What happened
Google Cloud’s Gemini Enterprise Agent Platform documentation describes Gemini Omni 1.2 Flash as a multimodal model for video, image and text tasks. Its model ID is gemini-omni-1.2-flash. The page says it is optimized for video generation, editing, extension and standalone video upscaling, and can return video with synchronized audio alongside text responses in one model. The listed output resolutions are 360p, 720p, 1080p and 4K, with generated or edited clips up to 15 seconds.
The model accepts text, images and video, but not audio as an input modality. Its maximum input is 131,072 tokens and maximum output is 260,640 tokens. The documentation lists support for video from text, image and references, generation from first and last frames, sound generation, editing, extension and upscaling. Sound generation covers speech, music and sound effects. Content Credentials (C2PA) are also listed as supported. Structured output, system instructions, context caching, RAG Engine, tuning, image generation and agentic video understanding are not supported. Neither function calling nor grounding, code execution, URL context or computer use is supported.
The duration limits vary by operation. Video understanding accepts input up to 10 minutes, while editing and upscaling accept up to 15 seconds. Extension can take a video of any duration, but uses up to its last 10 seconds; it produces a 3-to-10-second addition, concatenated onto up to the last 30 seconds of the input. Generation and editing outputs run from 3 to 15 seconds, except first-and-last-frame generation, which is limited to 3 to 10 seconds. Prompts can contain up to three videos; supported aspect ratios are 16:9 and 9:16. The page lists a 15-second maximum for video inputs with or without audio.
The model is available globally through Standard PayGo, Provisioned Throughput and fixed quota. Batch inference is not supported. The Interactions API and Count Tokens API are supported; GenerateContent, Gemini Live and Chat Completions APIs are not. Google lists the launch stage as GA and the release date as November 5, 2026. Its listed retirement date is November 5, 2027 or later. The page links to pricing but does not state a price in the model description. Deploying the example app requires a Google Cloud project with billing and the Agent Platform API enabled.
What the archive adds
New Runtime’s earlier coverage of Google Cloud model listings distinguished models by their workflow affordances and omissions, including Gemini Nano Banana 2.1 on October 6, 2026. This is a new model release in that ongoing coverage, with a video-specific capability and constraint profile rather than a restatement of the earlier image-model note.
Trend signal
This supports the pattern Generative media becomes application infrastructure: Google presents video generation, editing, extension and upscaling as composable operations with distinct limits, not just a one-off artifact prompt. The lack of function calling and batch inference also makes the boundaries of integration part of the model’s product profile.
Why it matters
For a team building a media product, the operational question is whether the model fits the shape of the workflow, not simply whether it can generate a polished clip. The separate limits for understanding, editing and extension affect how inputs must be chunked and which jobs can be automated in one step. A 10-minute understanding limit does not mean a 10-minute editing input; extension can accept a longer source but only uses a short trailing segment. Those differences should shape product promises and orchestration.
The model also is not a general agent endpoint: it lacks function calling, grounding and computer use. Teams should plan to wrap it in their own workflow logic rather than expect it to coordinate tools. Global availability and multiple consumption modes create deployment options, but the absence of batch inference may matter for offline render queues. The practical value is therefore strongest for short, interactive video workflows where synchronized audio and direct editing are central.
Limits and uncertainty
Google’s documentation states the model’s listed capabilities, limits, APIs and availability, but does not report independent quality evaluations, reliability rates or comparative performance against other video models. The page links to a pricing page without specifying costs in the model entry, so the economics of generating, editing or extending clips remain to be checked there. It also does not explain how quality changes across resolutions or how often generated audio stays aligned in production use. The retirement date is phrased as November 5, 2027 or later, not a guaranteed cutoff.
Reader takeaway
Gemini Omni 1.2 Flash is a generally available Google Cloud option for short video generation and editing with synchronized audio, including 4K output. Treat its operation-specific duration limits, absent batch inference and lack of tool calling as design constraints. Validate the exact workflow and price before committing to it as a production dependency.