• For Ad free site experience, please register now and confirm your email address. Advertisements and popups will not be displayed to registered users.Being a registered member will also unlock hidden sections and let you request for your favourite fakes.

Bollywood Actress High Quality AI generated celebs only

Hindi actress fakes and nude edits from Bollywood movies
Trying to teach AI non-popular models which to get more accurate skin tone, build, etc. Not the best, but closer to the actual person type than whatever base SDXL model knows about the model
Using IPAdapter: GitHub - tencent-ailab/IP-Adapter: The image prompt adapter is designed to enable a pretrained text-to-image diffusion model to generate images with image prompt.

Ankita Dave (Random IG thot):


Same prompt, same seed, 0.7 IPAdapter strength instead of 0.4:


Asheema Vardaan (DevDD, AltBalaji), 0.4 strength:
 
Last edited:
diffusedbaba said:
Vidya Balan is seriously underrated.
Some attempts with poses using controlnet
These are sometimes so realistic that I feel I'll get into trouble for making these lol


Others (pretty much a full gallery):

Click to expand...
Lovely work Bro, may I make a suggestion with Vidya Balan she doesn't look convincing as a slim woman. She is more suited to being the Voluptous type , big Boobs and wide hips thick legs. However this is merely my opinion and you are your own Artist and each one of us have our preferences.
 
wizardedits said:
There seems to be one thing that the AI model is not capturing is the height / proportions.
Like shenaz is quite average at 5'4" but looks like she's tall like 5'9" or more here.
Click to expand...
Yeah, AI as we currently have it is not deterministic. It is at best trying to co-relate things with each other based on the training data.

So, if you fed it several images of shenaz and if they had height tagged, then the AI would roughly co-relate to height, etc, but if the training data only has say.. images till her waist, etc, then it can't even guess.

So when it doesn't know anything, it'll fill in the blanks and hallucinate the closest estimate and when it fails, then you get weird issues like.. 3 hands, botched anatomy, etc.

As training data improves in the future, things should improve. SD3 looks good on that front and follows prompt very well. For now, SDXL is the best we have at least for this purpose as it's the best with bodies.
 
diffusedbaba said:
Yeah, AI as we currently have it is not deterministic. It is at best trying to co-relate things with each other based on the training data.

So, if you fed it several images of shenaz and if they had height tagged, then the AI would roughly co-relate to height, etc, but if the training data only has say.. images till her waist, etc, then it can't even guess.

So when it doesn't know anything, it'll fill in the blanks and hallucinate the closest estimate and when it fails, then you get weird issues like.. 3 hands, botched anatomy, etc.

As training data improves in the future, things should improve. SD3 looks good on that front and follows prompt very well. For now, SDXL is the best we have at least for this purpose as it's the best with bodies.
Click to expand...
Please tell us from where did you learn. I also want to learn this.
 
You must log in or register to reply here.
Back
Top