v1.2.0

daydreamlive/DEMONv1.2.0Jun 18, 2024by FrankLeeeee

AI Summary

This release introduces significant enhancements to the Open-Sora video generation model, including new training methodologies, compression networks, and improved evaluation metrics.

Key Highlights

Added a video compression network.
Introduced rectifiable-flow training.
Enhanced data availability and multi-stage training capabilities.
Improved model conditioning for easier and more effective usage.
Better evaluation metrics implemented.

New Features

Video compression network
Rectifiable-flow training
Enhanced multi-stage training with more data
Easy and effective model conditioning
Better evaluation metrics

Full Release Notes

Today, we are happy to announce the v1.2 release of the Open-Sora project. This release is built upon v1.1 and features extra exciting components:

1. 📦 Video compression network
2. 🔀 Rectifie-flow training
3. 🪅 More data and better multi-stage training
4. 🚀 Easy and effective model conditioning
5. 👍 Better evaluation metrics

If you want to know more about this project, you can
- read our [technical report](https://github.com/hpcaitech/Open-Sora/blob/main/docs/report_03.md)
- view our [gallery](https://hpcaitech.github.io/Open-Sora/)
- try our [checkpoints](https://github.com/hpcaitech/Open-Sora?tab=readme-ov-file#open-sora-12-model-weights) and [gradio demo](https://huggingface.co/spaces/hpcai-tech/open-sora)