Until I read the title, I thought it was a LoRA of my ex-wife.

The irony of this LoRA is that the outputs look like actors with make-up appliances, which is true to the name. Actual zombies could rot into the actual face/head volume- these images show skin/rot effects clearly above the head/skull volume (I assume your training data was images of real actors wearing appliances?)

Zombies in the past = BRAINS O_O

Zombies in The Walking Dead = LIPS O,...,O

ну такое...

I, for one, quite like this one; both conceptually and as a tangible work(/frame) of translation art*... with some mighty gut-punch potential. My only critique dwells on the occasional looseness of identity/feature matching/retention (like in the examples with the clearly shaved-headed cap-sporting bearded man, whose zombie gets randomly overfitted with longish gray hair around the cap). With that said, I myself am well aware of how difficult it is to cajole Kontext Dev into reliably exact-matching identity features, having spent weeks on end trying to retune it towards various concepts and utilities (my latest/ongoing have been "statue to person" LoRAs (check em out at my HF space , near the bottom of gallery); "statue to person" has been so much harder than "person to statue"!)...

Now I'm kind of curious: How exactly are you producing your dataset? (My guess would be: you might be pulling frames of walkers directly from The Walking Dead and then using some version of Kontext to "de-zombify" them, whilst retaining same backgrounds/compositions/poses (or maybe removing the bg altogether), and then using de-zombified outputs as control data and the Walking Dead sources as primary data for training. Am I correct?)

* ..."of translation art"?! ... Yes. Imho, LoRA training, at its best, is simply that kind of art; as the artistic choices and process of a LoRA maker are most alike in their logic and spirit to those of a literary translator.