Hirasawa Seiji style lora trained on local machine using NAI as base. Doesn't generate males very good since almost all images solo or pov scenes. Works good on NAI in my opinion (probably since optimized for NAI). First version is overtrained - use 0.5 weight, other operate on full weight fine. Sometimes hands are molten and thumbs are missing. Often draws highleg without prompting - keep this in mind.
Managed to train Lartoria, Lartria, Artoria, Altria and Morgan into model.
Lora has lots of steam and sidelocks.
描述:
changed dim and alpha to 32, added 2 new images to dataset, corrected and cleaned tags, set different lr. looks like some concepts got in as well, for some reason straight-on pov is hard to achieve. 1.0 weight works fine for me, generetes oversize booba, adjust sizes by replacing huge with medium etc.
训练词语:
名称: hizen_equaldim_lowlr-000012.safetensors
大小 (KB): 36990
类型: Model
Pickle 扫描结果: Success
Pickle 扫描信息: No Pickle imports
病毒扫描结果: Success
