# Gemini Omni 1.2 Flash makes 15-second AI video, lets you edit it by chat, and costs 30% less

> Google's video model is generally available at $0.07 a second for 720p, with conversational edits that change lighting or backgrounds without regenerating the whole clip.

- URL: https://devisdead.ai/news/gemini-omni-1-2-flash-makes-15-second-ai-video-lets-you-edit-it-by-chat-and-cost
- Published: 2026-11-05T19:13:13Z
- Author: DevIsDead Newsroom
- Section: AI
- Tags: google, gemini, video generation, deepmind
- Originally reported by The Decoder (https://the-decoder.com/googles-gemini-omni-1-2-flash-stretches-ai-video-clips-to-15-seconds-and-cuts-pricing-by-30-percent/).

Google DeepMind has made Gemini Omni 1.2 Flash, its video generation model, generally available, with longer clips, chat-based editing and a lower price, Matthias Bastian reports for [The Decoder](https://the-decoder.com/googles-gemini-omni-1-2-flash-stretches-ai-video-clips-to-15-seconds-and-cuts-pricing-by-30-percent/).

## What actually shipped

- **Length:** clips now run from 3 to 15 seconds, up from a 10-second ceiling.
- **Price:** 30% cheaper, at $0.07 per second for 720p.
- **Where:** the Gemini API, Google AI Studio, Google Cloud Console, Flow, Google Vids, Google Ads, and the Gemini app for AI Ultra subscribers.
- **Provenance:** every video carries an invisible SynthID watermark.

## Editing by conversation

The feature Google is pushing hardest is Conversational Editing. Instead of rerolling a whole clip because the lighting is off, you ask for the change and the model adjusts just that, whether it's the background, the light or the aspect ratio, across the full 15 seconds.

Google pairs it with a cheaper workflow: draft at 360p or 720p, fix details with follow-up edits, and only upscale the final version to 1080p or 4K. If you've paid for ten full-quality generations to get one usable shot, you can see the appeal.

## Consistency claims

Google also says the model sticks much more closely to references. It accepts character images, product photos, short motion references up to three seconds long, and audio cues, and is supposed to keep faces, clothing, packaging text and brand colors stable across camera angles. Its demo is a virtual try-on: one photo of a model plus product shots of three outfits becomes a 15-second clip with full 360-degree turns that keeps the person and each garment's cut intact.

Other claimed improvements include less morphing during motion and camera moves, more natural dialogue with better lip-sync and multiple speakers, and complex continuous shots like FPV drone flights with fewer accidental jump cuts.

## Benchmarks vs. vibes

All of that is Google's description of Google's model. The price and length are facts you can check today; "significantly better" consistency is something to test with your own references before you rebuild a pipeline around it. For anyone building product video or ad tooling, though, the combination of cheap drafts and targeted edits is the part that changes the math.

**[Scoreboard](/scoreboard) update:** AI +1. AI video gets longer, cheaper and editable by asking.

Stay undead.