MiniMax H3 reference to video in ComfyUI - keep the same person in any scene

Every setting, the model files with the folder each one goes in, and the workflows: https://forgebun.com/go/yt_h3ref Hand it one photo of a person. It gives you back a completely different shot, in a place you never photographed, with that same person in it. Your photo never appears in the result - it is read, and then set aside. This is not image to video, even though almost everyone starts by assuming it is. What we cover: the node that makes this different, and why your photo does not show up in the output why the prompt has to do far more work here than anywhere else getting a likeness that actually holds, instead of a stranger who looks a bit like them the model file this needs, which is not the one the other H3 workflows use the one dial worth knowing, and why the file ships on twenty steps the two decoders you need, and what happens when one is missing Free, no card, one page: https://forgebun.com/go/yt_h3ref There is a paid course on the portal that goes further with this model. The workflow above is free either way, and it is the whole thing, not a trimmed version. 0:00 One photo, a street she never stood on 0:12 A shot you never filmed 0:35 Not image to video 0:56 The prompt has to do more work 1:42 Getting a likeness that holds 2:36 It needs a different model file 3:09 The one dial worth knowing 3:53 Why it ships on twenty steps 4:30 Two rules that have not changed 4:59 Two decoders, and you need both 5:17 The workflow is free 5:56 Your turn