lightx2v/Wan2.1-Distill-Models

🎬 Wan2.1 Distilled Models

⚡ High-Performance Video Generation with 4-Step Inference

Distillation-accelerated versions of Wan2.1 - Dramatically faster while maintaining exceptional quality

---

---

🌟 What's Special?

⚡ Ultra-Fast Generation 4-step inference (vs traditional 50+ steps) Up to 2x faster than ComfyUI Real-time video generation capability	🎯 Flexible Options Multiple resolutions (480P/720P) Various precision formats (BF16/FP8/INT8) I2V and T2V support
💾 Memory Efficient FP8/INT8: ~50% size reduction CPU offload support Optimized for consumer GPUs	🔧 Easy Integration Compatible with LightX2V framework ComfyUI support available Simple configuration files

---

📦 Model Catalog

🎥 Model Types

🖼️ Image-to-Video (I2V) Transform still images into dynamic videos 📺 480P Resolution 🎬 720P Resolution

📝 Text-to-Video (T2V) Generate videos from text descriptions 🚀 14B Parameters 🎨 High-quality synthesis

🎯 Precision Variants

Precision	Model Identifier	Model Size	Framework	Quality vs Speed
🏆 BF16	lightx2v_4step	~28-32 GB	LightX2V	⭐⭐⭐⭐⭐ Highest quality
⚡ FP8	scaled_fp8_e4m3_lightx2v_4step	~15-17 GB	LightX2V	⭐⭐⭐⭐ Excellent balance
🎯 INT8	int8_lightx2v_4step	~15-17 GB	LightX2V	⭐⭐⭐⭐ Fast & efficient
🔷 FP8 ComfyUI	scaled_fp8_e4m3_lightx2v_4step_comfyui	~15-17 GB	ComfyUI	⭐⭐⭐ ComfyUI ready

📝 Naming Convention

# Pattern: wan2.1_{task}_{resolution}_{precision}.safetensors

# Examples:
wan2.1_i2v_720p_lightx2v_4step.safetensors                          # 720P I2V - BF16
wan2.1_i2v_720p_scaled_fp8_e4m3_lightx2v_4step.safetensors         # 720P I2V - FP8
wan2.1_i2v_480p_int8_lightx2v_4step.safetensors                    # 480P I2V - INT8
wan2.1_t2v_14b_scaled_fp8_e4m3_lightx2v_4step_comfyui.safetensors  # T2V - FP8 ComfyUI

💡 Explore all models: Browse Full Model Collection →

🚀 Usage

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 (720P I2V FP8 example)

huggingface-cli download lightx2v/Wan2.1-Distill-Models \
    --local-dir ./models/wan2.1_i2v_720p \
    --include "wan2.1_i2v_720p_scaled_fp8_e4m3_lightx2v_4step.safetensors"

2. Clone LightX2V repository

git clone https://github.com/ModelTC/LightX2V.git
cd LightX2V

3. Install dependencies

pip install -r requirements.txt

Or refer to Quick Start Documentation to use docker

4. Configure inference

For the 720P I2V FP8 weights downloaded above, use the I2V FP8 configuration.

Set dit_quantized_ckpt to the local wan2.1_i2v_720p_scaled_fp8_e4m3_lightx2v_4step.safetensors file and set size to [720, 1280]. Choose an attention backend supported by your GPU and installed environment, and enable CPU offload if needed.

For other tasks or weight formats, see the Wan inference guide.

5. Run inference

Edit the I2V FP8 launch script:

• Set lightx2v_path, CUDA_VISIBLE_DEVICES, and the input/output paths.
• Set model_path to the complete Wan2.1-I2V-14B-720P base model directory.
• Use the FP8 configuration prepared above for --config_json.
• Change --size 480 832 to --size 720 1280.

Run from the LightX2V repository root:

bash scripts/wan/inference/distill/run_wan_i2v_distill_fp8.sh

Documentation

• Quick Start Guide: LightX2V Quick Start
• Complete Usage Guide: LightX2V Model Structure Documentation
• Configuration Guide: Configuration Files
• Quantization Usage: Quantization Documentation
• Parameter Offload: Offload Documentation

Performance Advantages

• ⚡ Fast: Approximately 2x faster than ComfyUI
• 🎯 Optimized: Deeply optimized for distilled models
• 💾 Memory Efficient: Supports CPU offload and other memory optimization techniques
• 🛠️ Flexible: Supports multiple quantization formats and configuration options

Community

• Issues: https://github.com/ModelTC/LightX2V/issues

⚠️ Important Notes

1. Additional Components: These models only contain DIT weights. You also need:

   • T5 text encoder
   • CLIP vision encoder
   • VAE encoder/decoder
   • Tokenizers

   Refer to LightX2V Documentation for how to organize the complete model directory.

If you find this project helpful, please give us a ⭐ on GitHub