Modelpublic

lightx2v/Wan-NVFP4

sourceHugging Faceapache-2.0updated 9mo agoView on Hugging Face

72likes432downloads

Model Card

🎬 Wan-NVFP4-4Steps Models

NVFP4 Quantization-Aware Step Distillation for Blackwell Architecture

![GitHub](https://github.com/ModelTC/LightX2V) ![HuggingFace](https://huggingface.co/lightx2v/)

✨ Features

—⚡ 4-Step Inference: Dramatically accelerated end-to-end generation approaching real-time performance (tested on RTX 5090 single GPU)
—🎯 NVFP4 Quantization: Reduced memory and bandwidth usage, optimized for Blackwell architecture
—🔧 LightX2V Integration: Optimal performance and stability on the official framework
—🚀 High-Quality Generation: Maintains Wan2.1's superior video quality while achieving unprecedented speed

🚀 Quick Start

bash

# 1. Install LightX2V
git clone https://github.com/ModelTC/LightX2V.git
cd LightX2V
uv pip install -v .

# 2. Install NVFP4 Kernel
pip install scikit_build_core uv
git clone https://github.com/NVIDIA/cutlass.git
cd lightx2v_kernel

MAX_JOBS=$(nproc) CMAKE_BUILD_PARALLEL_LEVEL=$(nproc) \
uv build --wheel \
  -Cbuild-dir=build . \
  -Ccmake.define.CUTLASS_PATH=/path/to/cutlass \
  --verbose --color=always --no-build-isolation

pip install dist/*whl --force-reinstall --no-deps

# 3. Run inference
cd examples/wan
python wan_i2v_nvfp4.py   # Image-to-Video
python wan_t2v_nvfp4.py   # Text-to-Video

🎬 Generation Results

<div style="background: #f8fafc; border: 1px solid #e2e8f0; border-radius: 8px; padding: 16px; margin: 16px 0;"> <p style="font-style: italic; color: #475569; margin: 0; padding: 12px; background: white; border-radius: 6px; border-left: 4px solid #3b82f6;"> "A cinematic, hyper-realistic 3D animation, in the somber and beautiful style of Sekiro: Shadows Die Twice. In a vast field of silvery-white pampas grass, under a luminous full moon, the shinobi Wolf stands ready for a final duel..." </p> </div>

<table style="width: 100%; border-collapse: collapse; margin: 20px 0;"> <tr> <th style="text-align: center; padding: 12px; background: #f1f5f9; border: 1px solid #e2e8f0; font-weight: 600;">Input Image</th> <th style="text-align: center; padding: 12px; background: #f1f5f9; border: 1px solid #e2e8f0; font-weight: 600;">Wan2.1-I2V-14B-480P</th> <th style="text-align: center; padding: 12px; background: #f1f5f9; border: 1px solid #e2e8f0; font-weight: 600;">wan2.1i2v480pnvfp4lightx2v4step</th> </tr> <tr> <td style="text-align: center; padding: 12px; border: 1px solid #e2e8f0;"> <img src="https://cdn-uploads.huggingface.co/production/uploads/680de13385293771bc57400b/9lybVJ9QSkbNC4QiP1ygo.png" style="max-width: 200px; height: auto; border-radius: 6px;"> </td> <td style="text-align: center; padding: 12px; border: 1px solid #e2e8f0;"> <video controls style="width: 260px; height: 180px; border-radius: 6px; object-fit: cover;" src="https://cdn-uploads.huggingface.co/production/uploads/680de13385293771bc57400b/jA3eRiYWjBAif6PDnx_Q.mp4"></video> </td> <td style="text-align: center; padding: 12px; border: 1px solid #e2e8f0;"> <video controls style="width: 260px; height: 180px; border-radius: 6px; object-fit: cover;" src="https://cdn-uploads.huggingface.co/production/uploads/680de13385293771bc57400b/VJfHDcXEQ7zlixizKFrD7.mp4"></video> </td> </tr> </table>

<div style="background: #f8fafc; border: 1px solid #e2e8f0; border-radius: 8px; padding: 16px; margin: 16px 0;"> <p style="font-style: italic; color: #475569; margin: 0; padding: 12px; background: white; border-radius: 6px; border-left: 4px solid #10b981;"> "高对比度，高饱和度，短边构图，日落，中焦距，柔光，背光，暖色调，边缘光，中近景，日光，晴天光，一位外国白人女性的近景，她身穿黄色格子连衣裙，戴着耳环。随着仰拍镜头的上升，女子抬起头来，眼睛里含着泪水，看着前方说着话..." </p> </div>

