NeoChapter3版本在版本2的基础上，在1个输入层和2个输出层进行了调整，尝试修复了肢体和远景手部崩坏情况，并且增加了细节(包括脸部以及人物躯干、衣服及场景细节)，以及对整体构图进行了改善，并且考虑到训练TE的因素，将clip换回训练时原始的clip，以便更好地贴合tag表现，能画出版本2一些无法表现的特性，有将三个版本做xy图，对比其细节构图变化，及肢体改善，放在示例图片最后一张。已上传没有嵌入VAE的版本在files里，您可以根据自己的喜好随意挂载或嵌入合适的VAE。
The NeoChapter3 version has been adjusted based on version 2, with modifications made to one input layer and two output layers. It attempts to fix issues with limb and distant hand distortion, while also adding more details, such as facial features, character torso, clothing, and scene details. The overall composition has been improved, taking into consideration the training factors of TE. In order to better align with tag performance, the clips have been reverted back to their original training state, allowing for the portrayal of certain characteristics that version 2 was unable to depict. Three versions are then depicted in an XY graph, comparing the changes in detailed composition and limb improvement. This is shown on the last example image.The uploaded version without VAE is in the files, you can mount or embed the appropriate VAE according to your preference.