LAMP
LAMP is an official implementation of the research paper accepted at CVPR 2024, designed for few-shot text-to-video generation. This software enables users to learn specific motion patterns from a small set of 8 to 16 video samples and apply them to generate new videos based on text prompts. The framework utilizes a Stable Diffusion v1.4 backbone and requires a single GPU with over 15 GB VRAM for efficient training. It supports both general video generation and advanced video editing tasks. The repository includes source code, pre-trained checkpoints, and training data for various motion types such as birds flying, fireworks, helicopter movement, and horses running. Users can train their own models using custom video datasets collected from public sources or utilize provided examples via Google Drive, Baidu Disk, or Colab notebooks. The system is built on Python 3.8, PyTorch 1.12.1, and runs on Ubuntu with CUDA 11.3. It offers a streamlined workflow for researchers and developers to create dynamic video conte