--- title: Latte-1 app_file: demo.py tags: - Text-to-Video sdk: gradio sdk_version: 4.37.2 short_description: Text-to-Video emoji: 🤗 colorFrom: blue colorTo: indigo --- ## Latte: Latent Diffusion Transformer for Video Generation
Official PyTorch Implementation [![Arxiv](https://img.shields.io/badge/Arxiv-b31b1b.svg)](https://arxiv.org/abs/2401.03048) [![Project Page](https://img.shields.io/badge/Project-Website-blue)](https://maxin-cn.github.io/latte_project/) [![HF Demo](https://img.shields.io/static/v1?label=Demo&message=OpenBayes%E8%B4%9D%E5%BC%8F%E8%AE%A1%E7%AE%97&color=green)](https://openbayes.com/console/public/tutorials/UOeU0ywVxl7) [![Static Badge](https://img.shields.io/badge/Latte--1%20checkpoint%20(T2V)-HuggingFace-yellow?logoColor=violet%20Latte-1%20checkpoint)](https://huggingface.co/maxin-cn/Latte-1) [![Static Badge](https://img.shields.io/badge/Latte%20checkpoint%20-HuggingFace-yellow?logoColor=violet%20Latte%20checkpoint)](https://huggingface.co/maxin-cn/Latte) This repo contains PyTorch model definitions, pre-trained weights, training/sampling code and evaluation code for our paper exploring latent diffusion models with transformers (Latte). You can find more visualizations on our [project page](https://maxin-cn.github.io/latte_project/). > [**Latte: Latent Diffusion Transformer for Video Generation**](https://maxin-cn.github.io/latte_project/)
> [Xin Ma](https://maxin-cn.github.io/), [Yaohui Wang*](https://wyhsirius.github.io/), [Xinyuan Chen](https://scholar.google.com/citations?user=3fWSC8YAAAAJ), [Gengyun Jia](https://scholar.google.com/citations?user=_04pkGgAAAAJ&hl=zh-CN), [Ziwei Liu](https://liuziwei7.github.io/), [Yuan-Fang Li](https://users.monash.edu/~yli/), [Cunjian Chen](https://cunjian.github.io/), [Yu Qiao](https://scholar.google.com.hk/citations?user=gFtI-8QAAAAJ&hl=zh-CN) > (*Corresponding Author & Project Lead) ## News - (🔥 New) **Jul 11, 2024** 💥 **Latte-1 is now integrated into [diffusers](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/transformers/latte_transformer_3d.py). Thanks to [@yiyixuxu](https://github.com/yiyixuxu), [@sayakpaul](https://github.com/sayakpaul), [@a-r-r-o-w](https://github.com/a-r-r-o-w) and [@DN6](https://github.com/DN6).** You can easily run Latte using the following code. We also support inference with 4/8-bit quantization, which can reduce GPU memory from 17 GB to 9 GB. Please refer to this [tutorial](docs/latte_diffusers.md) for more information. ``` from diffusers import LattePipeline from diffusers.models import AutoencoderKLTemporalDecoder from torchvision.utils import save_image import torch import imageio torch.manual_seed(0) device = "cuda" if torch.cuda.is_available() else "cpu" video_length = 16 # 1 (text-to-image) or 16 (text-to-video) pipe = LattePipeline.from_pretrained("maxin-cn/Latte-1", torch_dtype=torch.float16).to(device) # Using temporal decoder of VAE vae = AutoencoderKLTemporalDecoder.from_pretrained("maxin-cn/Latte-1", subfolder="vae_temporal_decoder", torch_dtype=torch.float16).to(device) pipe.vae = vae prompt = "a cat wearing sunglasses and working as a lifeguard at pool." videos = pipe(prompt, video_length=video_length, output_type='pt').frames.cpu() ``` - (🔥 New) **May 23, 2024** 💥 **Latte-1** is released! Pre-trained model can be downloaded [here](https://huggingface.co/maxin-cn/Latte-1/tree/main/transformer). **We support both T2V and T2I**. Please run `bash sample/t2v.sh` and `bash sample/t2i.sh` respectively. - (🔥 New) **Feb 24, 2024** 💥 We are very grateful that researchers and developers like our work. We will continue to update our LatteT2V model, hoping that our efforts can help the community develop. Our Latte