What Google Actually Launched — And What It Can Do

Google’s Flow platform now includes an avatar feature built on the company’s Omni Flash model. The workflow is straightforward: scan your likeness once, generate an AI video scene, drop yourself inside it. No camera crew. No reshoots. No second take.

Elias Roman, VP of product management at Google Labs, demonstrated the feature live. He had previously scanned himself to create a digital clone, and during the demo he inserted that clone into an AI-generated video clip on demand. “This is for creators who want to bring themselves into their content but don’t want to have to shoot themselves,” Roman said. The process was seamless enough that the word “seamless” barely covers it — scan, generate, place, done.

The feature carries echoes of a capability OpenAI briefly offered through its now-defunct Sora app. Where Sora called them cameos or characters, Google calls them avatars. The terminology matters: avatar signals a persistent, reusable digital version of you, not a one-off visual trick.

This is not a research preview or a waitlisted experiment tucked inside Google Labs. The avatar feature sits inside Flow, a product-level tool built for everyday creators, and it extends across the Gemini app and YouTube. That distribution footprint is the real story. Google is not testing whether people want to self-deepfake — it is building the infrastructure to make it a default content creation behavior.

For the creator economy, the immediate upside is obvious: produce content at scale without being physically present for every shoot. For everything else — consent, misuse, identity rights — the questions arrive just as fast as the renders do.

The ‘Creator Convenience’ Framing — And What It Obscures

Google’s pitch for the avatar feature inside Flow is straightforward: Elias Roman, vice president of product management at Google Labs, demonstrated the tool by inserting his pre-scanned digital likeness into AI-generated video clips — no camera crew, no shoot day, no logistics. “This is for creators who want to bring themselves into their content but don’t want to have to shoot themselves,” Roman said. For a solo YouTuber managing production alone, or an indie filmmaker burning through a tight budget, that value proposition is real and immediate.

The convenience framing, though, does specific work. It positions the technology as a personal productivity upgrade — something you do to yourself, for yourself. That framing collapses the moment you ask a different question: what stops the same pipeline from accepting someone else’s face?

The answer is not obvious, because Google has not made it obvious. Coverage of the Flow announcement has treated this almost exclusively as a creator workflow story. Almost none of it has pressed Google on what the identity-verification layer actually looks like — whether a likeness scan requires any authentication tied to a verified account, whether the system cross-checks the submitted face against the account holder’s identity, or whether a bad actor uploading photographs of another person hits any meaningful technical barrier at all.

This is not a hypothetical edge case. Deepfake abuse — non-consensual intimate imagery, fraud, reputational manipulation — already exists at scale without Google’s help. What Google has built is a mainstream, consumer-grade interface that lowers the technical floor for synthetic likeness generation and distributes it across Gemini and YouTube simultaneously. Wrapping that infrastructure in the language of creator convenience does not neutralize the risk; it just makes the risk easier to ignore in early coverage. The consent question does not disappear because the default use case is benign.

Why ‘Consent Starts With You’ Is a Dangerously Incomplete Safeguard

The self-only framing sounds reassuring until you examine what it actually requires: that Google’s identity verification holds under adversarial conditions. Every major face-swap and voice-cloning tool launched with similar assurances. FaceApp, Reface, and a dozen commercial deepfake platforms all positioned consent as the user’s responsibility. Within months of each launch, researchers and bad actors demonstrated workarounds. Google’s avatar feature for Flow — which uses the Omni Flash model to insert a scanned likeness into any AI-generated video clip — is technically more sophisticated, but sophistication has never been the bottleneck for misuse. Motivation is.

The normalization problem runs deeper than any single product policy. When Google ships face-scanning to Gemini and YouTube at consumer scale, it establishes the behavior as routine. Hundreds of millions of people learn that uploading a facial scan and receiving a video clone is a normal, frictionless transaction. That mental model doesn’t stay inside Google’s walled garden. It migrates to less scrupulous platforms, lowers the social threshold for requesting someone else’s likeness, and trains users to treat face data as just another content input. The technical baseline shifts, and every downstream tool benefits from the normalization Google created.

Legislators are already moving. The EU AI Act classifies certain biometric identification systems as high-risk. California’s AB 602 and AB 2602 established likeness-rights protections specifically targeting AI-generated replicas. Illinois has enforced its Biometric Information Privacy Act aggressively enough to produce nine-figure settlements. A mainstream Google product that turns facial geometry into a reusable video asset will accelerate every one of those legislative tracks. Lawmakers don’t draft bills in response to obscure research tools — they respond to products that 14-year-olds are using on YouTube.

Consent starting with the user is not a safeguard. It is an assumption dressed as a policy. The history of analogous tools, the normalization effect of Google-scale distribution, and the existing legislative momentum all point to the same conclusion: the consent conversation cannot end where Google’s terms of service end.

