I discovered a bug in one of the tools used in my textual inversion training process, which actually means everything released so far is not as accurate as it could have been! This explains why the eyes and other features were sometimes so bad. The flip side of this is now I'm struggling with overbaking the new versions, so I'm still learning how to improve on these but I believe these v2 versions (more to follow for the other released embeddings) are an improvement over the original. Will also stop Mercy having that double ponytail she sometimes had on v1!
I've included some comparison shots to show the difference
I discovered a bug in one of the tools used in my textual inversion training process, which actually means everything released so far is not as accurate as it could have been! This explains why the eyes and other features were sometimes so bad. The flip side of this is now I'm struggling with overbaking the new versions, so I'm still learning how to improve on these but I believe these v2 versions (more to follow for the other released embeddings) are an improvement over the original. Will also stop Mercy having that double ponytail she sometimes had on v1!
I've included some comparison shots to show the difference