Google released Gemini Omni 1.1 Flash to developers on 27 August at 16:00 UTC, adding scene extension, first-and-last-frame interpolation, a low-cost draft mode and video-reference conditioning. The documentation describes gemini-omni-1.1-flash as "a high-performance multimodal model designed for high-speed video generation, editing, and cinematic control." It is live in Google AI Studio, the Gemini API, the Gemini Enterprise Agent Platform and Google Flow, with scene extension reaching AI Plus, Pro and Ultra subscribers in the Gemini app globally.
What is actually new
Scene extension appends up to 10 seconds per turn, to a stated maximum total length of 40 seconds. Video reference accepts up to three seconds of existing footage to hold a character or environment consistent. Aspect ratios are 16:9 by default and 9:16. SynthID watermarking is built in. Adobe Firefly, Figma Weave and Runway are named as production users; Figma Weave's creative director says the model "takes teams beyond generating videos to truly directing them."
What the common framing gets wrong
Three claims are circulating that Google's own pages contradict. First, "4K video generation" is a post-process: the model card marks both 4K and 1080p as upscaled, and native generation stops at 720p — 360p in draft mode. Second, "40-second videos" conflates a ceiling with a generation. Forty seconds is reachable only by stacking 10-second turns onto the end of a clip; the model cannot produce forty seconds in one shot and cannot insert in the middle. Third, "60% faster and one-third the cost" belongs to the 360p draft path alone — that is a resolution discount being read as an efficiency gain.
Production-ready in one place, Preview in another
The blog post frames the release as production-ready. The Google Cloud documentation page for the enterprise variant is titled "Gemini Omni 1.1 Flash Preview". Availability surface and launch stage are not the same thing across Google's three delivery channels, and coverage has flattened them into one.
The geofence nobody carried
Extending an uploaded video with added dialogue is unavailable in the EEA, Switzerland and the UK. Voice editing is not supported anywhere. Neither restriction appears in most write-ups.