Open-Sora-Plan

Why we included this project

Open-Sora Plan is an open source text-to-video project that reimplements the approach behind OpenAI's Sora in code anyone can run and extend. It ships training and inference recipes, pretrained checkpoints, and an architecture that scales across GPUs, so teams can generate long, high-resolution videos from text prompts without depending on closed commercial APIs. The code also sits alongside arXiv papers covering the model design and its video autoencoder, which makes it a practical reference for anyone studying how large video diffusion models are built. The V1.5 release, trained entirely on Huawei Ascend hardware, gives teams on that stack a ready starting point. Development stays active and outside contributions are welcome.