Gemini Omni Flash lengthens video clips and lowers costs
Google makes Gemini Omni 1.2 Flash more widely available for video generation. The model extends to 15 seconds, gains a conversational editing mode, and reduces its price by 30%.
Google DeepMind makes Gemini Omni 1.2 Flash generally available for video generation, with clips from 3 to 15 seconds, a conversational editing mode, and a price 30% lower than the previous model. The announcement, dated November 5, 2026, targets both creative uses and marketing workflows, as it combines longer duration, simpler iterations, and better-framed output. Each generated video also includes an imperceptible SynthID watermark. The whole aims at both production and usage control.
Un modèle plus long, moins cher, et pensé pour corriger sans tout recommencer
The most visible novelty concerns clip duration. Gemini Omni 1.2 Flash increases from a maximum of 10 seconds to 15 seconds, with a price announced at 7 cents per second for 720p video. Google also states that editing can be done through continuous conversation: the user adjusts a background, lighting, or aspect ratio without restarting the entire clip. On paper, this avoids discarding an almost perfect sequence due to a misplaced detail. Teams testing this type of tool know that this is often where time is lost.
Google also presents a stepwise workflow. Drafts are first created in 360p or 720p, corrections are then made on precise details, and the final version is assembled in 1080p or 4K. This logic applies to studios as well as content teams, as it reduces the cost of successive trials. The model is accessible via the Gemini API, Google AI Studio, Google Cloud Console, Flow, Google Vids, the Gemini app for Google AI Ultra subscribers, and Google Ads.
Des références mieux suivies, de la tenue du visage à la typo sur l’emballage
Google also says it improves fidelity to visual references. The model accepts images of characters, product photos, short motion clips up to 3 seconds, and audio cues. According to the announcement, faces, clothing, packaging typography, and brand colors must remain consistent from one camera angle to another. This is a decisive point for advertising, e-commerce, and product demos, where visual drift can sometimes break a scene’s credibility.
An example provided by Google shows a virtual try-on test: from a photo of a model and product images of three outfits, the model generates a 15-second sequence with 360-degree rotation while preserving the model’s identity and the cut of each garment. The company also says it has reduced visual drift and morphing between camera movements. In other words, faces and clothing should less often turn into digital modeling clay.
Ce que change cette version pour les créateurs de vidéo IA
Gemini Omni 1.2 Flash aims to make AI video generation more usable daily, not just more spectacular. Google claims that dialogues are smoother, with clearer lip-sync and better distinction between multiple speakers. The model also supports more complex continuous camera movements, such as drone FPV shots or wide zoom-outs, with fewer unwanted cuts. In this context, the challenge is not only image quality: it is the ability to produce cleaner sequences without multiplying reruns.
The general availability through multiple Google services also confirms a broad integration strategy, from creative studios to advertising products. For teams already testing AI-generated video, the combination of longer duration, lower cost, and iterative editing can change how they prototype a spot, product demonstration, or brand scene. It remains to be seen, in practice, whether the promise of consistency holds on longer and more ambitious projects [to be verified].
By the numbers
- 15 seconds — maximum clip duration, Google announcement of November 5, 2026.
- 10 seconds — previous model limit, according to Google.
- 30% — price reduction announced compared to the predecessor.
- 0.07 dollar per second — announced rate for 720p video.
- 3 seconds — maximum duration of source clips or motion cues accepted as input.