Google DeepMind has rolled out Gemini Omni 1.2 Flash, an update to its video generation model that focuses on longer clips, more consistent output, and lower cost. The model is generally available across the Gemini API and Gemini for business, and it is also accessible through Google Vids and Google Cloud for production use.
What changed
- Longer clips: Native generation and editing now extend to 15 seconds, up from 10 seconds in Omni 1.1. Duration can be set in one-second increments from 3 to 15 seconds.
- Reference adherence: The model accepts character, product, or style images, short motion clips of up to three seconds, and audio cues, and it is designed to keep those details stable from the first frame to the last.
- Consistency: Google says characters, objects, and scenes show less visual drift or morphing across camera transitions, including when interpolating between keyframes or extending a clip.
- Conversational editing: Users can refine a clip over multiple turns, such as swapping a product or background, changing lighting or weather, or switching aspect ratio, without regenerating the full clip.
Pricing and workflow
Google reports a 30% price reduction compared with Gemini Omni 1.1, listing $0.07 per second for 720p video. The company recommends a draft-and-refine workflow: generate at 360p or 720p, correct details with follow-up edits, then upscale the final version to 1080p or 4K. This approach is intended to keep iteration costs down.
Practical considerations
For teams building product imagery, virtual try-on, or branded video ads, the combination of longer duration and identity or garment consistency reduces the need for manual reshoots. Security and governance teams evaluating the model for production should review how reference images and brand assets are handled under their own data policies, since the sources describe the feature set but do not detail data retention or access controls.