Wan2.1-T2V-1.3B	wan2.1_t2v_1_3b_nvfp4_lightx2v_4step
<video controls style="width: 260px; height: 180px; border-radius: 6px; object-fit: cover;" src="https://cdn-uploads.huggingface.co/production/uploads/680de13385293771bc57400b/dwr0pPbtIe2fHg0hmEM5M.mp4"></video>	<video controls style="width: 260px; height: 180px; border-radius: 6px; object-fit: cover;" src="https://cdn-uploads.huggingface.co/production/uploads/680de13385293771bc57400b/cm-S4EaZlCOShlXxOnJ-3.mp4"></video>

⚡ Performance Comparison

Test Environment: RTX 5090 Single GPU | LightX2V Framework

<table style="width: 100%; border-collapse: collapse;"> <tr> <td style="vertical-align: top; padding-right: 20px;"> <h4 style="margin: 0 0 15px 0;">📸 Image-to-Video (I2V-14B-480P)</h4> <table style="width: 100%; border-collapse: collapse;"> <tr> <th style="text-align: left; padding: 8px; border-bottom: 2px solid #e2e8f0;">Metric</th> <th style="text-align: left; padding: 8px; border-bottom: 2px solid #e2e8f0;">Original Model</th> <th style="text-align: left; padding: 8px; border-bottom: 2px solid #e2e8f0;">Optimized Model</th> <th style="text-align: left; padding: 8px; border-bottom: 2px solid #e2e8f0;">Speedup</th> </tr> <tr> <td style="padding: 8px; border-bottom: 1px solid #f1f5f9;"><strong>Single-step Denoising</strong></td> <td style="padding: 8px; border-bottom: 1px solid #f1f5f9;"><span style="color: #64748b; font-weight: bold;">12.10s</span></td> <td style="padding: 8px; border-bottom: 1px solid #f1f5f9;"><span style="color: #2563eb; font-weight: bold;">3.40s</span></td> <td style="padding: 8px; border-bottom: 1px solid #f1f5f9;"><span style="background: #16a34a; color: white; padding: 4px 8px; border-radius: 12px; font-weight: bold;">3.5x</span></td> </tr> <tr> <td style="padding: 8px;"><strong>End-to-End</strong></td> <td style="padding: 8px;"><span style="color: #64748b; font-weight: bold;">498.90s</span></td> <td style="padding: 8px;"><span style="color: #2563eb; font-weight: bold;">17.65s</span></td> <td style="padding: 8px;"><span style="background: #16a34a; color: white; padding: 4px 8px; border-radius: 12px; font-weight: bold;">28x</span></td> </tr> </table> </td> <td style="vertical-align: top; padding-left: 20px;"> <h4 style="margin: 0 0 15px 0;">🎬 Text-to-Video (T2V-1.3B-480P)</h4> <table style="width: 100%; border-collapse: collapse;"> <tr> <th style="text-align: left; padding: 8px; border-bottom: 2px solid #e2e8f0;">Metric</th> <th style="text-align: left; padding: 8px; border-bottom: 2px solid #e2e8f0;">Original Model</th> <th style="text-align: left; padding: 8px; border-bottom: 2px solid #e2e8f0;">Optimized Model</th> <th style="text-align: left; padding: 8px; border-bottom: 2px solid #e2e8f0;">Speedup</th> </tr> <tr> <td style="padding: 8px; border-bottom: 1px solid #f1f5f9;"><strong>Single-step Denoising</strong></td> <td style="padding: 8px; border-bottom: 1px solid #f1f5f9;"><span style="color: #64748b; font-weight: bold;">2.00s</span></td> <td style="padding: 8px; border-bottom: 1px solid #f1f5f9;"><span style="color: #2563eb; font-weight: bold;">0.70s</span></td> <td style="padding: 8px; border-bottom: 1px solid #f1f5f9;"><span style="background: #16a34a; color: white; padding: 4px 8px; border-radius: 12px; font-weight: bold;">2.9x</span></td> </tr> <tr> <td style="padding: 8px;"><strong>End-to-End</strong></td> <td style="padding: 8px;"><span style="color: #64748b; font-weight: bold;">83.50s</span></td> <td style="padding: 8px;"><span style="color: #2563eb; font-weight: bold;">6.54s</span></td> <td style="padding: 8px;"><span style="background: #16a34a; color: white; padding: 4px 8px; border-radius: 12px; font-weight: bold;">12.8x</span></td> </tr> </table> </td> </tr> </table>

⚠️ Notes

System Requirements

—Required Hardware: NVIDIA RTX 50-series GPUs (RTX 5090/5080/5070/5060) or other Blackwell architecture GPUs

Dependencies

—Prepare T5 / CLIP / VAE components yourself (same as Wan2.x structure)

Performance Tips

—Use Blackwell + NVFP4 for best performance
—Enable CPU offload for GPUs with limited memory

🤝 Community

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

If you find this project helpful, please give us a ⭐ on [GitHub](https://github.com/ModelTC/LightX2V)

</div>