Innovate Futures @ Benji

Wan 2.2 Pusa V1.0 - Multi-Features In One LoRA Fine Tune Model.

Added 2025-09-05 13:01:48 +0000 UTC

Tutorial Video : https://youtu.be/Lu_WgtRP-0w

This video dives into running the Pusa version 1.0 LoRA model in ComfyUI, comparing its performance between the official Hugging Face release and the more streamlined WAN Video Comfy wrapper. You'll see real side-by-side tests for both text-to-video and image-to-video generation, highlighting issues of video quality. We also explore the unique video extension features of Pusa , including start and end frame handling, and test whether the model truly enhances output or if you’re better off using base Wan 2.2 alone. This is essential viewing for AI video creators, ComfyUI users, and anyone working with Wan 2.2 or Pusa models who wants honest insights before investing time or resources.

This content is suitable for AI video enthusiasts, ComfyUI practitioners, generative model researchers, and developers experimenting with text-to-video and image-to-video workflows. It matters because it provides a clear, practical evaluation of a newly released LoRA model, helping users avoid inefficiencies, unnecessary downloads, and subpar results—saving time and computational resources.

Resources:

https://yaofang-liu.github.io/Pusa_Web/

Official Repo https://huggingface.co/RaphaelLiu/Pusa-Wan2.2-V1

WanVideo Comfy Repo https://huggingface.co/Kijai/WanVideo_comfy/tree/main/Pusa

Pusa for video extension Wan Video Wrapper example

https://github.com/kijai/ComfyUI-WanVideoWrapper/blob/main/example_workflows/wanvideo_2_2_14B_Pusa_extension_example_01.json

Attached experiment workflows that used in this tutorial: