Gemini Omni 1.1 Flash: draft at 360p, upscale to 4K

The expensive part of AI video was never the final render. It was the eleven versions you threw away getting there. Gemini Omni 1.1 Flash shipped today with a cheap preview tier that makes those eleven takes affordable, and that is a bigger deal than the resolution ceiling.

What actually happened

Google released Gemini Omni 1.1 Flash on August 27 with four changes that matter operationally.

360p drafts. Preview generations run up to 60% faster and cost about a third of 720p. This is the whole point. Iterate at draft resolution, commit once.

Scene extension to 40 seconds. The model reads up to 10 seconds of prior context to hold visual consistency across the join, instead of restarting the scene every clip.

First and last frame control. Specify both keyframes and let the model handle the camera move between them. Deterministic endpoints, generated middle.

4K upscaling. Finish at delivery resolution without regenerating from scratch.

Inputs are multimodal: text, images, and video references up to three seconds. It is available in Google AI Studio, through the Gemini Enterprise Agent Platform API, in Google Flow for Plus, Pro, and Ultra subscribers, and scene extension is in the Gemini app for those tiers.

On price, use the published API number rather than anyone's estimate. Google's pricing page lists video output at $17.50 per million tokens, at 5,792 tokens per second of 720p video — roughly $0.10 per second of 720p. Google has not published separate per-resolution rates; the 360p savings are stated as a relative comparison, not a line item.

Why a cheap draft tier matters for your business

Run the math on your actual workflow, not the demo. A 30-second product clip at 720p is about $3 of output. Trivial. Twelve iterations to get the framing right is $36 — still fine. Now make it forty SKUs with seasonal refreshes and you are in real money, and 90% of it is spent on takes nobody will ever watch.

The draft-then-commit pattern is the fix, and it is not new — it is how every film production has worked forever. What changed is that the API finally exposes the cheap tier so you can encode it. Structure the pipeline in two passes: generate every candidate at 360p, gate on human or model review, then render and upscale only the survivors. Log the prompt and seed for the approved take so the December refresh is a re-render, not a re-discovery.

One caution before you commit a workflow: Google published a speed and cost comparison for drafts, not a fidelity guarantee. A shot that reads fine at 360p can fall apart at 4K. Validate that the draft actually predicts the final on your content before you automate approval off it.

Sources: Google Blog, Gemini API pricing.