H3 can already do this. The socket is right there on the node and almost nobody plugs anything into it.
Normally you hand the model one picture and it decides where the shot goes. Here you hand it both ends: the frame you want it to start on, and the frame you want it to finish on. It makes the journey between them.
How to use it
Load your first picture into the first image slot.
Load your last picture into the
last_frameslot.Write what happens between them, not what the picture looks like.
Run it.
The one rule that catches everybody
Both pictures have to be the same pixel size, and the clip has to be that same shape too.
If they do not match, nothing warns you. It does not crop and it does not letterbox. It stretches, and your subject quietly comes out wider or taller than she should be. I shipped that mistake once before I understood it, so the workflow now renders at the same shape as the demo pictures rather than a default widescreen size.
A trick worth knowing
For a smooth push in, use the same picture twice: once as it is, and once cropped tighter and scaled back to the identical size. You get one continuous move on one face.
Two different people as your two ends is the classic way to get a face that morphs into someone else halfway through. If that is not what you want, keep it to one.
What is in the file
Stock ComfyUI H3 nodes only. Nothing custom, nothing of mine you would have to install. Drop it in and it runs.
My other free workflows and the written guides are over here if you want them.
Description
First and last frame, at 608 x 1088. That size is on purpose: it matches the shape of the demo pictures, because a pair that does not match the render shape gets stretched rather than cropped.