Skip to main content
The Video node generates AI video clips from a prompt — on its own, or conditioned by the images, videos, and audio you connect. Two engines cover different jobs: Default for one coherent subject, Compose for combining several distinct elements or authoring multi-shot clips. Pick the engine by intent, not by how many inputs you have. Default keeps a single subject coherent; Compose gives you per-element control when a scene combines multiple products, characters, or views.
What you connect picks the route:
  • Nothing — text-to-video from the prompt alone.
  • Images — condition the clip as references or frames (see below). Up to 9.
  • Videos — act as references that condition the new clip’s look and motion. Up to 3. A connected video is not extended frame-by-frame — it steers the generation.
  • Audio — conditions the soundtrack and timing. Up to 3, and it needs at least one image or video alongside it.

​
Reference vs. frames

The image role setting decides how connected images condition the clip:
  • Reference (the default) — images are style and subject references. The clip does not start on them; instead they hold a consistent look and identity across distinct shots. This is the right choice for continuity across a sequence.
  • Frames — the clip literally starts on your first image, and optionally ends on a second. Use it to animate a specific still or to continue from a previous shot’s end frame. Connecting the same image to many shots in frames mode makes them all open identically.

​
Name your references in the prompt

In reference mode, a connected asset is only used when your prompt names it: @Image1, @Video1, @Audio1 — numbered by connection order. Reference every asset you want the engine to use:
  • Style transfer: “in the style of @Image1”
  • Character consistency: “@Image1 as the same subject across scenes”
  • Motion transfer: “@Image1’s subject performing @Video1’s motion”
  • Scored speech: “@Image1 speaker delivering @Audio1”

​
Pace multi-scene clips

One Default clip can carry several scenes. Write “Cut scene to …” between beats and “At N seconds …” to time actions — the engine honors scene cuts and timestamps, so a single generation can replace a chain of one-shot clips.
Video is the most credit-intensive node type, and longer, higher-resolution clips cost more. Failed generations are refunded automatically — see how credits work.
Once your clips are generated, wire them onward: into a Composition for editing, a Flam node as a reference, or straight into Studio for the final cut.