--- license: apache-2.0 tags: - diffusion-single-file - comfyui - distillation - LoRA - video - video genration base_model: - Wan-AI/Wan2.2-I2V-A14B pipeline_tags: - image-to-video - text-to-video library_name: diffusers --- # 🎬 Wan2.2 Distilled Models ### ⚡ High-Performance Video Generation with 4-Step Inference *Distillation-accelerated version of Wan2.2 - Dramatically faster speed with excellent quality* ![img_lightx2v](https://cdn-uploads.huggingface.co/production/uploads/680de13385293771bc57400b/tTnp8-ARpj3wGxfo5P55c.png) --- [![🤗 HuggingFace](https://img.shields.io/badge/🤗-HuggingFace-yellow)](https://huggingface.co/lightx2v/Wan2.2-Distill-Models) [![GitHub](https://img.shields.io/badge/GitHub-LightX2V-blue?logo=github)](https://github.com/ModelTC/LightX2V) [![LightX2V Studio](https://img.shields.io/badge/LightX2V_Studio-Create_Online-orange)](https://x2v.light-ai.top/) [![License](https://img.shields.io/badge/License-Apache%202.0-green.svg)](LICENSE) > 🌐 **Create online with [LightX2V Studio](https://x2v.light-ai.top/)** — > an AI avatar and multimodal content production platform built on LightX2V. > Use Wan2.2 video generation in your browser, with no local setup required. > [See the online usage guide below](#method-2-lightx2v-studio--online-creation). --- ## 🔥 News - 2026.04.12: We are excited to release the [Wan2.2-I2V-A14B-4step-720p-high](https://huggingface.co/lightx2v/Wan2.2-Distill-Models/blob/main/wan2.2_i2v_A14b_high_noise_lightx2v_4step_720p_260412.safetensors) and [Wan2.2-I2V-A14B-4step-720p-low](https://huggingface.co/lightx2v/Wan2.2-Distill-Models/blob/main/wan2.2_i2v_A14b_low_noise_lightx2v_4step_720p_260412.safetensors) models. Compared to previous iterations, this version was trained on a high-quality 720p dataset and features an optimized low-noise training algorithm. These enhancements significantly boost the model's performance in fine-grained detail rendering and visual texture. ## 🌟 What's Special?
### ⚡ Ultra-Fast Generation - **4-step inference** (vs traditional 50+ steps) - Approximately **2x faster** using LightX2V than ComfyUI - Near real-time video generation capability ### 🎯 Flexible Options - **Dual noise control**: High/Low noise variants - Multiple precision formats (BF16/FP8/INT8) - Full 14B parameter models
### 💾 Memory Efficient - FP8/INT8: **~50% size reduction** - CPU offload support - Optimized for consumer GPUs ### 🔧 Easy Integration - Compatible with LightX2V framework - Online creation with [LightX2V Studio](https://x2v.light-ai.top/) - ComfyUI support - Simple configuration files
--- ## 📦 Model Catalog ### 🎥 Model Types
#### 🖼️ **Image-to-Video (I2V) - 14B Parameters** Transform static images into dynamic videos with advanced quality control - 🎨 **High Noise**: More creative, diverse outputs - 🎯 **Low Noise**: More faithful to input, stable outputs #### 📝 **Text-to-Video (T2V) - 14B Parameters** Generate videos from text descriptions - 🎨 **High Noise**: More creative, diverse outputs - 🎯 **Low Noise**: More stable and controllable outputs - 🚀 Full 14B parameter model
### 🎯 Precision Versions | Precision | Model Identifier | Model Size | Framework | Quality vs Speed | |:---------:|:-----------------|:----------:|:---------:|:-----------------| | 🏆 **BF16** | `lightx2v_4step` | ~28.6 GB | LightX2V | ⭐⭐⭐⭐⭐ Highest Quality | | ⚡ **FP8** | `scaled_fp8_e4m3_lightx2v_4step` | ~15 GB | LightX2V | ⭐⭐⭐⭐ Excellent Balance | | 🎯 **INT8** | `int8_lightx2v_4step` | ~15 GB | LightX2V | ⭐⭐⭐⭐ Fast & Efficient | | 🔷 **FP8 ComfyUI** | `scaled_fp8_e4m3_lightx2v_4step_comfyui` | ~15 GB | ComfyUI | ⭐⭐⭐ ComfyUI Ready | ### 📝 Naming Convention ```bash # Format: wan2.2_{task}_A14b_{noise_level}_{precision}_lightx2v_4step.safetensors # I2V Examples: wan2.2_i2v_A14b_high_noise_lightx2v_4step.safetensors # I2V High Noise - BF16 wan2.2_i2v_A14b_high_noise_scaled_fp8_e4m3_lightx2v_4step.safetensors # I2V High Noise - FP8 wan2.2_i2v_A14b_low_noise_int8_lightx2v_4step.safetensors # I2V Low Noise - INT8 wan2.2_i2v_A14b_low_noise_scaled_fp8_e4m3_lightx2v_4step_comfyui.safetensors # I2V Low Noise - FP8 ComfyUI ``` > 💡 **Browse All Models**: [View Full Model Collection →](https://huggingface.co/lightx2v/Wan2.2-Distill-Models/tree/main) --- ## 🚀 Usage ### Method 1: LightX2V (Recommended ⭐) **LightX2V is a high-performance inference framework optimized for these models, approximately 2x faster than ComfyUI with better quantization accuracy. Highly recommended!