MY RECOMMENDATION - this is v1 so increase weight to ~1.25

Prompt: r3st0rt, two combined images, first image is always a reference, look at the identical elements in the first image and restore them in the second image if they have been changed

in short, you upload 2 combined images of the first frame and the last frame, and in order to preserve the context and nothing new pops up during the generation, this lora throws the context and looks for all the same elements from the first frame and returns them to the second

but what kind of use-case can this have, why? For what? For example, if you want to interpolate between first frame and last frame of a certain video with an inpaint/outpaint, for example v2v via ltx 2.3, or the most basic ff2lf, but with changing the object/environment, this lora gives much more stability and control

also there MUST BE combined 2 images together - first frame at the top second at the bottom