>>3027400>>3027401The model is only 12 GB and the text encoder is 8GB. Although that seems like they'd require a 24GB GPU together, I believe the workflow loads them one at a time so as long as you have at least 12GB/16GB I think you'd be good.
Does anyone in the thread have a 16GB VRAM card they can test with?
Pic related is a test showing ZImage can understand phasing like "white panties with a pink heart" and get the colors and location correct.