video_diffusion_play

Generate videos from text using the latest Wan2.2 model from https://wan.video and Alibaba's Tongyi Lab

"Over-the-shoulder shot, two shot, telephoto lens, warm colors, high contrast lighting, soft lighting, daylight, daylight, close-up shot, center composition.In an eye-level shot, a foreign girl sits by a window. She is wearing a red and black striped sweater over a blue turtleneck, and her short, smooth golden hair is tucked behind her ears. Her gaze is gentle as she looks at someone off-camera with a slight smile, her expression natural and friendly. Her hands are gently folded on the table, where an open book and a white ceramic cup with steam gently rising are placed. The background is out-of-focus, with soft daylight streaming in through the window. Faint warm yellow lights and scattered red decorations are visible, creating a cozy atmosphere."

Hardware Configuration

• Intel/AMD system with at least 24 cores
• 256 GB DDR5
• NVIDIA 5090 w/ 32GB VRAM
• 150GB free storage space for models

Prerequisites

System setup with:

• kernel drivers for NVIDIA GPU
• docker with NVIDIA-Runtime support
• 'make' installed with your package manager (e.g. apt install make)

Steps for setup and execution

1. Build docker container

$ make image

2. Run sample text to video script

$ make run

On NVIDIA 5090 card, it takes about 45 minutes to generate about 81 frames of 1280x720p video

References

• https://github.com/Wan-Video/Wan2.2
• https://huggingface.co/DFloat11/Wan2.2-T2V-A14B-DF11