
A 3 way blended merge of 3 families of drawing stylisation I've trained, namely lineage in "Western Ink" "Hedcut" and cutting room floor colouring book experiments, all yielding this LoRa "GAz Ink"
Preface prompts with ink drawing to get it going and it'll pull the aesthetic towards a black and white inking, in the tradition of western media comic books, with some minor pull from the halftoning of the hedcut model and colouring book training.
Tested and working decently in photo realism models along the lines of Realistic Vision and Juggernaught, with some testing also done in Dreamshaper 8 and Mistoon 2.
It seems to handle external lora influence quite well when inclined toward the ink aesthetic, if it pulls away too much or if colour starts leaking, try weigh up things like "black and white" "monochrome" "black ink" etc to force it down the path you want.
Works best at CFG 5-8 weighted between 0.8 and 1
描述:
Early stages model.
Has decent success at full strength in most realism models, as well as a few tested anime merges, though they tend to leak more, but I guess it could be a VAE+Clip Skip thing
Does not need to be invoked with a tag most of the time but can be steered back with wnkrt if needed.
More polished versions to come.
训练词语: wnkrt
名称: WesternInkX.safetensors
大小 (KB): 73841
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success