Bearded faces aren't pretrained, either expose the datasets so that we can train it

BadToBest / EchoMimic

Lifelike Audio-Driven Portrait Animations through Editable Landmark Conditioning

https://badtobest.github.io/echomimic.html

Apache License 2.0

2.35k stars 274 forks source link

Bearded faces aren't pretrained, either expose the datasets so that we can train it #132

Open rahuljustbaat opened 1 month ago

rahuljustbaat commented 1 month ago

Please either release datasets and how to train ourselves or please pre train bearded face as well. Attaching sample output with not so good lipsync because of lack of datasets in training(bearded,sikh muslim use cases) Issues

LipSync isn't good
Beard and mouth gets blurry
Eye blinking is also not working at default config(steps:30)
Teeth not visible in output
Quality of original picture not retained, like beard isn't retained

Input image below: rahul

@yuange250 @JoeFannie

rahuljustbaat commented 1 month ago

Lipsync not good, teeth and eye blink also not happening

https://github.com/user-attachments/assets/9baa3cdc-204e-44eb-bc37-a485cd0b718f