StepVideo is a state-of-the-art (SoTA) text-to-video pre-trained model with 30 billion parameters and the capability to generate videos up to 204 frames.
- Model: https://modelscope.cn/models/stepfun-ai/stepvideo-t2v/summary
- GitHub: https://github.com/stepfun-ai/Step-Video-T2V
- Technical report: https://arxiv.org/abs/2502.10248
For original BF16 version, please see ./stepvideo_text_to_video.py. 80G VRAM required.
We also support auto-offload, which can reduce the VRAM requirement to 24GB; however, it requires 2x time for inference. Please see ./stepvideo_text_to_video_low_vram.py.
video.mp4
For FP8 quantized version, please see ./stepvideo_text_to_video_quantized.py. 40G VRAM required.