How do you keep a character the same person from shot to shot?
Use H3's reference to video mode. It takes reference images, video and audio in one generation, so a character turnaround sheet, a 3D layout of the room and a voice sample go in together, and the prompt says which subject is in which picture.
Cite each image inside the definition of the subject it shows. A standalone <Picture N> entry acts as a shot anchor, and the one clip written that way lost the character's coat.
<Picture N> is numbered by the order images are wired into the node, not by anything in the prompt. H3 also keeps no voice between clips, so attach the same short voice reference to every clip where that character speaks.



























