佟丽娅288分辨率直播模型带底丹带src素材
本帖最后由 随风漂流 于 2024-10-1 22:15 编辑模型仅供学习娱乐，请勿用于商用，训练美女模型主要是个人爱好
清晰度是有保障的，带遮罩效果
使用 deepfacelab-MVE 训练的,模型复用请使用 MVE版本
模型下边牙齿效果很好，清晰可见
关于相似度，换脸主要是看脸型，人脸美颜开的越高，换脸后相似度越低
==---------------- Model Options ----------------==
== ==
== resolution: 288 ==
== face_type: wf ==
== models_opt_on_gpu: True ==
== archi: liae-udt ==
== ae_dims: 480 ==
== e_dims: 72 ==
== d_dims: 64 ==
== d_mask_dims: 32 ==
== masked_training: True ==
== eyes_prio: False ==
== uniform_yaw: False ==
== 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: False ==
== pretrain: False ==
== autobackup_hour: 0 ==
== write_preview_history: False ==
== target_iter: 0 ==
== random_flip: True ==
== batch_size: 16 ==
== use_fp16: False ==
== retraining_samples: False ==
== mouth_prio: False ==
== blur_out_mask: False ==
== adabelief: True ==
== loss_function: SSIM ==
== random_hsv_power: 0.0 ==
== random_downsample: True ==
== random_noise: False ==
== random_blur: False ==
== random_jpeg: True ==
== random_shadow: none ==
== background_power: 0.0 ==
== random_color: False ==
== cpu_cap: 8 ==
== preview_samples: 4 ==
== force_full_preview: True ==
== lr: 5e-05 ==
== session_name: ==
== maximum_n_backups: 24 ==
== random_src_flip: True ==
== random_dst_flip: True ==
== gan_patch_size: 36 ==
== gan_dims: 16 ==
== gan_smoothing: 0.0 ==
== gan_noise: 0.0 ==
== ==
佟丽娅直播模型带lab底丹带素材：
***本内容需购买可见***
感谢分享~~~~~~~~ mve训练的可以用原版复用的吧 jandown 发表于 2024-9-29 09:31
mve训练的可以用原版复用的吧
有些是可以的，但是如果有些参数MVE开启的情况下，好像不行，具体没怎么测试，一直用的这个论坛的MVE版本，很稳定
有些功能很好用，比如随机颜色改善光影
随风漂流 发表于 2024-9-29 09:55
有些是可以的，但是如果有些参数MVE开启的情况下，好像不行，具体没怎么测试，一直用的这个论坛的MVE版本 ...
我用过几个MVE跑过的丹，都是可以的，默认原版不会加载MVE新增的参数 这是什么？ 角度全吗，一共多少张？
页:
[1]