** #### Quick Start 1. Download model (using I2V FP8 as example) ```bash huggingface-cli download lightx2v/Wan2.2-Distill-Models \ --local-dir ./models/wan2.2_i2v \ --include "wan2.2_i2v_A14b_high_noise_scaled_fp8_e4m3_lightx2v_4step.safetensors" ``` ```bash huggingface-cli download lightx2v/Wan2.2-Distill-Models \ --local-dir ./models/wan2.2_i2v \ --include "wan2.2_i2v_A14b_low_noise_scaled_fp8_e4m3_lightx2v_4step.safetensors" ``` > 💡 **Tip**: For T2V models, follow the same steps but replace `i2v` with `t2v` in the filenames 2. Clone LightX2V repository ```bash git clone https://github.com/ModelTC/LightX2V.git cd LightX2V ``` 3. Install dependencies ```bash pip install -r requirements.txt ``` Or refer to [Quick Start Documentation](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/getting_started/quickstart.html) to use docker 4. Select and modify configuration file For the FP8 I2V weights downloaded above, choose: - [FP8 configuration](https://github.com/ModelTC/LightX2V/blob/main/configs/distill/wan22/wan_moe_i2v_distill_quant.json) - [RTX 4090 FP8 configuration with CPU offload](https://github.com/ModelTC/LightX2V/blob/main/configs/distill/wan22/wan_moe_i2v_distill_4090.json) Set `high_noise_quantized_ckpt`, `low_noise_quantized_ckpt`, and `t5_quantized_ckpt` in the selected JSON to your local checkpoint paths. Use an attention backend supported by your GPU and installed environment. For BF16 distilled weights, use the [BF16 configuration](https://github.com/ModelTC/LightX2V/blob/main/configs/distill/wan22/wan_moe_i2v_distill_model.json) instead, and update `high_noise_original_ckpt` and `low_noise_original_ckpt`. 5. Run inference Edit the [I2V launch script](https://github.com/ModelTC/LightX2V/blob/main/scripts/wan22/inference/distill/run_wan22_moe_i2v_distill.sh): set `lightx2v_path`, `model_path`, `CUDA_VISIBLE_DEVICES`, and the input/output paths. Set `--config_json` to the configuration selected above. Keep `model_path` pointed at the complete Wan2.2-I2V-A14B model directory containing the required encoders, tokenizer, and VAE. Run from the LightX2V repository root: ```bash bash scripts/wan22/inference/distill/run_wan22_moe_i2v_distill.sh ``` #### LightX2V Documentation - **Quick Start Guide**: [LightX2V Quick Start](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/getting_started/quickstart.html) - **Complete Usage Guide**: [LightX2V Model Structure Documentation](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/getting_started/model_structure.html) - **Configuration File Instructions**: [Configuration Files](https://github.com/ModelTC/LightX2V/tree/main/configs/distill) - **Quantized Model Usage**: [Quantization Documentation](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/method_tutorials/quantization.html) - **Parameter Offloading**: [Offload Documentation](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/method_tutorials/offload.html) --- ### Method 2: LightX2V Studio — Online Video Creation [**LightX2V Studio**](https://x2v.light-ai.top/) is an online AI avatar and multimodal content production platform built on the LightX2V video generation inference framework. It focuses on AI avatar video creation and also offers text-to-video and image-to-video generation. Create a **silent video** in your browser: 1. Sign in and choose **Text-to-Video** or **Image-to-Video**. 2. Enter a prompt or upload an image. 3. Review the settings and credit cost. 4. Generate and download the result. No local installation or GPU setup is required. ![LightX2V Studio — Image-to-Video (Silent)](https://cdn-uploads.huggingface.co/production/uploads/680de13385293771bc57400b/asXni6E5a9foNHKUwKe7k.png) *LightX2V Studio — Image-to-Video (Silent)* **Current Wan2.2 deployment:** 4-step Rank64 LoRAs from [Wan2.2 Distill LoRAs](https://huggingface.co/lightx2v/Wan2.2-Distill-Loras): - **T2V:** `_1217` high/low-noise LoRAs. - **I2V:** `_1022` high/low-noise LoRAs. Both online workflows generate video **without audio**. --- ### Method 3: ComfyUI Please refer to [workflow](https://huggingface.co/lightx2v/Wan2.2-Distill-Models/blob/main/wan2.2_moe_i2v_scale_fp8_comfyui.json) ## ⚠️ Important Notes **Other Components**: These models only contain DIT weights. Additional components needed at runtime: - T5 text encoder - CLIP vision encoder - VAE encoder/decoder - Tokenizer Please refer to [LightX2V Documentation](https://lightx2v-zhcn.readthedocs.io/zh-cn/latest/getting_started/model_structure.html) for instructions on organizing the complete model directory. ## 🤝 Community - **GitHub Issues**: https://github.com/ModelTC/LightX2V/issues - **HuggingFace**: https://huggingface.co/lightx2v/Wan2.2-Distill-Models If you find this project helpful, please give us a ⭐ on [GitHub](https://github.com/ModelTC/LightX2V)