11-19 更新 实时
首页/ AI 大模型/ 正文

Google Gemini Omni 1.2 Flash 视频延长至 15 秒,降价 30%

AI 大模型 来源:The Decoder 2026-11-06 03:05 1 阅读 机器翻译

Google DeepMind 正在将其 Gemini Omni 1.2 Flash 视频生成模型全面开放可用。这款更新后的模型可生成最长 15 秒的片段,支持对话式编辑,成本比上一代降低 30%。

Google 表示,Gemini Omni 1.2 Flash 正通过 Gemini API、Google AI Studio、Google Cloud Console、Flow、Google Vids、面向 Google AI Ultra 订阅用户的 Gemini 应用以及 Google Ads 陆续推出。每一段生成的视频都带有不可感知的 SynthID 水印,以标明其来源。

片段时长现在可在 3 到 15 秒之间,此前最长为 10 秒;价格下降 30%,720p 视频为每秒 0.07 美元。生成的片段还可以通过持续的对话进行编辑。

一个突出功能是 Google 所称的“对话式编辑”(Conversational Editing)。用户可以逐步调整生成的片段,更改背景、光照或宽高比,而无需重新生成整个片段。Google 表示,这在整个 15 秒时长内都有效。配合分层工作流,该功能有望降低成本。先以 360p 或 720p 生成草稿,再通过后续编辑修正个别细节,最后仅将最终结果放大到 1080p 或 4K。

Google 承诺借助参考图像可获得更一致的结果

Google 还承诺参考图遵循度显著提升。根据公告,该模型可以接收人物图像、产品照片、最长 3 秒的短动作片段以及音频提示作为输入。人脸、服装、包装字体和品牌颜色在不同机位角度下应保持一致。

举例来说,Google 展示了一段虚拟试穿短片。仅凭一张模特照片和三套服装的产品图,该模型就能生成一段 15 秒的画面序列,包含 360 度转身,同时保持模特的身份特征以及每件服装的剪裁与垂坠感。

据称整体视觉一致性也有所提升。Google 表示,与上一代相比,在运动和镜头切换过程中的视觉漂移或变形更少。

除此之外,Gemini Omni 1.2 Flash 还应当能生成更流畅、听起来更自然的对白,唇形同步更清晰,并具备区分多个说话人的能力。该模型还能处理复杂、连续的镜头运动,例如 FPV 无人机飞行或戏剧性的拉远镜头,且不想要的跳切显著减少。

没有炒作的 AI 新闻——由人工甄选

译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。

查看英文原文(Google's Gemini Omni 1.2 Flash stretches AI video clips to 15 seconds and cuts pricing by 30 percent)

Matthias Bastian

Google Deepmind is making its Gemini Omni 1.2 Flash video-generation model generally available. The updated model creates clips up to 15 seconds long, supports conversational editing, and costs 30 percent less than its predecessor.

Google says Gemini Omni 1.2 Flash is rolling out through the Gemini API, Google AI Studio, the Google Cloud Console, Flow, Google Vids, the Gemini app for Google AI Ultra subscribers, and Google Ads. Every generated video carries an imperceptible SynthID watermark to indicate its origin.

Clips can now run between three and 15 seconds, up from a maximum of 10 seconds previously, and pricing drops 30 percent to $0.07 per second for 720p video. Generated clips can also be edited through an ongoing conversation.

A standout feature is what Google calls "Conversational Editing." Users can tweak a generated clip step by step, changing the background, lighting, or aspect ratio without regenerating the entire clip. Google says this works across the full 15-second length. Paired with a tiered workflow, the feature is supposed to cut costs. Drafts are first created at 360p or 720p, individual details are corrected with a follow-up edit, and only the final result is upscaled to 1080p or 4K.

Google promises more consistent results with reference images

Google also promises significantly better reference adherence. According to the announcement, the model can take character images, product photos, short motion clips of up to three seconds, and audio cues as inputs. Faces, clothing, packaging typography, and brand colors are supposed to stay consistent across different camera angles.

As an example, Google shows a virtual try-on clip. From a single model photo and product shots of three outfits, the model generates a 15-second sequence with 360-degree turns while preserving the model's identity and each garment's cut and drape.

Overall visual consistency is said to have improved as well. Google says there is less visual drift or morphing across motion and camera transitions than with the predecessor.

Beyond that, Gemini Omni 1.2 Flash is supposed to produce more fluid, natural-sounding dialogue with clearer lip-sync and the ability to distinguish between multiple speakers. The model further handles complex, continuous camera movements such as FPV drone flights or dramatic zoom-outs, with significantly fewer unwanted jump cuts.

AI News Without the Hype – Curated by Humans

原文为英文,正文由自动抓取并保留原文。

本文由采集器自动同步

原文发布于 The Decoder(Google's Gemini Omni 1.2 Flash stretches AI video clips to 15 seconds and cuts pricing by 30 percent),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。

阅读原文 ↗