[Edit:
Users without a running Triton + Sage Attention Installation please use version 1.1!
Do I really need Sage Attention? Short answer: No.
Most users, especially beginners, do not need to install Sage Attention. Just use my workflow v1.1. For typical 5-7 second videos, there is virtually no advantage, and version 1.1 works perfectly.
A noticeable difference only becomes apparent with longer videos (e.g. 10 seconds). If you like experimenting, see the last chapter, ‘Installing Triton and Sage Attention,’ at the bottom of the page for detailed information. If you get it running you can test my new workflow version 1.2 sageattention.
Many thanks to everyone who tested this workflow 👍]
Quick Overview:
Optimized Wan 2.2 workflow, runs perfect on RTX 3060 12 GB VRAM GPU and 32 GB RAM
Installation of Triton und Sage Attention not required
Fast and high quality video generation with LightX2V Lora
Reduced VRAM usage with BlockSwap
Text/Image 2 Video generation in one workflow
"Easy" installation/model downloading, all necessary sources are specified
"Easy" to use workflow, clearly structured, all necessary steps are explained
A 5 Second long high quality video generation takes about 10 - 15 minutes (see below).
Short introduction:
I struggled hard to find a suitable Wan workflow that runs with 12 GB VRAM delivering good and fast results. With starting up Wan 2.2 a couple of days ago I started a new try. Since it's running satisfactorily now I would like to publish it here and I hope it well help some others too.
This workflow is based on elements of a variety of allready published workflows. My "job" was only to put things together, optimize it for a small machine and create a most simple and hopfully user or even "beginner" friendly workflow.
The facts in short:
LightX2V Lora: fast (4 to 8 step) generation with high quality outputs.
BlockSwap: To avoid "out of memory" errors. Significantly reduces VRAM usage without noticeably affecting speed. It is possible to run some other tasks on the same machine while genarating.
Upscaling and Framerate multiply: 2x fast upscale, 4x fast framerate multiply, generates very smooth and high quality video ouputs up to 1440x960 at 60fps.
Tested generation times:
As a rough guide value for RTX 3060 GPU: generating a 5 second long high quality 1440 x 960 60 fps video with 6 steps it will take:
t2v: around 10 - 12 minutes,
i2v: around 15 minutes.
Short Conclusion:
I`m not an "expert" - just a user who wants to get it running on "available" hardware.
There are many things I don't really understand. If you find mistakes or better solutions please give me a hint.
And I really hope that even "beginners" have a chance to go the first steps...
For testing/understanding/experimenting/changing the workflow:
Klick "Toggle Link Visibility" to see the links.
Move nodes to see all "covered" nodes.
For quick testing you may lower the settings for: steps, clip lenght and video resolution.
And as usual: Have Fun 🙂🙂
Installing Triton and Sage Attention:
If you like testing/experimenting or frequently generate longer videos this short guide may help you to get Sage Attention running. But at first, what you can expect:
As a rough guide value for RTX 3060 GPU: generating a high quality 1440 x 960 60 fps video with 6 steps will take:
5 second T2V: around 10 - 12 minutes with or without Sage Attention
10 second I2V:
around 34 minutes without Sage Attention
around 22 minutes with Sage Attention active
Thanks to the advise of user uzaxpro414 and his video link: Faster Wan 2.2 - Install Triton + Sage Attention (Comfy UI Guide) I was able to get the installation of Triton and Sage Attantion running. And yes, it's easier than you might think.
The installation is based on a Comfyui Portable installation with embeddet python. This has the following advantages:
You can install the portable version parallel to any existing comfyui version.
All installations/settings are done in the embedded python only - so there is no risk to ruin your working system.
Cause the embedded python works like a "new isolated system", the installation of Triton and Sage Attention is very easy finally.
OK - the cons are: you have to install and maintain a second comfyui system and the portable version might not be so convenient like the Desktop version, but on the other hand you get a perfect test invironment.
How to instal Triton and Sage Attention:
Download and install Comfyui Portable from this link: ComfyUI(portable) Windows - ComfyUI update it and get the comfyui Manager running.
Install and test Triton and Sage Attention according to this youtube guide: Faster Wan 2.2 - Install Triton + Sage Attention (Comfy UI Guide).
If anything works without errors you can use my last workflow version 1.2 sageattention.
The video is a little bit confusing in the beginning but it provides exactly the nessecery steps. Just download the txt file with the short instructions and the Triton test file from the video description - this is very helpfull. Some basic python knowledge is usefull too.
Hopfully you will get it running too - good luck 😉
Please understand that I have neither the time nor the knowledge to help you with this installation. If you run into errors try to google it.