预训练模型WF512-高达282万迭代-带遮罩带直播模型
这是来自外网的一位老兄训练的WF512预训练模型，Dfud结构，共282万迭代，附带一个152万迭代的强力遮罩和直播模型，现以搬运白菜价公开。模型前天才刚发布，我就尽快把它搬运到百度网盘了，所以国内别的地方暂时是没有的，据这位老兄说，训练这个模型整整花了他12个月的时间！
因为是WF512模型，估计大多数朋友只能暂时收藏一下，等以后升级了显卡才能正式用吧，但附带的152万的遮罩还是不错的，虽然这个遮罩不是这位老兄训练的，源遮罩来自谷歌网盘，但由于国内限制通常大部分人无法正常访问。
好了，重点说明一下遮罩，这个遮罩主要是对应液体，舌头，头发及NSFW（Not Suitable For Work，直译工作不宜，国内通常就翻译为少儿不宜了）的内容，外网预览图尺寸太大，就放一张遮罩蒙板图吧，口部的形状内容请自行脑补。
另外，WF512预训练模型预览图如下：
预训练模型，遮罩和直播模型可以分开下载，需要的朋友可以自行选择，因为是搬运所以就白菜价了，只是隐隐觉得似乎有点对不起这位老兄12个月的辛苦了。
此外，还要更新说明一下，作者好像已经开始了女性专模的训练，经测试已经可以直接合成输出杰西卡的视频。
================== Model Summary ===================
== ==
== Model name: 512DFUDJessica_SAEHD ==
== ==
== Current iteration: 2825602 ==
== ==
==---------------- Model Options -----------------==
== ==
== resolution: 512 ==
== face_type: wf ==
== models_opt_on_gpu: True ==
== archi: df-ud ==
== ae_dims: 256 ==
== e_dims: 64 ==
== d_dims: 80 ==
== d_mask_dims: 22 ==
== masked_training: True ==
== uniform_yaw: True ==
== lr_dropout: y ==
== random_warp: False ==
== gan_power: 0.1 ==
== true_face_power: 0.0 ==
== face_style_power: 0.0 ==
== bg_style_power: 0.0 ==
== ct_mode: rct ==
== clipgrad: True ==
== pretrain: False ==
== autobackup_hour: 0 ==
== write_preview_history: True ==
== target_iter: 5000000 ==
== random_flip: False ==
== batch_size: 8 ==
== eyes_mouth_prio: True ==
== blur_out_mask: True ==
== adabelief: True ==
== random_hsv_power: 0.0 ==
== random_src_flip: False ==
== random_dst_flip: True ==
== gan_patch_size: 64 ==
== gan_dims: 16 ==
== ==
==------------------ Running On ------------------==
== ==
== Device index: 0 ==
== Name: NVIDIA GeForce RTX 3090 ==
== VRAM: 21.15GB ==
***本内容需购买可见***
迭代还是不错的，只是老外的，不过价格便宜
页:
[1]