Search papers, labs, and topics across Lattice.
The authors introduce a highly optimized training framework for large-scale video generation models, focusing on data processing, model architecture, training strategy, and infrastructure. This framework incorporates techniques like efficient video compression, curriculum-based pretraining, and alignment-focused post-training. The resulting MUG-V 10B model achieves state-of-the-art performance, particularly on e-commerce video generation, and the complete training stack is open-sourced, including Megatron-Core-based training code.
Training large video generation models just got a whole lot easier: MUG-V 10B matches state-of-the-art performance while open-sourcing the entire Megatron-Core-based training stack for near-linear scaling.
In recent years, large-scale generative models for visual content (\textit{e.g.,} images, videos, and 3D objects/scenes) have made remarkable progress. However, training large-scale video generation models remains particularly challenging and resource-intensive due to cross-modal text-video alignment, the long sequences involved, and the complex spatiotemporal dependencies. To address these challenges, we present a training framework that optimizes four pillars: (i) data processing, (ii) model architecture, (iii) training strategy, and (iv) infrastructure for large-scale video generation models. These optimizations delivered significant efficiency gains and performance improvements across all stages of data preprocessing, video compression, parameter scaling, curriculum-based pretraining, and alignment-focused post-training. Our resulting model, MUG-V 10B, matches recent state-of-the-art video generators overall and, on e-commerce-oriented video generation tasks, surpasses leading open-source baselines in human evaluations. More importantly, we open-source the complete stack, including model weights, Megatron-Core-based large-scale training code, and inference pipelines for video generation and enhancement. To our knowledge, this is the first public release of large-scale video generation training code that exploits Megatron-Core to achieve high training efficiency and near-linear multi-node scaling, details are available in https://github.com/Shopee-MUG/MUG-V.
