Google Releases Gemini Omni 1.2 Flash With 15-Second Clips and 30% Price Cut

Google has made Gemini Omni 1.2 Flash generally available, extending maximum clip length to 15 seconds and cutting API pricing by roughly 30 percent. The update marks a measurable step forward in the video-generation race, though Google acknowledges several capabilities remain in development and independent benchmark scores for the new version have not been published.

What changed

The headline changes are length and price. Clips now run from 3 to 15 seconds in one-second increments, up from a 10-second ceiling on Omni 1.1 Flash. Google prices 720p output at $0.07 per second, down from an effective $0.10 per second on the prior model, with similar proportional savings at other resolutions [1][2]. A full 15-second 720p clip costs about $1.04; 1080p (upscaled) runs roughly $0.104 per second, and 4K about $0.209 per second [10].

The model is rolling out across the Gemini API, Google Cloud (with enterprise-grade infrastructure and Provisioned Throughput), Flow, Google Vids, Google Ads, and the Gemini app for AI Ultra subscribers [1][2]. Every output carries an imperceptible SynthID watermark [1][2][8].

Editing workflow and reference handling

Google frames cost control as a workflow: draft at 360p or 720p, refine with multi-turn conversational edits across the full 15-second clip, then upscale only the final take [1][5]. The conversational editing lets users swap products, backgrounds, lighting, camera angle, or aspect ratio without regenerating from scratch [1][2].

Reference adherence is a focus. The model accepts character, product, or style images; motion clips up to three seconds; and audio cues (the latter gated behind allowlisting on Google Cloud) [1][2]. Google claims faces, garments, packaging typography, and brand colors stay consistent across camera angles, and shows a virtual try-on demo where a single model photo and front/back product shots of three outfits produce a 15-second 360-degree sequence with consistent identity and garment detail [1].

Consistency improvements also extend to first-to-last frame interpolation, camera moves (FPV flythroughs, zoom-outs, orbits with fewer jump cuts), and speech: clearer lip-sync, more natural dialogue, and distinct multi-speaker conditioning [1][8].

Availability and limits

The model is generally available on Google Cloud and the paid tier of the Gemini API as of November 5, 2026 [11]. Consumer access is rolling out across Flow, Google Vids, Google Ads, and the Gemini app [1][7]. According to Marktechpost's reading of the API documentation, editing uploaded videos is not available in the EEA, Switzerland, or the UK [5]; the model card [6][9] does not mention this restriction. Uploaded videos for editing can be up to 15 seconds; for extension, any duration is accepted but only the last 10 seconds are used [11]. (Marktechpost reports a 10-second cap for editing inputs [5], which conflicts with the official documentation.)

Google's own blog lists known gaps: "development is ongoing to further improve things like fast sports action, rapid script changes and complex physics" [1]. The DeepMind model card notes that "maintaining complete consistency throughout edits, generating scenes with complex motion, or rendering perfectly accurate text remains a challenge" [9].

No independent benchmarks yet

Google has not published benchmark scores for version 1.2. Its quality claims rest on side-by-side demos against Omni 1.1 Flash [1][5]. An earlier Omni Flash version took the #1 spot on the Arena text-to-video board in June, but those rankings predate 1.2 [5]. Early partner feedback is positive but self-selected: Adobe, Photoroom, QuickFrame AI, invideo, Monks, Whering, Figma Weave, STAN, Picsart, Runway, Artlist, FLORA, Krea, and Resemble AI are quoted in the release materials [1][2].

Competitive context

Beam.ai's analysis positions Omni 1.2 Flash as "the obvious default for short product and social video" at its price point, while noting that on Artificial Analysis leaderboards the top-ranked text-to-video model, Wan 3.0, costs about $12/minute, and the top image-to-video model, Seedance 2.5, costs more than five times as much per minute as Gemini Omni 1.2 at 1080p [4]. HeyGen Video 1 at 768p ($2.40/minute) and MiniMax H3 at 768p ($4.80/minute) sit near the top of the text-to-video board [4].

OpenAI removed its Sora 2 models and Videos API on September 24, 2026, and has not announced a replacement; its new GPT Image 2.6 adds animated GIF output but no video API [4]. Google's Veo 3.1 offers 4-, 6-, or 8-second clips [7]; on OpenRouter, Veo 3.1 Fast is listed from $0.10/second and Veo 3.1 Lite from $0.05/second [3]; Google's Vertex AI documentation prices Veo 3.1 at $0.20 per generated video at 720p or 1080p [7].

Pricing mechanics

On the Gemini Developer API, video output is billed by tokens: 5,792 tokens per second of 720p video at $12 per million output tokens, yielding the ~$0.07/second effective rate [10]. Omni 1.1 Flash used the same token rate but was priced at $17.50 per million video tokens (~$0.10/second); effective December 3, 2026, all 1.1 requests automatically route to 1.2 at the new rates [10]. OpenRouter lists the model at $1.50/M input tokens and $9.00/M output tokens, with video+audio at $0.023172/second (presumably 360p) [3]; the per-second prices at higher resolutions ($0.104 at 1080p, $0.209 at 4K) come from Google's pricing documentation [10].

What remains unknown

• Independent benchmark results for 1.2 vs. competitors
• Real-world reliability of conversational editing at scale across diverse prompts
• Whether the 30% price cut holds under sustained production loads with Provisioned Throughput
• Timeline for lifting audio-reference allowlisting and regional editing restrictions
• Progress on the acknowledged gaps in sports action, rapid script changes, and complex physics

Google says development is ongoing on those fronts [1]. For now, the release delivers a concrete spec bump — longer clips, lower price, broader editing — backed by enterprise infrastructure and a watermark, with quality claims that await independent verification.