Struggling to get good SRC resolution in final video

[effectivedragonParticipant]

Hi. Basically the deepfake I am creating is basically overlaying myself onto myself but with a few modifications for a project (better to do that than plastic surgery lel) and the way I have done it is basically copying a select amount of photos from the DST folder into the SRC, and unsurprisingly considering there is an identical face pose the software does not take too long to train and in terms of the facial motion, I can easily get very good results. However I do have a nagging issue in the sense that after 25,000 iterations which I am on atm, that it still is not at the full quality of the source SRC images and is not as detailed. This is accentuated by the fact I am quite freckly so its quite noticeable… I am using settings from a pretrained model, below I have posted the settings I am using. Note I have a VRAM of 8GB (using a RTX2070 SUPER) but the trainer says I have 6.53GB. How do I get to the quality of the SRC as a final finishing touch?

== resolution: 224 ==
== face_type: wf ==
== models_opt_on_gpu: True ==
== archi: liae-udt ==
== ae_dims: 512 ==
== e_dims: 64 ==
== d_dims: 64 ==
== d_mask_dims: 32 ==
== masked_training: True ==
== eyes_mouth_prio: True ==
== uniform_yaw: True ==
== blur_out_mask: False ==
== adabelief: True ==
== lr_dropout: n ==
== random_warp: True ==
== random_hsv_power: 0.1 ==
== true_face_power: 0.0 ==
== face_style_power: 0.0 ==
== bg_style_power: 0.0 ==
== ct_mode: none ==
== clipgrad: True ==
== pretrain: False ==
== autobackup_hour: 0 ==
== write_preview_history: False ==
== target_iter: 0 ==
== random_src_flip: False ==
== random_dst_flip: False ==
== batch_size: 8 ==
== gan_power: 0.0 ==
== gan_patch_size: 28 ==
== gan_dims: 16 ==

[defalafaParticipant]

don´t know your routine, but the settings are not good for 25k inter.

just schon let run just random warp first.

eyes_mouth_prio: True == should be off
uniform_yaw: True == should be off

[Author]

Posts