Development Workshop:
ComfyUI - WAN2.1 IMG to VIDEO
This article accompanies this workflow: link
Foreword :
English is not my mother tongue, so I apologize for any errors. Do not hesitate to send me messages if you find any.
This guide is intended to be as simple as possible, and certain terms will be simplified.
Workflow description :
The aim of this workflow is to generate video from an existing image in a simple window.
Purpose of this article :
The goal is to work together on developing new versions of the workflow and to test it in advance to improve it or avoid bugs.
Work in progress :
New interface :
New upscaler :
ComfyUI Upscaler TensorRT replaces the native upscaler for a gain in speed and stability.
In addition, the models are automatically downloaded if they are not present.
New model optimisation :
Please note this requires Triton to be installed.
New LoRA loader :
More simple, easy and convenient.
New model loader to unload part to RAM:
You can easily choose how many GB you want to move to RAM and thus either use a larger model or gain stability at a very small loss in speed.
Known issue :
The new "Upscaler TensorRT" node sometimes does not install correctly.
You must therefore install a specific python module:
.\python.exe -s -m pip install wheel-stub
If you have already installed the node you can then install the prerequisites from the node folder:
..\..\..\python_embeded\python.exe -s -m pip install -r requirements.txt
Torch compile dont work :
Download triton depending on your python version, and put it in the python folder :
Open a terminal in this folder and run :
.\python.exe -s -m pip install .\triton-3.2.0-cp312-cp312-win_amd64.whl
Change history :
Version 2.0 b1 : New UI and new upscaler,
Version 2.0 b2 : TorchCompile added for faster generation,
Version 2.0 b3 : New LoRA loader,
Version 2.0 b4 : LoRA with TorchCompile fix try,
Version 2.0 b5 : New model loader,
Version 2.0 b6 : New optimisation menu