Depth-aware compositing with Flux2 Klein 9b?
I'm doing background replacement using flux2 klein 9b. Just plainly swapping the background of image 1 to image 2 works perfectly with just prompting, no mask needed. However, the background does not end up looking accurate. It is simply just swapped behind the character, it is not organically part of the scene.
For example, image 1 contains a woman sitting on a bed in a bedroom taking a selfie. Image 2 contains another bedroom. After swapping, she should end up sitting on the new bed from image 2, but instead it just ends up being in the background, while the woman is in the foreground as originally.
I tried various prompting techniques, but it doesn't seem to work. Either flux re-renders the woman actually sitting on the new bed, or just plain background swap. I don't want flux to re-render the woman, I want it to build the new background, the new bedroom around her organically, or if it's better to put it, not to put the woman on the new bed, but put the new bed under the woman. The woman's perspective, position, distance to the camera must remain absolutely the same as on the original image. so flux must figure out spacial adjustments how to build up the new bedroom around her so she is organically placed on the new bed, so pushed forward from the perspective of image 2, not just a plain background swap. Does this make sense? Can you guys help me with suggesting some solutions?
I tried to ask AI of course to give me some ideas, also tried to mask out the exact position on image 2 where she should be placed, also read something about using depth maps to bring everything together, but it just didn't make sense and I didn't find a good image-to-image tutorial for this kind of thing!
Thanks in advance!