Google opens Gemini Omni 1.2 Flash to all with 15-second clips and 4K video

Google flipped Gemini Omni 1.2 Flash to general availability on November 5, 2026, pushing the multimodal video model out through the Gemini API and Gemini for business. The marquee change is length: native clips now run a full 15 seconds, while earlier versions topped out at 10. Google frames the release around three wins — longer native video, multi-turn conversational editing, and tighter adherence to reference material.

Conversational editing is the feature creators will feel fastest. Instead of scrapping a clip and re-prompting from zero, users refine it through back-and-forth prompts, and the model leans on context from earlier exchanges to keep scenes consistent across edits. Reference following got a lift too: users can now feed 7 to 10 reference images into a single generation pass, which should help lock down brand assets and recurring characters.

Output quality climbs alongside the control. The model generates video at up to 4K resolution and attaches synchronized audio, so clips arrive with sound rather than as silent footage waiting on a separate soundtrack. Every generated video carries Google's built-in SynthID watermark.

There are plenty of doors in. Developers reach it through the Gemini API and Google AI Studio, creators through Google Flow, and third-party platforms Runware and Segmind also carry the model. Some access is free for eligible users in environments like YouTube Shorts, and subscription tiers — Google AI Plus, Pro and Ultra — offer another route.

Watch the meter. According to Crypto Briefing, API pricing for 720p output runs $0.02 to $0.10 per second, which puts a full 15-second 720p clip somewhere between $0.30 and $1.50. Teams generating at scale will need to model spend carefully as clips get longer and resolution creeps toward 4K.

The Omni line first showed up on May 19, 2026, at Google I/O. Version 1.1 Flash brought scene extension and first-and-last frame control, letting users pin how a clip starts and ends while the model fills in the motion. Version 1.2 keeps those tools and layers on finer command over output resolution — and the general availability label tells businesses Google now considers it production-ready, not just a lab toy.