wan2video 0.1.0

Generate videos using Wan2.2 TI2V model with GGUF quantization and GPU acceleration

wan2video

Generate videos using Wan2.2 TI2V (Text+Image to Video) 5B model with GGUF quantization and GPU acceleration.

Quick Start

# Install and run (requires sd-cli in PATH)
uvx wan2video "a cat walking on the grass, sunny day"

# Or install first
pip install wan2video
wan2video "a beautiful sunset over the ocean"

Prerequisites

• NVIDIA GPU with CUDA support (12GB+ VRAM recommended)
• stable-diffusion.cpp compiled with CUDA (sd-cli in PATH)
• ffmpeg (for mp4 output)

Setup sd-cli

wan2video setup

Or compile manually:

git clone --recurse-submodules https://github.com/leejet/stable-diffusion.cpp
cd stable-diffusion.cpp && mkdir build && cd build
cmake .. -DSD_CUBLAS=ON -DGGML_CUDA=ON
make -j
# Add build/bin/sd-cli to your PATH

Usage

# Text to video (480x320, 2 seconds)
wan2video "your prompt here"

# 720p video
wan2video "prompt" -W 1280 -H 720

# 1080p video
wan2video "prompt" -W 1920 -H 1080

# Image to video
wan2video "prompt describing motion" -i input.jpg

# More frames (5 seconds)
wan2video "prompt" -f 81

# Custom output
wan2video "prompt" -o my_video.mp4 --steps 30 --seed -1

Options

Option	Default	Description
-W/--width	480	Video width
-H/--height	320	Video height
-f/--frames	33	Frames (33≈2s, 81≈5s)
-s/--steps	20	Sampling steps
--cfg-scale	7.0	CFG guidance scale
--seed	42	Random seed (-1 for random)
-i/--image	None	Input image for I2V
-o/--output	output.mp4	Output file path

Performance (RTX 4070 Ti 12GB)

Resolution	Frames	Time
480x320	33	~16s
1280x720	33	~2.5min
1920x1080	33	~9min

License

MIT