discord channel is created for discussions. Coders are welcome to contribute. - (🔥 New) **Jan 9, 2024** 💥 An updated LatteT2V model initialized with the [PixArt-α](https://github.com/PixArt-alpha/PixArt-alpha) is released, the checkpoint can be found [here](https://huggingface.co/maxin-cn/Latte-0/tree/main/transformer). - (🔥 New) **Oct 31, 2023** 💥 The training and inference code is released. All checkpoints (including FaceForensics, SkyTimelapse, UCF101, and Taichi-HD) can be found [here](https://huggingface.co/maxin-cn/Latte/tree/main). In addition, the LatteT2V inference code is provided. ## Setup First, download and set up the repo: ```bash git clone https://github.com/Vchitect/Latte cd Latte ``` We provide an [`environment.yml`](environment.yml) file that can be used to create a Conda environment. If you only want to run pre-trained models locally on CPU, you can remove the `cudatoolkit` and `pytorch-cuda` requirements from the file. ```bash conda env create -f environment.yml conda activate latte ``` ## Sampling You can sample from our **pre-trained Latte models** with [`sample.py`](sample/sample.py). Weights for our pre-trained Latte model can be found [here](https://huggingface.co/maxin-cn/Latte). The script has various arguments to adjust sampling steps, change the classifier-free guidance scale, etc. For example, to sample from our model on FaceForensics, you can use: ```bash bash sample/ffs.sh ``` or if you want to sample hundreds of videos, you can use the following script with Pytorch DDP: ```bash bash sample/ffs_ddp.sh ``` If you want to try generating videos from text, just run `bash sample/t2v.sh`. All related checkpoints will download automatically. If you would like to measure the quantitative metrics of your generated results, please refer to [here](docs/datasets_evaluation.md). ## Training We provide a training script for Latte in [`train.py`](train.py). The structure of the datasets can be found [here](docs/datasets_evaluation.md). This script can be used to train class-conditional and unconditional Latte models. To launch Latte (256x256) training with `N` GPUs on the FaceForensics dataset : ```bash torchrun --nnodes=1 --nproc_per_node=N train.py --config ./configs/ffs/ffs_train.yaml ``` or If you have a cluster that uses slurm, you can also train Latte's model using the following scripts: ```bash sbatch slurm_scripts/ffs.slurm ``` We also provide the video-image joint training scripts [`train_with_img.py`](train_with_img.py). Similar to [`train.py`](train.py) scripts, these scripts can be also used to train class-conditional and unconditional Latte models. For example, if you want to train the Latte model on the FaceForensics dataset, you can use: ```bash torchrun --nnodes=1 --nproc_per_node=N train_with_img.py --config ./configs/ffs/ffs_img_train.yaml ``` ## Contact Us **Yaohui Wang**: [wangyaohui@pjlab.org.cn](mailto:wangyaohui@pjlab.org.cn) **Xin Ma**: [xin.ma1@monash.edu](mailto:xin.ma1@monash.edu) ## Citation If you find this work useful for your research, please consider citing it. ```bibtex @article{ma2024latte, title={Latte: Latent Diffusion Transformer for Video Generation}, author={Ma, Xin and Wang, Yaohui and Jia, Gengyun and Chen, Xinyuan and Liu, Ziwei and Li, Yuan-Fang and Chen, Cunjian and Qiao, Yu}, journal={arXiv preprint arXiv:2401.03048}, year={2024} } ``` ## Acknowledgments Latte has been greatly inspired by the following amazing works and teams: [DiT](https://github.com/facebookresearch/DiT) and [PixArt-α](https://github.com/PixArt-alpha/PixArt-alpha), we thank all the contributors for open-sourcing. ## License The code and model weights are licensed under [LICENSE](LICENSE).