Google Just Taught Gemini to Make Videos From a Chat: 9 Tricks to Try First

Google just changed who gets to direct a video. On 5 November 2026, Google DeepMind made Gemini Omni 1.2 Flash generally available, and the new model does something older video tools never did. It lets you make a video by talking to it.

You type a sentence. You get back a short clip with sound. Then you keep talking. Change the background. Warm up the light. Swap the product. The model edits the same clip instead of starting over, and the rest of the scene stays put. That is the whole idea, and it is a bigger shift than the longer clips or the lower price.

Here is what actually shipped. Gemini Omni 1.2 Flash is a video generation and editing model from Google DeepMind. It takes text, images, and video as input, and it returns video with native audio. A single clip runs from 3 to 15 seconds, in one second steps, up from a 10 second ceiling in the previous version. At 720p, Google lists the price at $0.07 per second, which is 30 percent less than Omni 1.1 Flash. A full 15 second clip at that rate costs about $1.05 before retries or edits.

It is also important to say what this is not. This is full video generation with sound, not a silent looping GIF. The model builds moving images and audio together, and every output carries an imperceptible SynthID watermark so the clip can be checked as AI made. If you have only ever made GIFs or still images with an AI tool, this is a different machine.

Where can you actually use it? Google says the model is rolling out across its products and developer platforms. That includes the Gemini API and Google AI Studio for developers, Google Cloud for businesses, plus Flow, Google Vids, Google Ads, and the Gemini app, where 15 second videos are rolling out to Google AI Ultra subscribers. Access varies by product and region, so check the tool you already use before you plan a project around it.

The nine tricks below come straight from what Google published in its launch announcement and the model documentation. They cover how to prompt, how to edit, how to keep costs down, and where the hard limits sit. Try them in order on your first clip, and you will learn the tool in one afternoon.

Omni 1.2 Flash at a glance

15 seclongest single clip$0.07per second at 720p6reference images maxDraft at 360p or 720p, upscale winners to 1080p or 4KEvery clip carries an invisible SynthID watermark

Source: Google DeepMind launch announcement, 5 November 2026.

1. Talk to Your Video Like a Director

A professional cinema camera on a tripod filming a cozy living room set with a sofa and warm lamps

Trick 1: edit by chatting, never by starting over. The headline feature of Omni 1.2 Flash is multi-turn conversational editing, and it works across full 15 second clips. You generate a clip, look at it, and type the fix in plain language. For example, you can swap a product or a background, change the lighting or weather, or switch the aspect ratio. The model revises the same clip and keeps everything else.

Think of it the way a director talks to a crew. You do not rebuild the set because the lamp is wrong. You say the lamp is wrong, and someone fixes it. Before this release, a bad background meant a fresh prompt and a fresh roll of the dice. Now the background is one sentence away.

Trick 2: change one thing per turn. This is the habit that makes trick 1 pay off. Ask for five changes at once and you cannot tell which one broke the shot. Change the background, watch the result, then change the light. Google says the model holds character, framing, and performance steady across turns, and its demo swaps backgrounds while two characters keep acting.

2. Draft Cheap, Then Upscale Only the Winner

A white sneaker on a rotating pedestal in a bright product photography studio with two softbox lights

Trick 3: draft at low resolution, upscale only the keeper. Google spells out this workflow. Draft fast at 360p or 720p, fix details with follow-up edits instead of re-generating the whole clip, then upscale your best creations to 1080p or 4K.

The money logic is simple. At the listed 720p rate of $0.07 per second, a 15 second draft costs about $1.05 and a 10 second draft about $0.70. Those are small numbers, but they multiply fast. Five full price re-rolls of a 15 second clip cost more than five dollars, before edits and upscales.

Build a habit. Generate short and low first. Watch the whole clip before you judge it. Edit in chat until the scene is right. Only then spend on the high resolution version. Most of your clips are tests and will never need the upscale. The one that works gets the polish.

A product ad, a try-on clip, or a social spot can be roughed out in cheap drafts all morning, then finished in the afternoon. The expensive step comes last, when you already know the clip works.

3. Feed It Photos and Watch Faces Stay Honest

A bright flat lay of printed photos of objects, folded clothes, a watch, a perfume bottle and a leather handbag on a white table

Trick 4: use up to six reference images. Omni 1.2 Flash accepts up to six reference images, with up to three images per character for up to two characters. That matters because the most common failure in AI video is drift. A face changes. A jacket changes color. A logo warps. Reference images pin those details down, and Google says this version holds faces, garments, packaging, and brand colors steadier across camera angles.

Give the model more than one angle. A front photo plus a side photo tells it more than a single portrait. A product photo plus a setting photo shows how the two belong together. Build a small reference set before you generate.

Trick 5: turn product photos into try-on and spin clips. This is the most practical use Google shows off. Combine one model photo with front and back shots of three outfits, and the model generates a 15 second virtual try-on sequence with turns and match cuts, keeping model and garments consistent. The same idea works for a catalog spin. Photograph the product once, then let the model rotate it and build the ad around it.

4. Pick Your Exact Length, Down to the Second

A silver stopwatch beside a coiled strip of orange film on a bright wooden desk

Trick 6: choose any length from 3 to 15 seconds. You can specify duration in one second steps, from 3 seconds up to 15. That sounds small, but it changes how you write a prompt. A 5 second clip is a single beat. A 10 second clip is a setup and a payoff. A 15 second clip can hold a full product explainer or a complete social ad spot in one generation, which is exactly how Google frames it.

