HardGreyWorld is built upon the shoulders of DotAnime 3.0 after a lot of rework.
My main idea was to taint and dirty the inital clean generations made by the LoRA by using subtle bleeding made by:
reducing the initial LoRA from 128 Dim to 32
merging the LoRA in 2 checkpoints (ShinV1 and Miriku V3)
two block merge of these checkpoints (one into each other)
re-extract two LoRAs through diff/svd (especially since i figured out how to do it fast on my good old laptop)
merge the resulting LoRAs using Cosine Similarity at various degree
figure out proportion to use of the two extreme (very colorful and clean, very monochrome and dirty)
merge those for a single result
It works best with simple anime style checkpoints but still deliver interesting effects with other checkpoints
It also play well as displayed with the DMD2 LoRA for quick gens (either integrated in model or added during generation).
No trigger word needed.
If the LoRA by itself does not trigger the greyscale effect with your checkpoint of choice (and if it was your goal), add greyscale and/or monochrome.
EDIT: the source LoRA vs this LoRA
