| A serene night scene in a forested area. [...] The video is a time-lapse, capturing the transition from day to night, with the lake and forest serving as a constant backdrop. | A soaring drone footage captures the majestic beauty of a coastal cliff, [...] The water gently laps at the rock base and the greenery that clings to the top of the cliff. | The majestic beauty of a waterfall cascading down a cliff into a serene lake. [...] The camera angle provides a bird's eye view of the waterfall. |
| A bustling city street at night, filled with the glow of car headlights and the ambient light of streetlights. [...] | The vibrant beauty of a sunflower field. The sunflowers are arranged in neat rows, creating a sense of order and symmetry. [...] | A serene underwater scene featuring a sea turtle swimming through a coral reef. The turtle, with its greenish-brown shell [...] |
Videos are downsampled to `.gif` for display. Click for original videos. Texts are trimmed for display, see [here](/assets/texts/t2v_samples.txt) for full texts. See more samples at our [gallery](https://hpcaitech.github.io/Open-Sora/).
* 📍 Open-Sora-v1 released. Model weights are available [here](#model-weights). With only 400K video clips and 200 H800 days (compared with 152M samples in Stable Video Diffusion), we are able to generate 2s 512×512 videos.
* ✅ Support training acceleration including accelerated transformer, faster T5 and VAE, and sequence parallelism. Open-Sora improve **55%** training speed when training on 64x512x512 videos. Details locates at [acceleration.md](docs/acceleration.md).
* ✅ We provide video cutting and captioning tools for data preprocessing. Instructions can be found [here](tools/data/README.md) and our data collection plan can be found at [datasets.md](docs/datasets.md).
* ✅ We find VQ-VAE from [VideoGPT](https://wilson1yan.github.io/videogpt/index.html) has a low quality and thus adopt a better VAE from [Stability-AI](https://huggingface.co/stabilityai/sd-vae-ft-mse-original). We also find patching in the time dimension deteriorates the quality. See our **[report](docs/report_v1.md)** for more discussions.
* ✅ We investigate different architectures including DiT, Latte, and our proposed STDiT. Our **STDiT** achieves a better trade-off between quality and speed. See our **[report](docs/report_v1.md)** for more discussions.
* ✅ Support clip and T5 text conditioning.
* ✅ By viewing images as one-frame videos, our project supports training DiT on both images and videos (e.g., ImageNet & UCF101). See [command.md](docs/command.md) for more instructions.
* ✅ Support inference with official weights from [DiT](https://github.com/facebookresearch/DiT), [Latte](https://github.com/Vchitect/Latte), and [PixArt](https://pixart-alpha.github.io/).
* [ ] Complete the data processing pipeline (including dense optical flow, aesthetics scores, text-image similarity, deduplication, etc.). See [datasets.md](/docs/datasets.md) for more information. **[WIP]**
Our model's weight is partially initialized from [PixArt-α](https://github.com/PixArt-alpha/PixArt-alpha). The number of parameters is 724M. More information about training can be found in our **[report](/docs/report_v1.md)**. More about dataset can be found in [dataset.md](/docs/dataset.md). HQ means high quality.
:warning: **LIMITATION**: Our model is trained on a limited budget. The quality and text alignment is relatively poor. The model performs badly especially on generating human beings and cannot follow detailed instructions. We are working on improving the quality and text alignment.
To run inference with our provided weights, first download [T5](https://huggingface.co/DeepFloyd/t5-v1_1-xxl/tree/main) weights into `pretrained_models/t5_ckpts/t5-v1_1-xxl`. Then download the model weights. Run the following commands to generate samples. See [here](docs/structure.md#inference-config-demos) to customize the configuration.
High-quality Data is the key to high-quality models. Our used datasets and data collection plan is [here](/docs/datasets.md). We provide tools to process video data. Currently, our data processing pipeline includes the following steps:
To launch training, first download [T5](https://huggingface.co/DeepFloyd/t5-v1_1-xxl/tree/main) weights into `pretrained_models/t5_ckpts/t5-v1_1-xxl`. Then run the following commands to launch training on a single node.
To launch training on multiple nodes, prepare a hostfile according to [ColossalAI](https://colossalai.org/docs/basics/launch_colossalai/#launch-with-colossal-ai-cli), and run the following commands.
* [OpenDiT](https://github.com/NUS-HPC-AI-Lab/OpenDiT): An acceleration for DiT training. We adopt valuable acceleration strategies for training progress from OpenDiT.
[Zangwei Zheng](https://github.com/zhengzangw) and [Xiangyu Peng](https://github.com/xyupeng) equally contributed to this work during their internship at [HPC-AI Tech](https://hpc-ai.com/).
