🎬 LightWan2.2-A14B

An extremely efficient Wan 2.2 14B variant: NVFP4 Quantization-Aware Step Distillation with Sparse Attention for Blackwell Architecture

📋 Table of Contents

• ✨ Features
• 🚀 Quick Start
• 🎬 Generation Results
• ⚡ Performance Comparison
• ⚠️ Notes
• 🤝 Community

✨ Features

• ⚡ 4-Step Inference: Two high-noise expert steps followed by two low-noise expert steps, enabling extremely fast Wan2.2 MoE generation on a single Blackwell GPU.
• 🎯 NVFP4 Quantization: Quantization-aware step distillation reduces memory traffic and compute cost while targeting Blackwell architecture.
• 🧩 Sparse Attention: Accelerates the costly O(n²) self-attention workload with sparse attention, reducing end-to-end latency for high-resolution video generation.
• 🔧 LightX2V Integration: Recommended runtime stack for stable deployment and best performance.
• 🚀 High-Quality Generation: Preserves the visual quality of Wan2.2-T2V/I2V-14B while dramatically improving inference speed.

🚀 Quick Start

Environment and weights

Follow the LightX2V environment setup and NVFP4 kernel installation. Use a Blackwell GPU with the required sparse-attention dependencies.

Download the matching T2V or I2V NVFP4 checkpoints from this repository.

In each launch script, set lightx2v_path, model_path, CUDA_VISIBLE_DEVICES, and the input/output paths. Keep model_path pointed at the complete corresponding Wan2.2 base model directory.

In the JSON selected by each script, set high_noise_quantized_ckpt and low_noise_quantized_ckpt to your downloaded NVFP4 files.

Run all commands below from the LightX2V repository root.

Single-GPU inference

Set --config_json in the corresponding launch script to the matching NVFP4 configuration:

Task	Launch script	Configuration
T2V	scripts/wan22/inference/distill/run_wan22_moe_t2v_distill.sh	configs/wan22/extreme/wan_moe_t2v_distill_nvfp4_sparse_attn.json
I2V	scripts/wan22/inference/distill/run_wan22_moe_i2v_distill.sh	configs/wan22/extreme/wan_moe_i2v_distill_nvfp4_sparse_attn.json

Prefix the configuration path with ${lightx2v_path}/ when setting --config_json in the script.

# Text-to-video
bash scripts/wan22/inference/distill/run_wan22_moe_t2v_distill.sh

# Image-to-video
bash scripts/wan22/inference/distill/run_wan22_moe_i2v_distill.sh

Multi-GPU inference

The following scripts are configured for eight GPUs and use their own _sp_parallel.json presets. Update the checkpoint paths in those presets before running.

# Text-to-video
bash scripts/wan22/inference/extreme/run_wan22_moe_t2v_extreme_sp_parallel.sh

# Image-to-video
bash scripts/wan22/inference/extreme/run_wan22_moe_i2v_extreme_sp_parallel.sh

🎬 Generation Results

"Two anthropomorphic cats in comfy boxing gear and bright gloves fight intensely on a spotlighted stage"

Resolution	Wan2.2-T2V-14B	LightWan2.2-A14B
480p
720p

⚡ Performance Comparison

Test Environment: RTX 5090 Single GPU | LightX2V Framework | End-to-End Latency

Method	Task	GPU Number	Resolution	NFE	E2E Latency	Speedup
Wan2.2-T2V-14B	T2V	1	480p	40	734.0s	1.0x
LightWan2.2-A14B	T2V	1	480p	4	9.1s	80.7x
Wan2.2-T2V-14B	T2V	1	720p	40	2668.0s	1.0x
LightWan2.2-A14B	T2V	1	720p	4	22.5s	118.7x
Wan2.2-I2V-14B	I2V	1	480p	40	787.0s	1.0x
LightWan2.2-A14B	I2V	1	480p	4	10.7s	73.9x
Wan2.2-I2V-14B	I2V	1	720p	40	2685.0s	1.0x
LightWan2.2-A14B	I2V	1	720p	4	26.7s	100.5x

⚠️ Notes

System Requirements

• Required Hardware: NVIDIA RTX 50-series GPUs or other Blackwell architecture GPUs.
• Recommended Runtime: lightx2v/lightx2v:26052801-cu130-5090.

Dependencies

• Prepare Wan2.2 T5 / VAE components following the standard LightX2V Wan2.2 model structure.
• For I2V, also prepare the required image encoder components and input image according to the LightX2V Wan2.2 I2V script.
• Use Blackwell + NVFP4 kernels for optimal speed and memory efficiency.

Performance Tips

• Use the provided extreme inference script for the 4-step high-noise / low-noise expert schedule.
• Sparse attention is most beneficial at higher resolutions where self-attention dominates latency.
• Enable CPU offload only when GPU memory is limited, since offload can reduce throughput.

🤝 Community

• 🐛 Issues: GitHub Issues
• 🤗 Models: HuggingFace Hub
• 📖 Documentation: LightX2V Docs

---

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

For questions or issues, please open an issue on LightX2V or contact lvchengtao0319@gmail.com.