Google rolls out Gemini Omni Flash

Google rolled out Gemini Omni Flash this week across the Gemini app, Google Flow and YouTube Shorts, extending its new video-generation model beyond a stage demo and into consumer-facing products. Google announced the release on May 19 in posts on its Keyword blog and in CEO Sundar Pichai’s I/O keynote transcript. The company said Gemini Omni Flash is the first model in a new Omni family and that it starts with video generation and editing. Google said developers and enterprise customers will get API access in the coming weeks. ### Where is Gemini Omni Flash actually showing up first? Google said the first rollout is inside three places: the Gemini app, Google Flow and YouTube Shorts. In the Omni announcement, Google described the model as able to take images, audio, video and text as inputs and generate video outputs, with image and audio outputs planned later. (blog.google) Sundar Pichai said in Google’s I/O keynote transcript that Gemini Omni Flash was “available starting today” and that Google would also roll it out to developers and enterprise customers through APIs in the coming weeks. That places the launch in both creator tools and Google’s broader software stack. ### What does Google say the model can do? (blog.google) Google CTO Koray Kavukcuoglu wrote that Omni is designed to “create anything from any input — starting with video.” The company said users can combine multiple input types, then refine the result through follow-up prompts rather than restarting from scratch. (blog.google) Google’s product post said the model keeps characters consistent across edits and is built to preserve scene continuity as prompts accumulate. The company also described Omni as grounded in Gemini’s “real-world knowledge,” a phrase Google used to distinguish it from simpler effects or filters. (blog.google) ### How fast is it in practice? PCMag reported that Google planned to make Omni Flash available the same day through the Gemini app, Google Flow and YouTube Shorts after showing it at I/O. In the upstream briefing for this story, PCMag said one generated clip took about a minute to process from source materials. (blog.google) PCMag also reported that Google DeepMind CEO Demis Hassabis demonstrated the model by altering live video scenes and generating stylized educational clips. That test did not amount to an independent benchmark, but it gave one early measure of the workflow Google is trying to push into mainstream products. (me.pcmag.com) ### What makes Omni different from older text-to-video tools? Google said Omni is multimodal in both input and editing, meaning users can start from existing photos, video, audio or text and continue revising through conversation. The company’s examples focused less on single-shot prompt generation and more on iterative changes to footage a user already had. (me.pcmag.com) BetaNews reported that Google presented Omni as a model trained from the ground up for multimodal generation and editing, and said the system was built to understand physical behavior including gravity and fluid dynamics. BetaNews also reported that generated content carries Google’s SynthID watermark. (blog.google) ### What should users watch for next? Google said image and audio outputs will be added to the Omni family over time, beyond the current video-first release. Pichai also said API access for developers and enterprise customers is expected in the coming weeks, which will be the next test of whether Omni moves from a consumer feature into a broader platform tool. (blog.google) (betanews.com)