Renders one still from a scene prompt, so a video prompt can be judged by looking at it rather than by reading it. Rendering video is not what this does.
Key frame preamble defaults
A video prompt describes a moving shot; an image model draws one instant. This preamble is what bridges the two, and it goes in front of the scene prompt unchanged, so the image still comes from what the orchestrator actually wrote. It is saved with the run and hashed, so two images are never mistaken for a like-for-like comparison when the wording changed between them.
Paste a video prompt from a comparison, or write a scene yourself. Coming from a comparison switches the preamble on; typing your own switches it off, since a hand-written scene already describes a still. Every render is saved and gets a shareable link that does not expire.
0 characters
Each model renders one image. Resolution defaults to the cheapest tier that model offers, and models that take no resolution parameter simply do not show one.