V2 has been trained with better training data, and is more flexible to use. Face is able to look around better, and model freaks out less when in outdoor environments.
If grainy or blurry, use the negative prompts "low resolution, jpeg artifacts, motion blur".
If the skin looks really plastic like, using the negative prompt "unreal engine, cg" will help out. Still has issues with being overexposed, since all their videos (where I get the frames to train) have really strong lighting.
Training method:
Model was trained over several different steps. First with close ups for "face, facial expression", then fullbody photos for "woman, body", and then a different set for "woman". There were a few more steps, but that's the basic idea.
Ari (Jang Eun Young) from Korean dance group WAVEYA.
If you want to say thanks, you can buy me a beer at: https://www.buymeacoffee.com/frozenviking
V2 info:
V2 has been trained with better training data, and is more flexible to use. Face is able to look around better, and model freaks out less when in outdoor environments.
If grainy or blurry, use the negative prompts "low resolution, jpeg artifacts, motion blur".
If the skin looks really plastic like, using the negative prompt "unreal engine, cg" will help out.
Still has issues with being overexposed, since all their videos (where I get the frames to train) have really strong lighting.
Training method:
Model was trained over several different steps. First with close ups for "face, facial expression", then fullbody photos for "woman, body", and then a different set for "woman". There were a few more steps, but that's the basic idea.
The base model is a custom merge.