the Cog Video X-2B model is trained for 15K steps. To preserve text-to-video performance, we apply the same noise level to every frame in 10% of the training steps. |