Omni-VFX
收藏资源简介:
Omni-VFX数据集是阿里巴巴集团AMAP、北京大学PKU、清华大学THU、中国科学院自动化研究所CASIA联合构建的一个高质量视觉特效数据集。该数据集包含55个不同的特效类别,通过图像编辑和First-Last Frame-to-Video (FLF2V)合成方法生成,旨在支持多种高质量的视觉特效视频生成。Omni-VFX数据集的构建过程经过了严格的手动筛选,以确保数据质量,并覆盖了广泛的特效类型。该数据集的开发旨在支持多条件视频生成任务,为视觉特效生成领域的研究提供了宝贵的数据资源。
The Omni-VFX dataset is a high-quality visual effects (VFX) dataset jointly constructed by AMAP of Alibaba Group, Peking University (PKU), Tsinghua University (THU), and the Institute of Automation of the Chinese Academy of Sciences (CASIA). It contains 55 distinct VFX categories, which are generated via image editing and the First-Last Frame-to-Video (FLF2V) synthesis method, aiming to support a variety of high-quality visual effects video generation tasks. The construction process of the Omni-VFX dataset has undergone strict manual screening to ensure data quality and covers a wide range of VFX types. Developed to support multi-condition video generation tasks, this dataset provides valuable data resources for research in the field of visual effects generation.
Omni-Effects数据集概述
基本信息
- 数据集名称: Omni-Effects
- 研究论文: arXiv:2508.07981
- 项目页面: OmniEffects Project Page
- HuggingFace资源:
数据集简介
- 目的: 提供统一的视觉特效(VFX)生成框架,支持空间可控的复合特效生成。
- 核心创新:
- LoRA-based Mixture of Experts (LoRA-MoE): 通过专家LoRA组整合多样特效,减少跨任务干扰。
- Spatial-Aware Prompt (SAP): 将空间掩码信息融入文本标记,实现精确空间控制。
- Independent-Information Flow (IIF): 隔离各特效的控制信号,避免混合干扰。
数据集内容
- Omni-VFX数据集:
- 来源:
- Open-VFX数据集
- Remade-AI提供的特效蒸馏
- 使用FLF2V合成的特效视频
- 限制: 部分视频因版权限制未公开。
- 来源:
- 支持的特效类型:
- "Melt it"
- "Levitate it"
- "Explode it"
- "Turn it into anime style"
- "Change the setting to a winter scene"
使用方法
-
安装: shell git clone https://github.com/AMAP-ML/Omni-Effects.git cd Omni-Effects conda create -n OmniEffects python=3.10.14 pip install -r requirements.txt
-
模型下载: 从HuggingFace下载检查点并放入
checkpoints文件夹。 -
脚本示例:
- 提示引导的特效生成:
sh scripts/prompt_guided_VFX.sh - 单特效空间控制:
sh scripts/inference_omnieffects_singleVFX.sh - 多特效空间控制:
sh scripts/inference_omnieffects_multiVFX.sh
- 提示引导的特效生成:
相关资源
- 预训练模型: CogVideoX1.5-5B-I2V-OmniVFX
- 提示列表: 见
VFX-prompts.txt
引用
bibtex @misc{mao2025omnieffects, title={Omni-Effects: Unified and Spatially-Controllable Visual Effects Generation}, author={Fangyuan Mao and Aiming Hao and Jintao Chen and Dongxia Liu and Xiaokun Feng and Jiashu Zhu and Meiqi Wu and Chubin Chen and Jiahong Wu and Xiangxiang Chu}, year={2025}, eprint={2508.07981}, archivePrefix={arXiv}, primaryClass={cs.CV} }




