2024.04.17
更新一个相对于之前更稳定的模型。
****************************
2024.04.16
经自己这几天的测试，感觉模型还是欠拟合，请等待后续版本的微调模型。
****************************
本人使用 SDXL 架构训练的第一个二次元画风模型。
以下记录训练参数：
Anything XL → XL-1 : lr=1e-5, batch size = 1, repeat = 1, epoch = 4, optimizer : Lion
XL-1 → XL-2 : lr=1e-5, batch size = 2, repeat = 2, epoch = 2, optimizer : Lion
XL-2 → XL-3 : lr=1（即初始值1e-6，最终稳定在3.4e-6左右）, batch size = 1, repeat = 1, epoch = 15, optimizer : DAdaptLion
XL-3 → XL-4 : lr=1（一直稳定在1e-6） , batch size = 1, repeat = 1, epoch = 8, optimizer : DAdaptLion
分辨率建议 968*1536左右，建议大于1024*1024，采样器选择euler最佳。
提示词基本上兼容sd1.5架构。由于底模原因，您可能需要添加 "masterpiece" 等质量相关提示词。
欢迎尝试。
Description
FAQ
Looks like we don't have an active mirror for this file right now.
CivArchive is a community-maintained index — we catalog mirrors that volunteers upload to HuggingFace, torrents, and other public hosts. Looks like no one has uploaded a copy of this file yet.
Some files do get recovered over time through contributions. If you're looking for this one, feel free to ask in Discord, or help preserve it if you have a copy.