Google puts four generative media models under one Vertex AI roof

What actually shipped, and in what state

Google announced four generative media updates for Vertex AI at Google Cloud Next. Only one of them, Imagen 3, arrived as a straightforward capability improvement. The other three carry qualifiers worth reading closely.

Lyria, the text-to-music model, is in private preview with allowlist. Veo 2's new editing and camera controls are in preview with allowlist. Chirp 3's Instant Custom Voice is new but its rollout state is described only as being included in Chirp 3. Imagen 3's improved generation, inpainting and object removal are the closest to generally usable.

So the headline of "four big updates" is really one shipped feature set and three gated ones. Teams evaluating this should plan around allowlist access rather than assuming they can call these models tomorrow.

The specific claim: a single platform across four media types

The load-bearing sentence in the announcement is about coverage, not any single model's quality.

These updates make Vertex AI the only platform with generative media models across video, image, speech and music.Montana Labs

The addition of Lyria for music is what completes the set. Video (Veo 2), image (Imagen 3) and speech (Chirp 3) were already present; music was the missing category. That is why this bundle was announced together — it converts a partial lineup into a full one.

Whether breadth matters depends on the buyer. A team that needs only image editing gains nothing from music generation living on the same platform. But a team building, say, marketing content pipelines that touch several media types benefits from one billing relationship, one IAM model and one set of APIs rather than stitching together separate vendors.

The most concrete numbers are in Chirp 3 and Veo 2

Two details in the announcement are specific enough to design against. Chirp 3's Instant Custom Voice creates a custom voice from just 10 seconds of audio input. That is a low enough threshold to change how voice-driven products handle onboarding — you can capture a usable sample in a single short recording.

Veo 2's editing story is described in terms of named techniques: interpolation for smooth transitions across frames, and outpainting to extend the frame of existing footage and reformat it for web and mobile. That last point is a reframing tool — taking traditional video and producing vertical or alternate aspect ratios without a reshoot.

Imagen 3's inpainting reconstructs missing or damaged portions of an image, and object removal edits are described as higher quality. These are retouching operations, positioned as replacing manual editing rather than generating net-new images from scratch.

The editing pivot is the real signal for Vertex AI users

The through-line across Veo 2 and Imagen 3 is editing existing assets, not just generating from a prompt. Camera controls, interpolation, outpainting, inpainting and object removal all operate on content that already exists.

That matters because enterprise media work is mostly modification, not creation from zero. A brand has footage and images it already owns and needs to repurpose, clean up and reformat. Positioning these models as refinement tools targets that reality more directly than another text-to-video demo would.

The practical implication for teams on Vertex AI: the near-term value here is likely in Imagen 3's editing and, once off the allowlist, Veo 2's reformatting — turning one asset into many formats — rather than in Lyria's music generation, which remains the most experimental and access-restricted piece of the announcement.