currently theres text inversion but it has its limitations and does not retain identity.
Or this is something totally different despite being a diffusion and cant be applied to SD, from what i see in the paper, the results are pretty spectacular and retaining identiy is off the charts, will this be public ?
Doyou have original 1024 output images somewhere ? results are downscaled to 796 in the resuts.png.Also do you have results on cartoon style and on human faces to see how it retains the identity and cartoon style
currently theres text inversion but it has its limitations and does not retain identity.
Or this is something totally different despite being a diffusion and cant be applied to SD, from what i see in the paper, the results are pretty spectacular and retaining identiy is off the charts, will this be public ?
Doyou have original 1024 output images somewhere ? results are downscaled to 796 in the resuts.png.Also do you have results on cartoon style and on human faces to see how it retains the identity and cartoon style