Homelander in Justice League (2017) [DF-UDT HEAD + WF]

[GdaTyler]

After wasting like two weeks on an AMP HEAD attempt that wasn't up to par, due to how heavy it was, decided to just say screw it and go back to df-udt. Gotta say it's been really good. The likeness is pretty close and the expressions are much better than expected. Also pretty sharp. I think AMP HEAD has some good uses, for like swapping really dissimilar face shapes like a bald subject with a hat, or vice versa. I'm currently attempting a pretty brutal fake which is swapping a character with a hat and glasses with a short haired subject and AMP barely managed to get it done, so will try DF-UDT soon for that. If DF-UDT can handle it, I'm just going to quit AMP lol. AMP seems to be better for WF fakes imo, and while it can be pretty amazing, you really need such a good faceset and to train it to hell.

As for the masking, trying to include more of the dst's neck so that it fits a bit better, but I'm still struggling with the content aware removal of the dst. :/ I didn't really use color matching for this fake that much, just mkl-m for some scenes with blue lighting. All in all, I think this was a more decent attempt, but still needs a lot of work. Also special thanks as always to groggy for the pretrained models, I'm pretty much useless without them.

 

[Voicechange_needer]

I know this thread has been posted for a long time but I just want to ask, can you share any experience about the ears or the upper hair. The ears always cant be covered. I have tried many ways but the results is disappointed. I usually use some apps to reduce the size of the head in the dst video, after that I train with that dst and merge with face scale -4 or -5. Sometime the result is acceptable, sometime it's not.

 

[GdaTyler]

I know this thread has been posted for a long time but I just want to ask, can you share any experience about the ears or the upper hair. The ears always cant be covered. I have tried many ways but the results is disappointed. I usually use some apps to reduce the size of the head in the dst video, after that I train with that dst and merge with face scale -4 or -5. Sometime the result is acceptable, sometime it's not.

I export the full mask for the Whole-Face result and also learnedpredicted mask. In AfterEffects I first use the square mask and apply some blur and aim to cut off right above the hairline and exclude the ear for side shots. then I pre-comp it and then use the lrnd prd mask over that. It's a pain, but I feel like it's been the best way I could do it.

You must train using HEAD facetype though to get the ears in the training area. And you can use that same HEAD faceset when you train again with a WF model. I'd recommend leaving facescale as is, it's usually pretty close for SAEHD.

 

[Voicechange_needer]

Thank you so much about the advices, and yes I usually train HEAD first and WF after to make the face more detailed. Your process is complicated, explained why your videos has that good quality. I will try later. Thanks again.