Skip to content
Get packs

Open-Sora Plan

Open research project aiming to reproduce OpenAI's Sora text-to-video model in open source.

GitHub Download v1.5.0
Stars
12.2k
Forks
1.1k
Licence
MIT
Language
Python
Last update
8 months ago
Type
AI model / agent

What it is

Open-Sora Plan is an open research project that tries to build a text-to-video generation model similar to OpenAI's Sora, which is closed. The team publishes model checkpoints, code, and technical reports for versions such as v1.5.0, which the README says is trained and run on Huawei Ascend chips. It is aimed at researchers and engineers who want to study or contribute to video generation models.

  • Generate short text-to-video clips using the released checkpoints, according to the README resources table
  • Study the WFVAE video autoencoder and the SUV sparse DiT architecture described in the reports
  • Try image-to-video or transition generation using the v1.2.0 I2V description in the news section
  • Read the versioned technical reports in the docs folder to understand design choices

Great for

  • Researchers studying open video diffusion models and their architecture
  • Engineers with Huawei Ascend hardware who want to run the v1.5.0 code
  • Contributors who want to join an open development effort on text-to-video models

Not ideal for

  • People who just want to make videos quickly, since the README describes setup and model files but not a simple end-user app
  • Users without access to suitable accelerators, since the README mentions 24G VRAM for an older version and Ascend training for v1.5.0 without listing general requirements
  • Those who need a stable, production-ready product, as the README describes rapid iteration and a GPU version still 'coming soon'
Step-by-step guide: how to use Open-Sora PlanInstall, first run, examples, tips and FAQs

How to get it

  1. 1

    Clone and build from source

    Clone the repository

    git clone https://github.com/PKU-YuanGroup/Open-Sora-Plan
  2. 2

    Run it

    • Clone the repository from GitHub
    • Read the README and the docs folder for the version you want (v1.5.0 uses the mindspeed_mmdit branch per the README)
    • Download the checkpoints listed in the Resource table from Hugging Face
    • Follow the setup steps in the repository docs, which were not included in the provided README excerpt
  3. 3

    Get the most out of it

    • Read Report-v1.5.0.md and the other version reports in docs/ to understand the architecture before running
    • Check the Hugging Face checkpoint links in the Resource table for the version that matches your hardware
    • Join the Discord or WeChat community linked in the README for updates and help
    • Watch the repository on GitHub to follow the planned GPU version

Downloads

Source code (.zip)main branch · straight from GitHubRelease v1.5.0Release notes and source

Strengths

  • Publishes papers, technical reports, and model checkpoints openly
  • Has an active community with Discord, WeChat, and regular updates in the README news
  • Released under MIT according to the license metadata

Trade-offs

  • The README says the v1.5.0 code is on a separate mindspeed_mmdit branch and the GPU version is still coming soon
  • Setup instructions are not included in the truncated README, so installation steps are unclear
  • Some documentation and badge links are in Chinese, which may slow non-Chinese readers

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.