Gemini Omni 1.2 Flash Guide: 15-Second Video Ads for About $1 Each — Edit by Chatting

A 15-second video ad now costs about $1.05 to generate — $0.07 a second at 720p, 30% cheaper than the last version — and if something in the clip is wrong, you fix it by typing a sentence instead of starting over. That is Gemini Omni 1.2 Flash, which Google made generally available on November 5, 2026. Daniel is a realtor with three listing photos and no video; a videographer quoted him $400 and next Tuesday. This Gemini Omni 1.2 Flash guide shows you how Daniel gets his clip for about a dollar in 10 minutes, and how to sell the same clip to every other agent in his office.

What changed on November 5

Because 15 seconds is exactly the length of a social ad spot, a listing teaser, or a product demo — and until now, making one meant a camera, a shoot, and an editor. Omni 1.2 Flash went generally available on November 5 across the Gemini API, AI Studio, Vertex AI, Flow, Vids, and the Gemini app (Ultra). What changed versus the previous version (all from Google’s announcement; performance claims are the vendor’s own words):

• Native 15-second clips, up from 10. You can pick any length from 3 to 15 seconds in one-second steps. What that means for you: a complete product story or ad spot in a single generation, no stitching.
• 30% cheaper: $0.07 per second at 720p. What that means for you: a full 15-second ad costs about $1.05 — the AI cost is now a rounding error next to what you charge.
• It holds onto your references. Feed it a model photo plus front-and-back shots of three outfits and it generates a 15-second virtual try-on sequence, keeping the model’s identity and each garment’s cut and drape consistent. What that means for you: your client’s actual product stays recognizable — no more “close enough” morphing.
• Edit by chatting. Swap a product or background, change the lighting or weather, switch aspect ratio — turn by turn, on the same clip. What that means for you: one good clip becomes ten ad variants for different audiences, without starting from scratch each time.
• More natural speech. Clearer lip-sync and distinct voices when several people talk. What that means for you: dialogue scenes and explainer clips finally sound like people, not robots.

Price and free quota

Plain numbers, from Google’s announcement (vendor pricing, not independently verified):

• 720p video: $0.07 per second. A 5-second clip ≈ $0.35; a 10-second clip ≈ $0.70; a 15-second clip ≈ $1.05. That is 30% less than Omni 1.1.
• Where you can use it: Gemini API and AI Studio for developers; Flow and Google Vids (Google says 15-second clips in Vids at no additional cost with a Google account or Workspace subscription); the Gemini app for Google AI Ultra subscribers; Vertex AI and Google Ads for businesses.
• The money-saving workflow Google itself suggests: draft fast at 360p or 720p, fix details with a follow-up edit instead of re-generating, then upscale only your best creations to 1080p or 4K. What that means for you: never pay for a polished render of a clip you haven’t approved yet.
• What we could not confirm: per-second prices for 1080p and 4K, and AI Studio’s free-tier limits for video. The announcement doesn’t state them — check AI Studio pricing before you quote a client, and treat anything beyond 720p as “unconfirmed” until you see it on Google’s pricing page.

Hands-on tutorial: your first 15-second video in 10 minutes

What you need: a computer, an internet connection, and a Google account.

Step 1 — Open Google AI Studio. Go to aistudio.google.com and sign in with your Google account. What you’ll see: Google’s workbench for trying its AI models. This is the front door for prototyping. Your first 15-second clip at 720p costs about $1.05 at the published rate (see pricing above); Google has not published a free video allowance for AI Studio, so don’t count on a free tier.

Step 2 — Pick video generation. Look for the video option and select Gemini Omni 1.2 Flash as the model. What you’ll see: a prompt box, plus places to attach reference images. Don’t worry about the settings panel on your first run — the defaults are fine.

Step 3 — Upload reference photos. Attach 1–3 photos: your product, and if there’s a person in the scene, a clear photo of them. These are your “reference images” — the AI holds onto the exact faces, products, and label text while it builds the scene around them. What you’ll see: thumbnails of your references. Good references are the whole trick — clear, well-lit, front-facing photos give you consistent results; blurry ones give you drift.

