Gemini Omni 1.2 Flash edits a finished clip. 15 seconds runs about $1.04.

Gemini Omni 1.2 Flash takes notes, and keeps taking them. Hand it a finished clip and ask for warmer light. Google says you get the same clip back, warmer, ready for the next note. The 15-second ceiling is nice. The notes are why you'd switch.

On most models a wrong light means a re-roll. You pay for the whole shot again and hope your actor comes back with the same face. That's reshooting the whole scene because an extra blinked.

Google released Omni 1.2 Flash on the paid tier of the Gemini API and on Gemini for business, and it's rolling into Google's own apps. What Google says changed for people making shots:

• Clips now run up to 15 seconds in one generation (1.1 stopped at 10), and you pick the length to the second, from 3 up.
• You can edit a finished clip turn by turn, changing the light or swapping the background without starting over.
• References go further. Besides stills, you can feed it motion clips of three seconds or less and audio cues. Google says faces, wardrobe and brand colors hold across camera angles.
• In the API, the 720p rate fell 30%, to $0.07 a second.

Already calling 1.1 through the API? From December 3 those requests get routed to 1.2 and billed at the new rate, so you'll be on it whether you switch or not.

Google's suggested loop is to draft cheap and only upscale the take you've already fixed with notes. On the pricing page, 360p runs about $0.023 a second, so an 8-second draft is roughly 18 cents. The same 8 seconds at 720p is $0.56, and a full 15 is about $1.04. Upscaled 1080p comes to about $0.104 a second, 4K about $0.209.

What you won't find on that page is a price for an edit. Billing counts output video at a per-second rate, plus input for whatever you send in, the clip you're editing included. So if a note re-renders the whole clip, it would cost at least a fresh take of the same length. Run two edits and look at your usage before you build a workflow around it.

We haven't run this one yet, so read it as a plan. You need a paid-tier Gemini API key; the free tier doesn't include this model.

Close-up of a man at a diner counter, late night, single pendant lamp overhead, steam rising from a coffee cup, locked-off camera on a 50mm lens, the sign behind him reads OPEN in red.

1. Open Google AI Studio and pick gemini-omni-1.2-flash. Set 8 seconds at 360p. Paste the prompt. That take is about 18 cents.
2. Send one note: "Warm the lamp to tungsten. Keep everything else the same." Budget another 18 cents plus input.
3. Scrub the two clips side by side. Watch his face, the steam and the letters on the sign.
4. Pass: only the lamp changed. Fail: write down the first frame that drifted. Either way, under 50 cents bought you more than a demo reel shows.

No budget? Google says Vids makes 15-second clips with this model at no additional cost on a Google account. That's enough to judge the look, but not the edit loop. We haven't checked which account types see it yet.

• These are Google's demos. Demos show the best take. Nobody demos take 14.
• Google names its own weak spots: fast sports action, rapid script changes and complex physics. Don't pitch a client a skateboard trick on this yet.
• Every output carries SynthID, Google's invisible watermark. Fine for most work; worth knowing before you hand files to a client.
• Fifteen seconds is a long leash. The longer the shot, the more room a face has to drift. If the beat doesn't need 15 seconds, generate 6 and cut on the action.

Use it for product spots and single beats you'll polish. Skip it for multi-shot scenes until someone has pushed it across cuts.

Holding one face across five shots is prep work you do before the first prompt: a character bible and a set of reference views. The Bootcamp's Continuity Studio walks you through building both. See what's inside.