Edit: Am back this time with automatic image prompt addition, you can also add some of your prompt too.
Edit: Majority of u have commented that ur machine does not have sageattention installed, some of how have never heard of sageattention or triton. Guys u need to do some research, use chatgpt and install it. Its very important to get sage attention (its a lossless method of weight distribution while loading ur wan model), it reduces the time a lot with no penalty. Search on reddit, there are several great explanations on how to install sage attention on windows.
Edit: Launching an update for my workflow, including the native teacache node...and man on man, did the speed increase. Resolution 480 by 480, 64 clip length, fps 16 and we have this beauty. Sampler : euler
Teacache has changed the game completely. Almost lossless
Edit: Re uploaded the json workflow file. Lemme know how it goes
I have been tinkling with the WAN 2.1. I have a low end pc, 8GB VRAM RTX 4060 and 16 GB RAM. I have tried many workflows but none of them worked okay for me. For some reason, only natives worked for me, Kijai's nodes my system was not able to pull.
Within my system, I was able to run the Q5 GGUF model generating 3 seconds video within 14 seconds. The quality of the I2V was quite okay, considering my low end pc. I used it to make insta reels here.
Only 3 clips within this reel is from Minimax, rest all from my system using the below workflow. Try it on your system if you have low VRAM 8GB. You need to download the GGUF models.
