Google has released Gemini Omni 1.2 Flash, a video generation and editing model capable of creating native clips up to 15 seconds long. The update lowers the cost of 720p video to $0.07 per second, a 30% reduction from the previous version.
The model is available on Google Cloud and the Gemini API. Developers gain access to longer clips, fewer re-renders, and a lower bill per second.
Key Specifications
- Availability: Google-hosted only via Gemini API or Google Cloud. No open weights or local deployment.
- Performance: Google claims better adherence to reference images and less visual drift. No specific benchmark scores for version 1.2 are published.
- Best Use Case: Native 15-second clips at 720p for $0.07 per second, totaling approximately $1.05 per clip.
- Limitations: Fast sports action, rapid script changes, and complex physics still require improvement.
What is Gemini Omni 1.2 Flash?
Gemini Omni 1.2 Flash is a multimodal video model from Google DeepMind. It accepts text, images, short video clips, and audio as input. The output is video with audio. Google’s developer documentation describes the Omni line as natively multimodal, with editing handled through the Interactions API.
Changes from Omni 1.1 Flash
The launch announcement lists six upgrades:
- Longer clips: Generation and editing now reach 15 seconds, up from 10 seconds in Omni 1.1.
- Duration control: Users set length in 1-second steps ranging from 3 to 15 seconds.
- Reference adherence: Users can combine character, product, or style images with motion clips up to 3 seconds.
- Consistency: Faces, garments, packaging text, and brand colours hold across camera angles with less morphing.
- Speech: Dialogue sounds more natural, featuring clearer lip-sync and distinct multi-speaker conditioning.
- Camera control: FPV flythroughs, zoom-outs, and orbits show fewer unwanted jump cuts.
How the Draft, Edit, Upscale Workflow Works
Google frames cost control as a workflow rather than just a price point. Users draft at 360p or 720p. They fix a detail with a follow-up edit instead of regenerating the whole clip. Then they upscale the best result to 1080p or 4K.
Multi-turn conversational editing now works across full 15-second clips. In practice, users can swap a product, background, lighting, or aspect ratio turn by turn. The documentation shows this through a previous_interaction_id, so the prior video is not re-uploaded.
Cost Breakdown
Google prices Omni 1.2 Flash at $0.07 per second of 720p video. Savings are similar at other resolutions. For context, the pricing page bills Omni 1.1 Flash at roughly $0.10 per second of 720p video. A full 15-second 720p clip on 1.2 works out to about $1.05.
Google Vids users can generate 15-second clips at no additional cost. Provisioned Throughput is available, and outputs carry a SynthID watermark.
Comparison with Other Models
| Feature | Gemini Omni 1.2 Flash | Gemini Omni 1.1 Flash | Veo 3.1 | Seedance 2.0 |
|---|---|---|---|---|
| Developer | Google DeepMind | Google DeepMind | ByteDance | |
| Parameters | Not disclosed | Not disclosed | Not disclosed | Not disclosed |
| Max single clip | 15 s | 10 s | 8 s | 15 s |
| Duration options | 3 to 15 s, 1 s steps | Not disclosed | 4, 6 or 8 s | 4 to 15 s |
| Output resolution | 360p, 720p; 1080p and 4K upscaled | 360p, 720p; 1080p and 4K upscaled | 720p, 1080p, 4K | 480p, 720p native |
| Native audio | Yes | Yes | Yes | Yes |
| Multi-turn editing | Yes, up to 15 s | Yes | Not listed | Not disclosed |
| 720p price | $0.07/s | About $0.10/s | Not listed on source page | Token-based, varies |
| Weights | Closed, API only | Closed, API only | Closed, API only | Closed, API only |
Benchmark Performance
Google has not published benchmark scores for version 1.2. Its claims rest on side-by-side demos against Omni 1.1. The DeepMind model page reports human-rater results for the Omni Flash line, including 1,003 MovieGenBench prompts. In June, an earlier Omni Flash version took #1 on the Arena text-to-video board. Those rankings predate 1.2, so treat them as family context only.
Early partner feedback is positive but self-selected. According to Google’s release page, invideo’s VP of Engineering said the model sticks to the prompt and that physics and sound feel real. QuickFrame AI says 15 seconds at 1080p lets users build a full CTV ad in 1 generation.
Developer Access
- Gemini API: Prototype in Google AI Studio and build with the developer docs.
- Google Cloud: Use the Agent Platform API with enterprise SLAs.
- Google apps: Flow, Google Vids, Google Ads and the Gemini app (rolling out to AI Ultra).
What it means
For people making video, the practical change is the ability to create a complete 15-second clip in a single generation without needing to stitch shorter segments together. The lower price point makes testing ideas cheaper, while the multi-turn editing feature allows fixing specific errors like a misplaced logo or incorrect lighting without starting over. However, creators dealing with high-speed action or complex physical interactions should still expect some trial and error.
