Gemini Omni (Google)

Gemini Omni is Google's multimodal model line for generating and editing video. Unlike its predecessor, Veo, which produces one clip per request, Omni works like a conversation: a developer or user generates a clip, then asks for changes in plain language over several turns. Announced at Google I/O in May 2026, Gemini Omni 1.1 Flash became the production model on the Gemini API on August 27, 2026, and as of October 2026 it is the video model Google points developers to, as the Veo 3.1 previews shut down on October 22, 2026. It also powers video creation in the Gemini app, Google Flow and YouTube Shorts.

What It Does

Omni 1.1 Flash generates clips of 3 to 10 seconds at 24 frames per second, with native audio (dialogue, sound effects and music), and extends them in 10-second steps up to 40 seconds. It accepts text, images, video and audio as input, including first and last frames, subject reference images and up to three short reference clips, and it can edit uploaded clips of up to 10 seconds. Output runs from a 360p draft through 720p (the default) to 1080p and 4K, but Google's documentation says the 1080p and 4K outputs are upscaled rather than generated natively. Only 16:9 and 9:16 aspect ratios are documented, which suits widescreen and vertical video.

The defining feature is the stateful Interactions API. Each turn can refer back to the last, so a shot can be refined step by step without restating the whole prompt. The state lives on Google's side, and turning storage off (store=false) disables follow-up edits. Google's documentation also notes that Omni does not edit voices and does not accept system instructions or sampling settings such as temperature.

Price and Access

Omni is priced like a language model rather than per second: $17.50 per million video-output tokens, billed at 5,792 tokens per second of 720p video, which Google rounds to about $0.10 a second. Google says a 360p draft is up to 60% faster and costs about a third as much, which makes iteration cheap. There is no free tier. At 720p, standard Veo 3.1 costs about four times as much.

Developers reach Omni through the Gemini API, Google AI Studio and Google's enterprise platform (Gemini Enterprise Agent Platform, formerly Vertex AI), where it is still in preview. Google also supplies it to partners including Adobe Firefly, Figma Weave and Runway, which has offered Gemini Omni Flash since June 2026 at the same 720p rate as Google. The earlier gemini-omni-flash-preview model retires on October 22, 2026 as well, so early adopters must move to the 1.1 model ID.

Limits and Safeguards

Google's restrictions are explicit. Editing or extending uploaded videos is not available in the European Economic Area, Switzerland or the UK, and uploading or editing certain recognizable people is not supported anywhere. Every output carries Google's invisible SynthID watermark, which can be detected programmatically and helps with the platform rules covered in AI content labeling and C2PA. Because Omni can edit real footage, consent for the people in source video matters more than it does for pure generation (see deepfakes).

How It Compares

On Artificial Analysis's arenas as fetched on October 20, 2026, Omni 1.1 Flash ranked 9th for text-to-video with audio and 4th for image-to-video, well ahead of Veo 3.1 but behind the leaders, Alibaba's Wan 3.0, Utopai X and ByteDance's Seedance 2.5. Its strengths are price, conversational editing and Google's documentation; its weak spots are short clips, upscaled high resolutions and regional limits on editing uploads. Against a creative suite such as Runway, it is a model rather than a studio, built for developers who want to put generation and editing into their own product. Omni shares its name with Google's Gemini language models and reflects the wider move toward multimodal AI that reads and writes several media at once. For the broader landscape, see text-to-video.