I've spent the last weeks trying to optimize Image gen with Wan 2.2. I've tried many Workflows but none of them looked "real". Most importantly, most of the WFs I've tried weren't able to add small details or fine tune in the image.

This WF is based on Clownshark sampler. It does use both High And Low model. Afaik images can be generated on LOW model only, HIGH is not mandatory. However, adding the HIGH model into the mix with this setup definitely improves the quality.

After the image has been generated, it is passed into a third sampler for a second pass to inject more details.

As for LORAS, in my setup i use LOW Character LORA on both High and Low.

If you find this helpful, a donation is always appreciated

Let me know if there's any issue. I really hope that as many people as possible test it so that i can find the flaws on my setup and work on it

The wf is kept as simple as possible on purpose so that you can use it to create more complex ones.