Not every shot needs a hero model. A lot of real work — testing a concept, filling a sequence with b-roll, iterating on a hook ten times before one lands — is bottlenecked by turnaround, not by the last 5% of fidelity. Gemini Omni Flash is built for that lane: prompt in, short clip out, in seconds, at a fraction of the cost of a premium tier. The win is throughput.

It also collapses four jobs into one model. Text-to-video, animating a still, generating from reference images, and editing an existing clip are usually four different tools; here they're four ways to use the same endpoint — the mode is chosen by what you give it. Pacing and audio live in the prompt rather than in toggles, which keeps the control surface tiny: one well-written line is the whole brief. On GenFire it runs in the browser with no Google Cloud setup, and the same model is one call away through the API, MCP, and CLI.