Google Flow integrates Gemini Omni and AI agents

Google Flow adds Gemini Omni Flash, creative agents and mobile apps so you can generate and edit videos, images and music from a single platform. Google is also updating Flow Music with clip-by-clip editing, alternate song versions and AI-guided music video creation.

The company introduced Flow last year as a tool for filmmakers. Since then, it has turned it into a creative studio with video and image generation and editing features, available in more than 140 countries. Earlier this year, it added Flow Music, powered by the Lyria 3 Pro music model.

An agent for every stage of the project

The main new feature is Google Flow Agent, an assistant that can help you from the first idea through to organizing the final files. It does more than generate content: it can also reason about the tasks you assign and work under your control.

In practice, you can use it to:

• Suggest dialogue between characters or recommend changes to a plot.
• Create several versions of a scene at the same time.
• Apply a modification to multiple assets through batch editing.
• Organize your materials into collections and rename them automatically.

The agent is now available to all Google Flow users globally. Google presents it as a collaborator that can speed up repetitive tasks, not as a replacement for creative decisions.

Gemini Omni Flash edits videos from instructions

Gemini Omni Flash combines Gemini's understanding capabilities with Google's media generation models. Its goal is to let you start with a video, image or other reference and transform it through conversational instructions.

For example, you could ask it to change the style of a scene, combine a real recording with generated elements or adjust a sequence without redoing the entire project. Google also highlights improved character consistency: their identity and voice should remain more stable across scenes.

The feature is available in Flow for Google AI subscribers worldwide. This does not mean editing is fully automatic or perfect, but it does reduce the need to switch between several tools when testing changes.

Custom tools without coding

Flow also lets you create your own tools and workflows using natural language. You can describe, for example, an image editor, a video resizer or a visual effect, and the platform will generate that resource without requiring coding experience.

You can also share your tools with other users. They can reuse and adapt them. One of the first examples is pixelBento, created by László Gaal to apply post-production effects such as lo-fi or glitch aesthetics.

All Flow users can use existing tools. Creating and modifying new tools is available to Google AI subscribers.

Flow Music adds clip-by-clip control

The Flow Music updates are designed for people who want to modify a song without having to start from scratch. You can now select a specific section and change only that fragment:

• Rewrite or translate the lyrics in one section.
• Change the rhythm or style of a musical drop.
• Take a section and extend it in another direction.
• Transform an entire song while keeping its melody and structure.

You can also create an alternate version of a track, such as turning a song into a lo-fi studio version without altering its core composition.

With Gemini Omni Flash, Google AI subscribers can also direct music videos through instructions about the style, characters, scenes and rhythm of the song.

Apps for creating on mobile

Google is launching mobile apps for both services. The web version will continue to include all features, while the apps are designed to let you continue a project when you are away from a computer.

For users over 18, Flow is available on Android in beta and will arrive on iOS soon. Flow Music is already available on iOS and will get an Android version later.

Google's direction is clear: Flow is no longer just a video generator. It is becoming a workspace with agents, conversational editing and tools created by users themselves. The important question will be how much real control creators retain and what limits emerge when these features are used in more complex projects.