Apps · Desktop app

WanGP

by deepbeepmeep · See repo · 11,277 GitHub stars

Video generation “for the GPU poor”: runs Wan, LTX-Video and HunyuanVideo on cards with as little as 6–8 GB by offloading cleverly.

WindowsLinux

Get WanGP

At a glance

Best for
Making video on a mid-range graphics card.
Inference engine
  • PyTorch
GPU support
  • NVIDIA CUDA
  • AMD ROCm
Opens
  • Safetensors (original weights)
Features
  • Chat window
  • Web chat UI
  • Offloads to RAM on small GPUs
Platforms
Windows and Linux, best on NVIDIA.

Install

Windows

git clone https://github.com/deepbeepmeep/WanGP

Linux

git clone https://github.com/deepbeepmeep/WanGP

21 models on this site run in WanGP

Model Made by Type Size Opens it as Quality tier
LTX-2.5 (22B, video + audio) Lightricks Video 21.0B Built-in support 88
LTX-2.3 Lightricks Video 18.9B Built-in support 86
Wan 2.2 I2V A14B Alibaba Video 14.0B Built-in support 86
Wan 2.2 T2V A14B Alibaba Video 14.0B Built-in support 86
HunyuanVideo 1.5 (8.3B) Tencent Video 8.3B Built-in support 84
LTX-2 (19B, video + audio) Lightricks Video 18.9B Built-in support 82
Wan 2.1 I2V 14B 720P Alibaba Video 14.0B Built-in support 78
Wan 2.1 T2V 14B Alibaba Video 14.0B Built-in support 78
Wan 2.1 VACE 14B Alibaba Video 17.3B Built-in support 78
Wan 2.2 Animate 14B Alibaba Video 17.3B Built-in support 78
HY-OmniWeaving (8.3B, unified video) Tencent Video 8.3B Built-in support 76
Wan 2.2 S2V 14B (speech to video) Alibaba Video 16.3B Built-in support 76
Wan-Dancer 14B (music to dance) Alibaba Video 17.3B Built-in support 76
Wan 2.1 I2V 14B 480P Alibaba Video 14.0B Built-in support 74
HunyuanVideo 13B Tencent Video 13.0B Built-in support 72
HunyuanVideo I2V 13B Tencent Video 13.0B Built-in support 72
Wan 2.2 TI2V 5B Alibaba Video 5.0B Built-in support 70
LTX-Video 13B 0.9.8 (distilled) Lightricks Video 13.0B Built-in support 68
Wan 2.1 T2V 1.3B Alibaba Video 1.3B Built-in support 55
Wan 2.1 VACE 1.3B Alibaba Video 2.2B Built-in support 55
LTX-Video 2B Lightricks Video 1.9B Built-in support 50

Alternatives

Which app should I use?

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.

Was this page helpful?