This is somewhat outdated now that ComfyUI support is available. You can check out my nf4 quants at this link if you want to use it:
https://civitai.com/models/1299436?modelVersionId=1466629
If you don't like comfy, my Gradio GUI works well, so see below. :)
Demo video here.
Just wanted to quickly get the word out about this. I've forked the Wan2.1 code and done some memory optimizations on it, added automatic quantization, etc, and gotten it to work on my 4090 running the 14B i2v model.
My fork is here:
https://github.com/envy-ai/Wan2.1-quantized/blob/optimized/README.md
See the instructions at the very top of the readme.
Note that this is for technical users who are familiar with the command line, including anaconda.
This is licensed Apache 2.0, same as the original.