Google发布Gemini Omni 1.2 Flash:支持15秒视频生成
Google Releases Gemini Omni 1.2 Flash: 15-Second AI Video at $0.07 per Second
视频生成领域的重要迭代,时长突破至15秒且价格下调30%,对Agent工作流和多轮编辑有直接影响,建议开发者关注其API接口与成本控制方案。
Google has released Gemini Omni 1.2 Flash, the newest version of its video generation and editing model. Gemini Omni 1.2 Flash generates and edits clips up to 15 seconds natively. It also cuts the price by 30% compared with Omni 1.1 Flash. The model is generally available on Google Cloud and the Gemini API. For developers, the takeaways are longer clips, fewer re-renders and a lower bill per second.
Google 发布了 Gemini Omni 1.2 Flash,这是其视频生成和编辑模型的最新版本。Gemini Omni 1.2 Flash 原生支持生成和编辑长达 15 秒的片段。与 Omni 1.1 Flash 相比,其价格降低了 30%。该模型已在 Google Cloud 和 Gemini API 上全面开放。对开发者而言,主要改进包括更长的片段、更少的重新渲染以及每秒更低的费用。
TL;DR
TL;DR(太长不看版)
- Runs on: Google-hosted only (Gemini API, Google Cloud). No open weights, no local deployment.
- Performance: Google reports better reference adherence and less drift. No 1.2 benchmark scores are published.
- Best: Native 15-second clips at $0.07 per second of 720p video, about $1.05 per 15-second clip.
- Worst: Google says fast sports action, rapid script changes and complex physics still need work.
- 运行环境:仅限 Google 托管(Gemini API、Google Cloud)。无开源权重,不支持本地部署。
- 性能:Google 报告称参考遵循性更好,漂移更少。未发布 1.2 版本的基准测试分数。
- 最佳情况:原生 15 秒片段,720p 视频每秒 0.07 美元,每个 15 秒片段约 1.05 美元。
- 最差情况:Google 表示,快速体育动作、快速脚本更改和复杂物理效果仍需改进。
What is Gemini Omni 1.2 Flash?
什么是 Gemini Omni 1.2 Flash?
Gemini Omni 1.2 Flash is a multimodal video model from Google DeepMind. It takes text, images, short video clips and audio as input. It outputs video with audio. Google’s developer docs describe the Omni line as natively multimodal, with editing handled through the Interactions API.
Gemini Omni 1.2 Flash 是 Google DeepMind 推出的多模态视频模型。它接受文本、图像、短视频片段和音频作为输入,输出带有音频的视频。Google 的开发文档将 Omni 系列描述为原生多模态,并通过 Interactions API 处理编辑功能。
What changed from Omni 1.1 Flash?
与 Omni 1.1 Flash 相比有哪些变化?
The launch lists 6 upgrades:
此次发布列出了 6 项升级:
- Longer clips: Generation and editing now reach 15 seconds, up from 10 seconds in Omni 1.1.
- Duration control: You set length in 1-second steps from 3 to 15 seconds.
- Reference adherence: You can combine character, product or style images with motion clips up to 3 seconds.
- Consistency: Faces, garments, packaging text and brand colors hold across camera angles with less morphing.
- Speech: Dialogue sounds more natural, with clearer lip-sync and distinct multi-speaker conditioning.
- Camera control: FPV flythroughs, zoom-outs and orbits show fewer unwanted jump cuts.
- 更长的片段:生成和编辑现在可达 15 秒,而 Omni 1.1 中为 10 秒。
- 时长控制:你可以以 1 秒为步长设置从 3 到 15 秒的时长。
- 参考遵循性:你可以将角色、产品或风格图像与最长 3 秒的运动片段结合使用。
- 一致性:在不同镜头角度下,面部、服装、包装文字和品牌颜色的变形减少,保持一致性。
- 语音:对话听起来更自然,唇形同步更清晰,并能区分多个说话者。
- 相机控制:第一人称视角飞行穿越、拉远镜头和环绕镜头中,不需要的跳切现象减少。
How does the draft, edit, upscale workflow work?
草稿、编辑、超分辨率工作流程是如何运作的?
Google frames cost control as a workflow, not just a price. You draft at 360p or 720p. You fix a detail with a follow-up edit instead of regenerating the whole clip. Then you upscale the best result to 1080p or 4K.
Google 将成本控制视为一种工作流程,而不仅仅是价格问题。你可以在 360p 或 720p 下进行草稿制作。通过后续编辑修复细节,而不是重新生成整个片段。然后将最佳结果超分辨率至 1080p 或 4K。
Multi-turn conversational editing now works across full 15-second clips. In practice, you can swap a product, background, lighting or aspect ratio turn by turn. The docs show this through a previous_interaction_id, so the prior video is not re-uploaded.
