
训练集来自儀式 (@gishiki_unc),这次更新可以画出真实背景+动漫角色,这一次更新我同时使用了原版的Illustrious-XL模型和以Illustrious-XL为base model训练的JRU2.5模型作为base model训练,具体的效果我跑了xy表格呈现给大家,可以看出,作为base model时,Illustrious-XL和JRU2.5没有显著差异,使用JRU2.5训练的lokr作用在Illustrious-XL时没有显著效果,但是使用Illustrious-XL训练的lokr作用在JRU2.5上时,会有别样的风格,不过以演示中的权重来看,画面会比较脏,或是较低的权重能够让效果有所改善,
The training set is from 儀式 (@gishiki_unc). This update enables the model to generate real backgrounds with anime characters. For this update, I used both the original Illustrious-XL model and the JRU2.5 model trained from Illustrious-XL as base models. I presented the results using an XY grid for comparison. It shows that there is no significant difference between Illustrious-XL and JRU2.5 when used as base models. The lokr trained with JRU2.5 has minimal effect on Illustrious-XL, but the lokr trained with Illustrious-XL brings a unique style when applied to JRU2.5. However, at the weights shown in the demonstration, the images appear cluttered; lowering the weights may improve the effect.
描述:
修正了偶尔人物脸会跑出画面外的问题
I fixed the occasional issue of characters' faces going off-screen.
训练词语:
名称: GUXLlokr8f-000175.safetensors
大小 (KB): 69092
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success