Client
Hustle
Year
2023
Train a new diffusion model based on selfies.
Summary
This project train a new diffusion model based on your face (upload 5-10 selfie). You can select defined set of “styles” for their avatar. Styles are themes ie. superheroes, painting, studio lighting, sci-fi, anime, etc. A secondary service using controlnet allow “avatar” to express defined set of emotions.
I realize that deep fakes - highly convincing as it may be might not be possible if you truly "Know thyself."
"She who sees the unseen"
When you are up close to a real photo/person, you can recognize their authenticity, and you feel comfortable in their presence. But deep fakes, often feel uncomfortable, uneasy, even if you can't pinpoint why. This is because the generated images lack the truth and authenticity that comes from really knowing someone.
SETUP
Experiments checklist
01
Train an avatar model and merge it with “portrait plus”
02
Apply styles to the “avatar” using embedding/lora
03
Apply emotions using control net trained modelsn
+182%
Week over week
+153%
Time spent
+247%
duration increased
