--- base_model: - Wan-AI/Wan2.2-I2V-A14B - Wan-AI/Wan2.2-TI2V-5B - Wan-AI/Wan2.1-I2V-14B-720P library_name: diffusers license: apache-2.0 tags: - diffusion-single-file - comfyui - distillation - LoRA - video - video generation - sparse-attention pipeline_tag: text-to-video --- # Light Forcing: Accelerating Autoregressive Video Diffusion via Sparse Attention This repository contains the weights and artifacts for **Light Forcing**, the first sparse attention solution tailored for autoregressive (AR) video generation models. [](https://huggingface.co/papers/2602.04789) [](https://github.com/chengtao-lv/LightForcing) Light Forcing introduces a *Chunk-Aware Growth* mechanism and *Hierarchical Sparse Attention* to capture informative historical and local context. It enables significant end-to-end speedups (e.g., up to 3.0ร on an RTX 5090) for models like Wan2.1 and Wan2.2 while maintaining high visual quality. ## ๐ Quick Start ### Fast Inference To use Light Forcing for video generation, please refer to the official [GitHub repository](https://github.com/chengtao-lv/LightForcing) for environment setup and model weights. **For short-video generation (e.g., 5s):** ```shell python inference.py \ --config_path configs/light_forcing_short.yaml \ --output_folder videos/light_forcing_short \ --checkpoint_path path/to/short_video_gen.pt \ --data_path prompts/MovieGenVideoBench_extended.txt \ --use_ema ``` **For long-video generation (e.g., 15s):** ```shell python inference.py \ --config_path configs/light_forcing_long.yaml \ --output_folder videos/light_forcing_long \ --checkpoint_path path/to/long_video_gen.pt \ --data_path prompts/MovieGenVideoBench_extended.txt \ --use_ema \ --num_output_frames 63 ``` --- # ๐จ LightVAE ## โก Efficient Video Autoencoder (VAE) Model Collection *From Official Models to Lightx2v Distilled Optimized Versions - Balancing Quality, Speed and Memory*  --- [](https://huggingface.co/lightx2v) [](https://github.com/ModelTC/LightX2V) [](LICENSE) --- For VAE, the LightX2V team has conducted a series of deep optimizations, deriving two major series: **LightVAE** and **LightTAE**, which significantly reduce memory consumption and improve inference speed while maintaining high quality. ## ๐ก Core Advantages
| ### ๐ Official VAE **Features**: Highest Quality โญโญโญโญโญ โ Best reconstruction accuracy โ Complete detail preservation โ Large memory usage (~8-12 GB) โ Slow inference speed | ### ๐ Open Source TAE Series **Features**: Fastest Speed โกโกโกโกโก โ Minimal memory usage (~0.4 GB) โ Extremely fast inference โ Average quality โญโญโญ โ Potential detail loss |
| ### ๐ฏ **LightVAE Series** (Our Optimization) **Features**: Best Balanced Solution โ๏ธ โ Uses **Causal 3D Conv** (same as official) โ **Quality close to official** โญโญโญโญ โ Memory reduced by **~50%** (~4-5 GB) โ Speed increased by **2-3x** โ Balances quality, speed, and memory ๐ | ### โก **LightTAE Series** (Our Optimization) **Features**: Fast Speed + Good Quality ๐ โ Minimal memory usage (~0.4 GB) โ Extremely fast inference โ **Quality close to official** โญโญโญโญ โ **Significantly surpasses open source TAE** |