Google adds an agent, natural-language tool building, and Gemini Omni to Flow

Three distinct surfaces, not one feature drop

Google's update to Flow bundles three separate interaction models under one announcement. Gemini Omni Flash is a generative video model. Google Flow Agent is a conversational planning layer. Google Flow Tools lets users build editing utilities by describing them. These are different frontends onto the same creative project, and Google is explicit that they carry different access rules.

Omni Flash and Tools creation are gated to Google AI subscribers globally. The Agent, and the ability to use existing Tools, are available to all Flow users. That split matters: the free tier gets orchestration and consumption, while the paid tier gets the generative model and the authoring rights. Google is drawing the monetization line between using and making.

Omni Flash framed against Nano Banana

Google describes Omni Flash directly by analogy: "You can think of Omni like Nano Banana, but for video." It positions the model as one that can "create anything from any input, starting with video," combining Gemini's reasoning with Google's generative media models.

The concrete claim worth noting is character consistency — Google says "identity and voice are preserved across every scene." For a video tool, that is the frontend problem that has historically broken the illusion: a character's face or voice drifting between shots. If Omni Flash holds identity across a conversational editing loop, the value is in the iteration workflow, not any single clip.

Tools turns the app into a platform users extend

Flow Tools is the most structurally interesting change. Rather than shipping a fixed set of editors, Google lets users describe a utility in natural language — "a particular image editor, video resizer or custom shaders" — and generate it with no code. Created tools can be shared and remixed by other users.

The named example is a partner-built tool: László Gaal's "pixelBento," which applies lo-fi and glitch post-processing effects. This reframes Flow from an application with a feature list into a small ecosystem where the feature surface is populated by its own users. The remix mechanic, restricted to subscribers, is how Google keeps that ecosystem inside the paid tier while letting everyone consume its output.

Section-level editing and the mobile-web divide

Flow Music gains granular control that mirrors the video side: highlight any part of a song, rewrite or translate lyrics, restyle a beat drop, or sample a section and extend it in a new direction. Covers preserve melody and structure while changing style. Omni Flash also arrives here to direct music videos conversationally.

Google is candid about the mobile tradeoff: "the web versions remain the go-to platforms for access to all capabilities and features," with the apps offering flexibility on the go. The Flow app is Android-only in beta with iOS pending; Flow Music is iOS-only with Android pending. Both are 18+. This is a deliberate partial rollout, not feature parity.

What the tiering signals for creative-tool builders

The specific implication here is architectural. Google split its creative suite into a free orchestration and consumption layer and a paid generation and authoring layer, then let users manufacture the feature set themselves through Tools. Teams building creative frontends should read this as a bet that the durable moat is the underlying generative model plus the sharing ecosystem — not any individual editor — and that the interface is increasingly something users assemble in natural language rather than something the vendor ships fully formed.