Ari (Jang Eun Young) from Korean dance group <a target="_blank" rel="ugc" href="https://www.youtube.com/@waveya2011/videos">WAVEYA</a>. Her sister MiU (Jang You Sun) can be found <a rel="ugc" href="https://civarchive.com/models/101366/miu-waveya">here</a>. If you want to say thanks, you can buy me a beer at: <a target="_blank" rel="ugc" href="https://www.buymeacoffee.com/frozenviking">https://www.buymeacoffee.com/frozenviking</a>Merge Model info:Some people preferred more of the look of V1, so I made a merge of V1 and V2 for those that dig the original look a little better.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.A ton of images were used, around 700'ish individual ones, hence the huge step count.The base model is a custom merge.

Ari WAVEYA

Ari (Jang Eun Young) from Korean dance group <a target="_blank" rel="ugc" href="https://www.youtube.com/@waveya2011/videos">WAVEYA</a>. If you want to say thanks, you can buy me a beer at: <a target="_blank" rel="ugc" href="https://www.buymeacoffee.com/frozenviking">https://www.buymeacoffee.com/frozenviking</a>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.

Files with SHA256: 21493b3509343cfe489e2573ea58fa91692b1d72bcbbbc9be697dad966656783