Two images in, one image out
The trick is giving the model two separate jobs. Image 1 decides what happens: the punch, the angry face, the fisheye angle. Image 2 decides who it happens to.
Why it works
"Replace", not "make"
Asking for a replacement keeps everything else from image 1: the bed, the ceiling light, the photos on the wall, the wide-lens distortion.
One job per image
The prompt says which image owns the pose and which owns the cat. Without that, the model has to guess what to keep.
A reference sheet, not a selfie
Four angles on white show the coat from every side, so the model can draw the cat at an angle the reference never shows.
Where it ended up
It went straight onto a lock screen. Same punch, a different cat.
The two sentences we paste, and the order to upload the images in. We email you a link to the full guide.
We’ll email you the link either way. The box only adds you to the newsletter. Vectis AI, Toronto, Canada.