Intro
This is a basic workflow for use in conjunction with WAN video, Sage Attention, and the new teacache support. Basic tests with 30 samples brought my generation speed down from 17 minutes to around 6. This is a massive upgrade in speed. A few people have had trouble installing it, so I thought I would post my workflow here to help out.
Note - this is the Kijai version. Models used are different than the Comfy supported ones. Click here for a link to Kijai's walkthrough with huggingface links and example workflows
To download the Teacache node:
go to your custom_nodes folder and type:
git clone https://github.com/welltop-cn/ComfyUI-TeaCache
Sage Attention
I am aware that Sage Attention is a pain to set up, and I do not yet have a guide here to do so. I can offer advice though - make sure that your Nvidia SDK, Torch version, Python Version, and sage attention version are lined up and compatible. This is where I had most of my issues. Start with Sage Attention, check compatibilities, and work backwards from there to get it.
I did dig around on Youtube and found a good tutorial on how to get it running (the speaker is not me). Keep in mind again that your versioning really matters - when I set it up, I had to downgrade my NVidia SDK version so everything was compatible.
Personal Tests
My test with this setup on a GTX4090+64 GB RAM (30 steps)
100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 30/30 [05:56<00:00, 11.89s/it]
My test using SDPA instead:
100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 30/30 [17:27<00:00, 34.91s/it]
6 minutes, versus 17.5 minutes. Almost 3x faster isn't bad.
Description
FAQ
Comments (24)
my WanVideoTeaCache is missing :(
oh no! I will add it to the instructions - go to your custom nodes folder, go to the command line by clicking on your folder name and typing cmd - then type:
git clone https://github.com/welltop-cn/ComfyUI-TeaCache
restart your comfyui and it should work. I will see if Comfy has a better one, too.
@_raydeStar Thanks i got it
@_raydeStar This is not Kijai's teacache you are linking or referring to.
This is the "official" teacache, passed through by another author.
@azeli ahh! you're right! Kijai's github also links back to it, so I am going to keep it up. I am exploring other options to see if there is much of a speed difference
@_raydeStar Native is better generally, just update your title
How are you getting such smooth video? All of mine are coming out very strange, blocky animation, color filter type overlays sliding across randomly etc.
Make sure you have all the Kijai models loaded. When Comfy released their own version, everything was new - even the VAE. It should be accessible through the Kijai Github that I provided.
@_raydeStar ok i will try. tx
how to use this without sage attention?
On the video model loader, you're going to want to click where it says sage_attn and change it to SDPA.
i got an oom with using SDPA thought, any advise on it?
hi,i use wan2.1 via pinokio,how can i use your file?i dont have comfyui
Okay wtf am I doing wrong ? Where do I get the nodes ? None of the nodes appear. I went to the ling, git clone, install requirements. Did this like a million times. Despite all , none of the nodes in this workflow appear, I even updated comfyui, still all red. wtf. Got 5090 , so I am using the version for that, here are all the specs.
Windows 11
NVIDIA 5090
I am using a portable version of comfyui
Python 3.12.8
VRAM 32GB
RAM 98GB
Comfy version 3.12.8 ( literaly just updated figured maybe it would help but nope ... but at least it works and nothing broken ... except the nodes aren't here)
Comfy frontend version 1.11.8
pytorch version 2.7.0.dev20250306+cu128
and yeah got sage attention installed but can't figure out where tf are these nodes XD , any tips ? Anyones guess is welcome , not just OP lol
Let's start by getting your nodes set up. We don't have evidence of incompatibility unless we can run it - I downloaded two from GitHub here - Kijai's, and the teacache node. It's also using Video Helper Suite.
I just wanted to be sure you have comfyui custom node manager installed as well. It doesn't come natively, and you have to install it. Pinokio might have it natively though.
@_raydeStar I ummm may have made a short message, I have a fully decked out comfy, was using Bizarro's workflow with teacache and sageattention, as I have said, fully updated everything, but still no signgs of the nodes for this one, not sure why. Kinda have a feeling the more of these video makers come out, the crazier it gets to setup a workflow. Anyways I will figure it out, might take a while but, if it doesn't work I just look for another source it's really no big deal
Have you tried zooming out in comfy? When I loaded this, the viewport was down and to the left of all the nodes for some reason.
Is there any way to control timeline in I2V Wan? i mean prompt scheduling? if yes, how? frustating the the AI must do all the storytelling
How to add a lora ? There is no clip input in the text encoder and no compatible model input in the sampler
Nice question 🤔🤔🤔ℹ️
There is a "WanVideo Lora Select" node in the node library under WanVideoWrapper section. Just pull the node into the workflow, select the Lroa you wish to use and connect the Lora node to the "Lora" point of the WanVideo Model Loader. If you wish to use multiple Lora just use multiple nodes and conect them together.
I generate... but when I upload to 700x700 the resolution goes down to 20 minutes... it's not worth it for me. If I generate at 520x520 or so it goes down to 8 or 9 minutes, all of that on a HDD. Thx. Pd 3090 gpu
How about this for an idea? Post a picture of the fucking workflow. Trying to steal clicks just to see the thing. piss off.
literally download the workflow and use it like a normal person