NUS-HPC-AI-Lab/VideoSys

Health v2 · maintenance only

Adoption not part of the score

2025 stars · 129 forks observed · 2026-09-23

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

VideoSys is an open-source Python library providing easy and efficient infrastructure for video generation, supporting training, inference, serving, and compression of diffusion-based video models. It integrates models like Open-Sora, CogVideoX, Latte, and Vchitect-2.0, and includes acceleration techniques such as Pyramid Attention Broadcast (PAB) and Dynamic/Data-Centric Sequence Parallelism.

Use cases

• generate videos with open-source diffusion models
• accelerate DiT-based video generation to real-time
• train video generation models with sequence parallelism
• run CogVideoX or Open-Sora inference efficiently
• reduce memory usage for video model training
• serve and compress video generation models

When to choose

• you want a unified, high-performance pipeline for open-source video generation models
• you need GPU acceleration techniques like PAB or sequence parallelism for video diffusion models
• you want to train or fine-tune DiT-based video models on multi-GPU setups

When to avoid

• you need simple image generation rather than video
• you want a plug-and-play consumer app with no GPU infrastructure
• you need commercial closed-model video generation APIs

Facets

library · maturity active

video-processing machine-learning deep-learning llm-training gpu-computing machine-learning deep-learning gpu-computing artificial-intelligence python video-generation diffusion-transformers model-inference distributed-training pytorch open-sora cogvideox latte sequence-parallelism attention-broadcast video gpu linux docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="NUS-HPC-AI-Lab/VideoSys")