Gemini Omni 1.2 Flash Pushes Video Generation Toward Longer, Editable Clips

Introduction

The video-generation race is shifting from simple visual novelty to dependable production workflows. Google DeepMind’s Gemini Omni 1.2 Flash targets that gap with a set of practical improvements: longer native clips, stronger reference adherence, conversational editing, better speech synchronization, and lower generation costs.

Key updates

• Native clips up to 15 seconds: The model increases the supported duration from 10 seconds in the previous release to 15 seconds. Creators can choose any duration from 3 to 15 seconds in one-second increments, which is useful for product explainers, short narratives, and social advertisements.
• Stronger reference adherence: Gemini Omni 1.2 Flash can combine character, product, or style images with motion references of up to three seconds and audio cues. Google’s examples include virtual try-ons, catalog spins, and branded advertisements. Faces, clothing shapes, packaging typography, and brand colors are intended to remain more stable through camera changes.
• Conversational editing: Instead of starting over, users can revise the same clip across multiple turns. They can swap a product or background, adjust lighting or weather, or change the aspect ratio. This workflow is especially relevant when one campaign needs several audience or placement variants.
• More natural speech: The update focuses on clearer lip-sync, more fluid dialogue, and conditioning for multiple speakers. These capabilities can support explainers, avatar videos, and animated scenes with several characters.
• More controlled camera motion: The model is designed to handle continuous movements such as FPV flythroughs, orbits, and dramatic zoom-outs while reducing unwanted jump cuts and abrupt scale changes.
• Lower-cost iteration: The listed price for 720p video is $0.07 per second, 30% below Gemini Omni 1.1. Users can draft quickly at 360p or 720p, make targeted edits, and upscale selected results to 1080p or 4K.

Why it matters

The most important shift is the move from one-shot prompting toward an iterative loop: generate, inspect, revise through dialogue, and export variants. For commercial teams, consistent products and characters may be more valuable than a single impressive demo. Longer clips can also reduce the amount of stitching required to tell a complete visual story.

The release still comes with boundaries. Google DeepMind says that fast sports action, rapid script changes, and complex physics remain areas of ongoing development. A 15-second limit therefore should not be interpreted as a guarantee of production-ready continuity in every difficult scene. The model is being made available through the Gemini API, Google Vids, Gemini for business, and related Google Cloud workflows. Enterprise users can access Google Cloud infrastructure and Provisioned Throughput, while outputs carry an imperceptible SynthID watermark.

Overall, Gemini Omni 1.2 Flash competes on controllability and cost as much as on visual quality. If the improved consistency holds up in real projects, the model could make short advertisements, product demonstrations, and explanatory videos easier to produce at scale.

Source: Google DeepMind