底模:SDXL

主页:https://www.liblib.art/modelinfo/acc221fc9d4e4a2394a2bc72b236ab4e

版本:Eps-pred 77%

  • 一个更新的版本,比目前Civitai的75%版本多训练了3个epoch,少图表现更好
  • A new version of traditional-para training that supports concepts around img count 150 (styles and chars), there have been 22 epoch of training on 12.7 million images so far.
  • This is still an intermediate version and can be run using the traditional A1111 Webui, meanwhile A1111 promises to support all versions of NoobAI XL (including the v-pred version) in the main branch these days
  • V-pred version and epsilon-pred version are still on the Laxhar Lab's to do schedule
  • This release is already a mid-late version that supports the concept of actual count numbers over 100, and even some characters with only 50 count! If there are characters and styles you like that meet the criteria but are not supported, it's probably because the actual data has not been collected, and Laxhar Lab welcomes your feedback to make the model better!ミ(・・)ミ

We hope you like this 77% version, and we plan to release 100% of the EPS-Ver next month, and at the same time begin development work on the full version of the V-pred release.

声明:本站所有文章,如无特殊说明或标注,均为本站原创发布。任何个人或组织,在未征得本站同意时,禁止复制、盗用、采集、发布本站内容到任何网站、书籍等各类媒体平台。如若本站内容侵犯了原著者的合法权益,可联系我们进行处理。0
发表评论 共有条评论