多轮对话式编辑现在可应用于完整的 15 秒片段。在实际操作中,你可以逐轮替换产品、背景、光照或宽高比。文档通过 previous_interaction_id 展示了这一点,因此无需重新上传之前的视频。
How much does Gemini Omni 1.2 Flash cost?
Gemini Omni 1.2 Flash 的费用是多少?
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 为 Omni 1.2 Flash 定价,720p 视频每秒 0.07 美元。在其他分辨率下节省的费用类似。作为参考,定价页面显示 Omni 1.1 Flash 的 720p 视频价格约为每秒 0.10 美元。在 1.2 版本上,一个完整的 15 秒 720p 片段成本约为 1.05 美元。
Google Vids users can generate 15-second clips at no additional cost. Provisioned Throughput is available, and outputs carry a SynthID watermark.
Google Vids 用户可以免费生成 15 秒片段。提供 Provisioned Throughput(预置吞吐量),输出内容带有 SynthID 水印。
How does it compare with other video models?
它与其他视频模型相比如何?
| 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 |
| 功能 | Gemini Omni 1.2 Flash | Gemini Omni 1.1 Flash | Veo 3.1 | Seedance 2.0 |
|---|---|---|---|---|
| 开发者 | Google DeepMind | Google DeepMind | ByteDance | |
| 参数量 | 未披露 | 未披露 | 未披露 | 未披露 |
| 单次最长片段 | 15 秒 | 10 秒 | 8 秒 | 15 秒 |
| 时长选项 | 3 至 15 秒,步长 1 秒 | 未披露 | 4、6 或 8 秒 | 4 至 15 秒 |
| 输出分辨率 | 360p、720p;1080p 和 4K 超分 | 360p、720p;1080p 和 4K 超分 | 720p、1080p、4K | 480p、720p 原生 |
| 原生音频 | 是 | 是 | 是 | 是 |
| 多轮编辑 | 是,最长 15 秒 | 是 | 未列出 | 未披露 |
| 720p 价格 | $0.07/秒 | 约 $0.10/秒 | 源页面上未列出 | 基于 Token,价格浮动 |
| 权重 | 闭源,仅限 API | 闭源,仅限 API | 闭源,仅限 API | 闭源,仅限 API |
How does it perform on benchmarks?
它在基准测试中的表现如何?
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.
Google 尚未发布 1.2 版本的基准测试分数。其主张基于与 Omni 1.1 的并列演示。DeepMind 模型页面报告了 Omni Flash 系列的人类评估者结果,包括 1,003 个 MovieGenBench 提示词。今年六月,早期的 Omni Flash 版本在 Arena 文本到视频排行榜上排名第一。这些排名早于 1.2 版本,因此仅将其视为家族背景信息。
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.
早期合作伙伴反馈积极,但属于自我选择样本。根据 Google 的发布页面,invideo 的工程副总裁表示该模型能严格遵循提示词,且物理效果和声音感觉真实。QuickFrame AI 称,1080p 下的 15 秒时长让用户能够在一次生成中构建完整的 CTV 广告。
Where can developers use it?
开发者可以在哪里使用它?
- 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).
- Gemini API:在 Google AI Studio 中进行原型开发,并使用开发者文档进行构建。
- Google Cloud:使用具有企业 SLA 的 Agent Platform API。
- Google 应用:Flow、Google Vids、Google Ads 以及 Gemini 应用(逐步向 AI Ultra 推出)。
Key Takeaways
关键要点
- Gemini Omni 1.2 Flash generates and edits native 15-second clips, up from 10.
- Price drops 30% to $0.07 per second of 720p video.
- Multi-turn editing now spans full 15-second clips without re-generation.
- No benchmark scores for 1.2 are published yet.
- Audio references remain allowlist-gated on Google Cloud.
- Gemini Omni 1.2 Flash 可生成和编辑原生的 15 秒片段,较之前的 10 秒有所增加。
- 价格下降 30%,降至 720p 视频每秒 0.07 美元。
- 多轮编辑现在可以覆盖完整的 15 秒片段,无需重新生成。
- 目前尚未发布 1.2 版本的基准测试分数。
- 音频引用在 Google Cloud 上仍受白名单限制。
Check out the launch post, the Google Cloud announcement and the developer docs.. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter. Wait! are you on telegram? now you can join us on telegram as well.
查看发布文章、Google Cloud 公告以及开发者文档。此外,欢迎在 Twitter 上关注我们,别忘了加入我们拥有超过 15 万成员的 ML SubReddit,并订阅我们的新闻通讯。等等!你在 Telegram 上吗?现在你也可以加入我们的 Telegram 群组了。
更进一步:量化金融体系
看懂新闻只是起点——沿量化金融路径,把它变成能交付的工程能力