Gemini Omni Flash

Gemini Omni Flash is Google's video model for the Gemini API, released in public preview on 30 June 2026 as gemini-omni-flash-preview. Google's own documentation now recommends it as the default model for video generation, ahead of Veo 3.1 — a notable shift, because Veo had been the company's flagship video line since 2024 and Omni Flash comes from the Gemini family instead. What separates it from Veo is the working method. Veo takes a prompt and returns a finished clip; Omni Flash is built for multi-turn work through the Interactions API, so a video can be refined across several conversational turns instead of being regenerated from a rewritten prompt. Google lists its strengths as video coherence, multi-input reasoning with text, images, audio and video supplied simultaneously, character consistency and factual accuracy. Inputs are text and images turned into short videos; refinement happens in natural language. Billing shows how differently the two models are built. Veo 3.1 is priced per second of finished video (0.40 USD at 720p and 1080p, 0.60 USD at 4K, less for the Fast and Lite variants). Omni Flash is priced in tokens like any other Gemini model: 1.50 USD per million input tokens covering text, image, video and audio alike, 9.00 USD per million text output tokens and 17.50 USD per million video output tokens, with video counted at 5,792 tokens per second of 720p footage — roughly 0.10 USD per second, comparable to Veo 3.1 Fast. There is no free tier. Google publishes no parameter count, architecture description, context window or knowledge cutoff for this model, and it still carries the preview label, meaning the behaviour and the model ID can change. It arrived on the same day as Nano Banana 2 Lite.

▸Videos

No videos yet.

▸Reviews