• [10/12/2023] Code and weights released!

pip install -r requirements.txt

Note: PyTorch 2.0+ is highly recommended for more efficiency and speed on GPUs.

All model weights for Show-1 are available on Show Lab's HuggingFace page: Base Model (show-1-base), Interpolation Model (show-1-interpolation), and Super-Resolution Model (show-1-sr1, show-1-sr2).

Note that our show-1-sr1 incorporates the image super-resolution model from DeepFloyd-IF, DeepFloyd/IF-II-L-v1.0, to upsample the first frame of the video. To obtain the respective weights, follow their official instructions.

To generate a video from a text prompt, run the command below:

python run_inference.py

By default, the videos generated from each stage are saved to the outputs folder in the GIF format. The script will automatically fetch the necessary model weights from HuggingFace. If you prefer, you can manually download the weights using git lfs and then update the pretrained_model_path to point to your local directory. Here's how:

git lfs install
git clone https://huggingface.co/showlab/show-1-base 

If you make use of our work, please cite our paper.

@misc{zhang2023show1,
    title={Show-1: Marrying Pixel and Latent Diffusion Models for Text-to-Video Generation}, 
    author={David Junhao Zhang and Jay Zhangjie Wu and Jia-Wei Liu and Rui Zhao and Lingmin Ran and Yuchao Gu and Difei Gao and Mike Zheng Shou},
    year={2023},
    eprint={2309.15818},
    archivePrefix={arXiv},
    primaryClass={cs.CV}
}

We are working with the university (NUS) to figure out the exact paperwork needed for approving commercial use request. In the meantime, to speed up the process, we'd like to solicit intent of interest from community and later on we will process these requests with high priority. If you are keen, can you kindly email us at [email protected] and [email protected] to answer the following questions, if possible:

• Who are you / your company?
• What is your product / application?
• How Show-1 can benefit your product?

• This work heavily builds on diffusers, deep-floyd/IF, modelscope, and zeroscope. Thanks for open-sourcing!