AI video agent vs a single-model video generator: what you call

A single-model generator turns a prompt into one clip, so you do the planning, cutting and audio yourself. A video agent takes a brief and decides which generation tools to call, then assembles the result. Runway's announcement describes Agent as going from idea to a finished, ready-to-publish video in a single conversation; on Sume the same split is generate_video for one model job and an Agent Completions or Format run for the assembled video.

The two shapes

When the single call is right

• You already have a storyboard and want a model to render each shot.
• You are comparing models on one prompt.
• You need the cheapest possible preview of a single shot.

When the agent is right

When the work is a chain of decisions, such as a script, takes, B-roll, voiceover, captions and a cut, hand the chain to a Format run. It stays inside a cap you set, and the run is unattended, so write the quality bar into the instruction.

Start with a capped run

The body below is a complete run request. Replace the instruction with your brief and keep the cap low while you learn what a run costs.

{
  "instruction": "Make a 30-second product video for a trail mug.",
  "generation_spend_cap_usd": 20
}