Reference to video

Reference-to-Video Upload Rules: 9 Image Slots, Positional Tags, and a Binding That Expires

Reference-to-video takes three arrays and a text field. The tags in your prompt are array indices rather than labels you choose, the ceiling is 12 files across all modalities, and the binding expires with the request. Here is the full input contract, the notation that differs on every surface documenting it, and the two jobs it will not do.