Modescope's swapper model alternative to inswapper?

[jp7189]

It seems the entire face swapper community is held back by the inswapper_128 low res model. I was looking at facechain and I noticed they seem to be using a different model for the swapping portion. Can any benefit be derived for use in Roop Unleashed? As far as I can tell (through machine translation) this model is open source.
Here's a link: https://www.modelscope.cn/models/iic/cv_unet_face_fusion_torch/summary

[C0untFloyd]

Well, that lowres output was just worked around by the subsampling method invented by hillobar from rope. It works good enough for me.
Anyway, I found this older reddit thread googling that model:
https://www.reddit.com/r/StableDiffusion/comments/16h0nv6/is_this_a_new_chinese_face_swapping_model/
especially https://www.reddit.com/r/StableDiffusion/comments/16h0nv6/comment/k0e5sup/
doesn't look that good to me.
I think if it would be any promising the facefusion guys would have already included it. The only other model which IMHO can partly compete with the insightface model is the ghost_unet1 one (and it has a native resolution of 256x256px).