Step 4 — Paste this prompt (copy it exactly). Here’s a complete, ready-to-use prompt for a 15-second product ad — a realtor’s listing teaser, adaptable to any product:

“A bright 15-second vertical video tour of a modern apartment living room. Start wide on the sunlit sofa and coffee table, then slowly push in toward the large window with a city view. A small vase of fresh flowers sits on the table; warm afternoon light, clean and airy, cinematic but realistic. No people, no text on screen, no narration — just gentle background music and the feeling of walking into a home you want to live in.”

Set duration to 15 seconds and resolution to 720p for your first run (cheapest). Press Generate. What you’ll get: a 15-second vertical clip that opens wide on the sunlit sofa, drifts slowly toward the window and the city view, and holds the vase of flowers and the warm afternoon light steady through the whole shot — no people, no text on screen, just the feeling of walking into a home you want to live in.

Step 5 — Edit by chatting, don’t regenerate. Watch the clip. This is multi-turn editing: fixing a video by talking, instead of re-generating the whole clip. If the light is wrong, type: “Change the lighting to warm evening light, keep everything else exactly the same.” If the camera move is wrong, type: “Make the camera move slower and smoother.” What you’ll see: the same clip, fixed — not a brand-new random take. This is the feature that saves you re-rolls: each fix targets one thing instead of gambling on a brand-new take.

Step 6 — Download. When you’re happy, download the clip. Congratulations — you just made a 15-second ad in under 10 minutes, for about a dollar, without touching a camera.

One real way to make money with it

The play: 15-second listing and product videos for local businesses, $149 each.

• Who’s the customer: realtors (listing teasers), restaurants (dish spotlights), boutiques (try-on clips), gyms and salons (atmosphere spots). Anyone who posts video to Instagram Reels or TikTok but can’t afford a shoot.
• What you sell: one 15-second vertical video from their photos, with two edit rounds included, delivered in 24 hours.
• Your cost: ~$1.05 per clip at 720p, plus a few cents for edit passes. Call it $2 all-in.
• Your margin: $149 − $2 = $147 per video. Five videos a week is ~$735 for work that takes about an hour each, most of it waiting for renders.
• Why they buy: their alternative is a $400 videographer and a week of scheduling. You show them their own listing moving — the sample sells itself.

7-day validation plan:

1. Day 1: Make 3 sample videos for 3 real local businesses from their public photos (a listing, a dish, a storefront). Cost: ~$3.
2. Days 2–3: DM or walk in: “I made this 15-second video for your listing as a sample. Want one for your next listing for $149, delivered tomorrow?” Show the video, not a pitch deck.
3. Days 4–5: First paying customer → deliver the clip plus two chat-edits (“make it brighter”, “slower camera”), exactly as the tutorial does.
4. Day 6: Ask for a testimonial and one introduction to another agent or owner.
5. Day 7: Count: one paid customer at $149 proves the model. Then package it: 4 videos a month for $499 as a retainer, priced on the $0.07/sec math so a price move never eats your margin.

Pitfalls and limits

1. Still in development on hard shots. Google says fast sports action, rapid script changes, and complex physics are still being improved. What that means for you: sell slow, static product and listing shots first; don’t promise a skateboard trick sequence.
2. Check every face and label. Consistency is better than the last version, not perfect. Watch the full clip for morphing faces, garbled packaging text, or drifting colors before you send it to a client. Regenerating costs about a dollar; sending a broken clip costs your reputation.
3. The Gemini app needs Ultra. If a client asks “can I do this in the Gemini app?”, the honest answer is: video generation there is rolling out to Google AI Ultra subscribers. Point beginners at AI Studio or Vids instead.
4. Prices beyond 720p are unconfirmed. Don’t quote 1080p or 4K rates to a client until you’ve seen them on Google’s pricing page. Sell 720p first — on a phone screen, nobody can tell the difference.
5. The watermark is real. SynthID is invisible to viewers but machine-readable. Tell clients upfront that the video is AI-made — it’s honest, and it’s increasingly what platforms expect.