Is there a way to add a person into an image (as if they were there in the first place), except for cut and paste solutions? Ideally something that would involve feeding that person’s image(s) onto some AI engine and then letting the software manipulate the target image so that the person is pasted there more naturally. A bit like how deepfakes work
This is not for illegal purposes, if this is somehow considered unethical the thread can be locked.
Do the edit in photoshop then run Stable Diffusion on the result, in img2img mode. Tune the settings and the text prompt (which probably needs to contain a bunch of photography and camera-related terms and some general description of the setting. It’s possible to make it do very small changes on a picture, particularly with skilled use of the masking tool. For example it could smooth any rough edges around your edit, or erase the shadows and then draw new ones in a more consistent manner. It can also completely change the faces, erase people, add more people, and, most commonly, mutilate any visible hands.
Alternatively you could just use the deepfake method on a single frame, if there’s already another person present in the image you could replace.
I didn’t even know the existence of this software. How far have we come?
Do you know if there is a way to run stable diffusion on a desktop with onboard graphics? I seem to find mixed answers
Not really, no. You want a fairly beefy graphics card, preferably an Nvidia since it’s easier to set up with CUDA than with ROCM-pytorch.