Match the length to the job. A logo reveal needs 3 to 5 seconds. A before and after needs 8 to 10. A joke with a setup and a punchline needs the full 15. Picking the length first stops you from writing a 15 second story into a 5 second clip, or stretching a 3 second idea across 15 seconds of filler.

Shorter is cheaper and faster to test. If you are testing an idea, generate at 5 or 8 seconds first. You will see whether the scene works in a fraction of the time, and you can extend the winner later.

The cheap to sharp workflow

1. Draft360p or 720p2. Chat editsone fix per turn3. Upscale1080p or 4K15 seconds at 720p costs about $1.05at the listed $0.07 per second rate

Only the winning clip gets the expensive upscale.

5. Stretch a Good Clip to 30 Seconds

A long strip of film with colorful nature scenes unrolling across a bright white desk

Trick 7: extend a winner instead of generating a longer clip. Native generation stops at 15 seconds, but extension takes you further. Each extension adds 3 to 10 seconds, and the total can reach 30 seconds in the Gemini API. The model uses the final 10 seconds of your clip as context, then appends the new footage to the end.

That last detail sets the rule. Extension only adds to the end. It cannot prepend footage or insert a scene in the middle, so plan your clip as a chain. Generate a strong opening, extend it once to reach the middle, extend again for the ending. Each step builds on the last 10 seconds, which keeps the scene consistent from shot to shot.

Use extension for stories, not for fixes. If the opening is wrong, do not extend it. Fix it with a chat edit first, or regenerate it. Every new second inherits whatever came before it.

Try this pattern. Generate a 10 second establishing shot, extend it by 10 seconds for the action, then extend once more for the payoff. Three generations, one continuous 30 second sequence.

6. Direct the Camera and the Voices

A professional video camera on a gimbal next to a studio microphone on a boom arm in a warm recording studio

Trick 8: write camera moves and dialogue into the prompt. Google highlights two kinds of control. First, camera work. You can direct continuous moves like sweeping FPV flythroughs, dramatic zoom-outs, and orbits, with fewer unwanted jump cuts. Its demo flies one continuous shot through a piazza, through bronze doors, and up through a domed ceiling.

Second, sound. The model generates scenes with more fluid, natural sounding dialogue, clearer lip-sync, and distinct multi-speaker conditioning. Two characters can talk and keep their own voices. Google describes a demo where four voice references reach the right characters with expressive lip-sync across a 15 second scene.

Write prompts like a shot list. Name the subject, the action, the setting, and the time of day. Then add the camera. Then add the sound. A prompt that says “slow orbit around a coffee cup on a sunny cafe table, steam rising, soft jazz in the background” gives the model far more to work with than “a coffee cup.”

Keep dialogue short. A line or two per character lands better than a speech in a 15 second clip. Let the voices carry the joke or the pitch, and let the camera carry the rest.

7. Know the Guardrails Before You Spend a Generation

A large monitor showing abstract colorful gradient frames beside a keyboard and a small gold shield ornament on a desk

Trick 9: learn the limits first, so no generation is wasted. Every tool has edges. Google is open about the weak spots. Fast sports action, rapid script changes, and complex physics are still under development, so a sprint finish or a collapsing tower is a risky first test. Pick scenes with smooth, readable motion.

First and last frame interpolation tops out at 10 seconds. You set the opening and closing frames and the model fills the middle. Clips of 11 to 15 seconds accept one reference video of up to three seconds, while shorter clips accept up to three. Audio inside a reference video is ignored, and uploaded audio references are not supported.

Voice editing has its own guardrail. The model card says video editing can change what people say, and Google is restricting that ability while it studies a safe release. You can generate dialogue, but rewriting speech in an existing clip is limited for now.

Two more facts. Editing or extending uploaded videos is unavailable in the European Economic Area, Switzerland, and the United Kingdom, although model generated videos can still be edited there. From 3 December 2026, API requests to Omni 1.1 Flash route to 1.2 automatically.

The ceilings to remember

15 secondslongest native clip30 secondslongest extended sequence10 secondsinterpolation ceilingSynthIDwatermark on every clip

Extension adds to the end only. It never prepends or inserts mid-clip.

The fastest way to learn this tool is to make one small clip today. Pick a product you own or a scene you can describe in one sentence. Generate it short and cheap. Then talk to it. Change the light, swap the background, stretch it by ten seconds, and watch how much of the scene survives each turn. That survival rate is the real product here.

Omni 1.2 Flash is not a replacement for a camera or an editor. It is a way to test an idea in minutes, to build an ad variant without re-shooting, and to turn a still photo into a moving scene with sound. The nine tricks above all point the same way. Start cheap, direct in plain language, fix one thing at a time, and save the expensive steps for the clip that already works.

If you try it, keep one habit from this article. Watch every clip all the way through before you judge it, and keep the version that surprised you. That is usually the one worth upscaling.

We craft smart, actionable content about money, health, style, and travel that fits real life. No fluff, just practical wisdom delivered with charm and backed by data.

Stay in the loop

Get weekly stories that actually fit real life — no spam, just value.


© 2026 Live Our Narrative. All rights reserved.

Made with Love
Live Our Narrative
Logo