Type | |
Stats | 5,550 |
Reviews | |
Published | May 13, 2024 |
Base Model | |
Hash | AutoV2 81123D319C |
this model follow Fair-AI-Public-License-1.0-SD ,please ignore the page information and use it under follow Fair-AI-Public-License-1.0-SD
v2.0:
①fix overly bright
②import some human preference datasets
③switching back to fp16 for retraining
④adjust some settings to match with pony,enhancing the LORA friendly.
v1.1: Fixed the issue of overly bright images. but may be import new bug.
更新版本1.1: 修复了部分图片过亮的问题,但可能引入新bug
(because the training was use the fp8 base on a 11g vram 2080ti, so there might be some problems. But currently no gpu can be used,so it's WIP that re trian use fp16.)
因为训练基于fp8+2080ti,可能引入了新的问题。不过现在没卡,以后可能会重跑
v1.0: add artist iamges for balance
v0.9: ver : only nai3 images.
训练测试的产物,后面想着跑都跑了就发了吧,肢体出错很少了,出图都是dtg+美学后自动上传,没经过过其他筛选。
This model from a series train testing,after the result look well so I upload it. It will a few body/hand mistake. small test about foot look like no bad.
All images is auto upload , from dtg + 45% pass rate aesthetic, without other filters.
模型画风来源是个人xp图50%,nai3的老五样50%,结果还行。正式训练会更换caption rules去兼容不同画风(如果有机会)
this model style 50% from me, 50% from five prompt&nai3,result is still acceptable.formal train will switch caption rules for different styles (if I have the chance)
License
This model is released under Fair-AI-Public-License-1.0-SD
Plz check this website for more information:
Freedom of Development (freedevproject.org)