Google Releases Gemini Omni 1.2 Flash With Longer Video Generation and 30% Price Cut

Google has launched the general availability of Gemini Omni 1.2 Flash, updating its generative video model with longer clip lengths, conversational video editing, and lower developer pricing, according to reporting by The Next Web (https://thenextweb.com/news/google-gemini-omni-1-2-flash-video-15-seconds-price-cut).

The model now generates up to 15 seconds of video in a single prompt, up from the 10-second ceiling established when Google previewed the system at its I/O conference in May. Alongside the duration increase, Google lowered API pricing by 30% to roughly $0.07 per second of 720p output, down from the $0.10 per second benchmark set in June.

Beyond raw duration and cost, the update introduces multi-turn conversational editing. Rather than forcing developers and creators to discard and regenerate an entire sequence when making revisions, the interface allows users to prompt iterative adjustments across chat turns—such as modifying lighting, swapping a background scene, or replacing a featured product while retaining the rest of the clip.

Google claims the new version significantly improves visual consistency across camera angles, reducing jump cuts and rendering artifacts while maintaining continuity for faces, clothing, packaging typography, and brand colors. In vendor-run tests conducted by Google DeepMind, human evaluators favored Gemini Omni 1.2 over version 1.1 for visual consistency 62% of the time.

Distribution is expanding across Google's commercial and consumer software stack. Gemini Omni 1.2 Flash is deploying into Google Ads, Google Vids, the Flow video tool, and the Gemini consumer application for Google AI Ultra subscribers. The launch also marks the first time an Omni model has reached general availability on Google Cloud for enterprise customers. Every output file embeds Google's invisible SynthID digital watermark for automated provenance identification.

Despite the iteration, Google acknowledged that the model still struggles with fast athletic action, sudden script transitions, and complex physical dynamics. The company also faces intense international pressure in synthetic video: ByteDance unveiled its Seedance 2.5 model in June with 30-second single-take outputs, and Chinese research labs currently hold nine of the top ten positions on a prominent industry benchmark where Google remains the sole U.S. vendor.