来群里扯淡吧：188546326tag:ink wash painting基于noob 1.1，一种水墨风格，通过对一张水墨画风格迁移出1500张图训练而成，可能是由于素材中部分图片背景有龙、狗子、野兽没有被识别，导致跑图经常会出现狗子（comfyui中尤其严重），后续不知道会不会改，原本ipa的话对各种画师串混合效果更好，适当降低lora或者触发词权重可以减少水墨成分,由于数据集分辨率的原因，画图时分辨率如果太高，会影响笔触返图区我放了一些用于训练的素材图，效果更好……Based on noob 1.1，An ink-wash style trained using IPA on 1,500 images based on a single ink painting. Due to some background elements in the training data—such as dragons, dogs, or beasts—not being properly recognized, the generated images often include unexpected dogs (especially in ComfyUI). It’s unclear whether this issue will be fixed in the future. Originally, with IPA, the style-mixing effect across different artists was better. Reducing the LoRA or trigger word weight slightly can help decrease the amount of ink-wash influence.Due to the resolution of the dataset, using a resolution that is too high during generation may affect the brushstrokes.I’ve placed some of the training material images in the reference section, which produce better results...

[noob 1.1] 水墨 ink wash painting

yi-noob-000005.safetensors

From the author: 来群里扯淡吧：188546326tag:ink wash painting基于noob 1.1，一种水墨风格，通过对一张水墨画风格迁移出1500张图训练而成，可能是由于素材中部分图片背景有龙、狗子、野兽没有被识别，导致跑图经常会出现狗子（comfyui中尤其严重），后续不知道会不会改，原本ipa的话对各种画师串混合效果更好，适当降低lora或者触发词权重可以减少水墨成分,由于数据集分辨率的原因，画图时分辨率如果太高，会影响笔触返图区我放了一些用于训练的素材图，效果更好……Based on noob 1.1，An ink-wash style trained using IPA on 1,500 images based on a single ink painting. Due to some background elements in the training data—such as dragons, dogs, or beasts—not being properly recognized, the generated images often include unexpected dogs (especially in ComfyUI). It’s unclear whether this issue will be fixed in the future. Originally, with IPA, the style-mixing effect across different artists was better. Reducing the LoRA or trigger word weight slightly can help decrease the amount of ink-wash influence.Due to the resolution of the dataset, using a resolution that is too high during generation may affect the brushstrokes.I’ve placed some of the training material images in the reference section, which produce better results...

Files with SHA256: 8b58c49e8b0c29eb80dfc7f29949f4ffb5eeb903e3ac2fba619d195495604379