The Déjà Vu Problem — We Have Been Here Before

When Elias Roman, vice president of product management at Google Labs, demoed the new avatar feature for Flow, the journalist watching felt an immediate wave of déjà vu. That reaction is data. The tech industry has run this exact play before — launch a face or voice cloning tool wrapped in creator-empowerment language, watch it go viral, then scramble for policy guardrails after the damage is done.

The pattern is not subtle. Early deepfake apps like DeepFaceLab arrived framed as filmmaking tools. AI voice cloners like ElevenVoice launched promising podcasters and audiobook narrators a shortcut. In both cases, the abuse cases — non-consensual intimate imagery, fraud, political disinformation — materialized faster than any platform response. The policy reaction lagged by months, sometimes years. Victims absorbed the cost of that gap.

Roman’s pitch for avatars follows the same script almost word for word. “This is for creators who want to bring themselves into their content but don’t want to have to shoot themselves,” he said during the demo. The framing centers convenience and creative freedom. It does not center what happens when someone uses another person’s scanned likeness without permission, or when a creator’s avatar is extracted and redeployed outside the platform it was built for.

The structural difference this time is scale and integration. OpenAI’s Sora app offered a comparable selfie-deepfake feature, but Sora was a standalone product that never reached mainstream saturation. Google’s avatar tool ships inside Flow, the Gemini app, and YouTube — platforms with billions of combined users. What took niche deepfake communities two or three years to weaponize, a Google-scale rollout could replicate in a single news cycle. The abuse arc does not disappear because the company behind the tool is larger. It compresses.

What Google Must Disclose — And What Creators Should Demand

Google has not published answers to the most basic questions a creator should ask before scanning their face into Flow: Where is the biometric data stored? Can users delete it on request? Does Google use scanned likenesses to train future versions of its Omni Flash model or any other AI system? Until those answers appear in plain language — not buried in a terms-of-service document — every creator who scans their face is signing a blank check.

The intellectual-property question is equally unresolved. When Google’s tool generates a video of your likeness, the ownership of that output is unclear. Does the creator hold full commercial rights to AI-generated footage of their own face? What license does Google retain over the avatar data and the videos it produces? Creators building sponsored content, licensing deals, or merchandise around their digital likeness need explicit written answers before they monetize a single frame.

Three transparency benchmarks Google must meet — now, not eventually. First, an explicit consent-verification step at the point of scanning, separate from the standard account sign-up flow, that spells out exactly how biometric data will be used. Second, on-device processing or a verified zero-retention policy, so that facial scan data never persists on Google’s servers beyond the active session. Third, a clear, public policy governing law-enforcement requests for likeness data — the same category of transparency Google already provides for search and Gmail data through its annual transparency report.

Creators should not wait for Google to volunteer this information. Demand a data-deletion confirmation in writing. Ask your legal representative to review the Flow and Gemini terms of service before you produce commercial content. Push Google’s creator-partnership teams for a straight answer on model training. The avatar feature is useful. The absence of answers around it is not a minor oversight — it is a structural gap that puts your face, your brand, and your rights at risk.

The Bigger Picture: Google Is Setting the Industry Norm

Google does not ship features in a vacuum. When the world’s most-used search, video, and mobile ecosystem embeds self-avatar technology across Flow, Gemini, and YouTube simultaneously, it sends a single unambiguous message to every competitor: this capability is now baseline. Adobe, Meta, and TikTok are not watching from the sidelines — they are reading the same product announcements and accelerating their own roadmaps. OpenAI already tested this territory with its now-defunct Sora app before pulling back. Google just made the retreat look like a mistake.

That competitive dynamic is the real story. The avatar feature itself — scan your face, drop your likeness into any AI-generated clip without picking up a camera — is genuinely useful for creators. Elias Roman, Google Labs’ VP of product management, framed it plainly: it exists for people who want to appear in their content without shooting themselves. That pitch lands. But once Adobe ships a version, once Meta bakes it into Reels creation tools, once TikTok rolls it into CapCut, the question of who controls a person’s digital likeness stops being theoretical.

The legal infrastructure is not ready for that world. The United States has no comprehensive federal statute governing synthetic likeness. A handful of state laws address nonconsensual deepfakes, primarily in the context of sexual imagery or election interference, but they do not touch the sprawling commercial gray zone where a creator’s avatar could be licensed, resold, or repurposed without their knowledge. Platform terms of service are similarly behind — written for a moment when this capability lived in research labs, not consumer apps.

Google is moving faster than lawmakers, faster than platform-policy teams, and faster than most users’ understanding of what they are agreeing to when they scan their face. That gap is not an accident of timing. It is a structural feature of how large technology companies operate: ship, normalize, then negotiate the rules after the behavior is already embedded. The governance vacuum exposed by this launch will not close on its own. It closes when regulators treat synthetic likeness as a civil rights issue, when platforms write enforceable consent standards, and when creators understand exactly what they are handing over.