@Shadyy @dailyvdz @Emma Morgan
I am using SD1.5 inpainting model, mask the image and set denoise to 1, which gives an OK output. after which I move the image to f222 and set denoise to 0.2 and run the image with same prompt. Some images give better output on first try. Some don't work at all. Some images take up to 10 to 15 tries like the 1st Nayanthara pic. Tamanna and Anushka shetty images took around 2 or 3 tries.
Prompt that works for me best most of the time "naked, nude, breasts, nipples" and other specifics depending on the image. Once you get the basic body shape, breasts correct, you have to work on the details separately like nipples, arms, fingers etc.
There is another way to create fakes using stable diffusion. To actually train the model using actress faces and bodies. If that works, you don't have to worry about inpainting and removing dresses. You can just use promps to make the person do whatever you want. Limit is your imagination.
I am working on training faces.
Some one elses trained model of kriti sanon
https://civitai.com/models/5069/kriti-sanon. It is not that good, but it will get better with time.
Desifakes should consider allowing the community to share these models in this forum, as training a model might be time and energy consuming.