Google is adding self-cloning AI avatars to Flow, letting users generate selfie-style videos of themselves with its new Omni Flash video model, according to Wired. The update turns personalized deepfake-style video from a specialist workflow into a product feature inside Google’s AI creation software.
Google announced the Flow overhaul at its annual I/O developer conference in Mountain View, California, where it also showed how Android XR smart glasses could turn cameras into AI eyes. The move lands after OpenAI wound down its Sora app in less than seven months, leaving Google to push a similar creator-facing idea into Flow, Gemini, and YouTube rather than a standalone AI social app.
Google Updates Flow With AI Avatars for Selfie-Style Video Generation
Flow is Google’s AI creation environment for generating and remixing images and videos. It launched last year under Google’s experimental Labs division, and the new avatar feature gives users a way to place a digital version of themselves inside AI-generated clips.
Elias Roman, vice president of product management at Google Labs, framed the tool as a production shortcut for people who want to appear in videos without filming every take.
“This is for creators who want to bring themselves into their content but don't want to have to shoot themselves,” Roman says.
The setup process is built around a live self-capture. Users go into Flow settings, scan a QR code on their phone, record themselves saying a string of numbers aloud, and move their head around so Google can capture multiple angles.
That process gives Google material to recreate both appearance and voice. Roman told Wired: “You can capture your voice and your visual identity from multiple angles and have that show up with pretty high levels of fidelity.”
The feature is not only for Flow. Wired reports that these avatars are also available through the Gemini app and YouTube, widening the potential audience beyond users who already work inside Google’s creative tools.
MLXIO analysis: Google is not just adding another AI filter. It is turning personal identity into an editable media asset — one that can be moved between scenes, backgrounds, outfits, and short-form formats.
Flow’s New Video Model Pushes Google Deeper Into Generative AI Media
The new engine behind Flow is Omni Flash, which succeeds Veo. Wired reports that Omni Flash is designed to produce richer detail across clips and improve character consistency, a weak point in earlier Flow generations where characters could warp across successive videos.
That consistency matters because avatars only work if the digital person remains recognizable from scene to scene. A convincing face in one shot is not enough. For creator workflows, the model has to preserve identity while changing settings, prompts, clothes, and framing.
Roman demonstrated that by generating a tongue-in-cheek video of himself teasing the Flow team in front of a dumpster fire. He then asked Flow to change the background and shirt color while keeping his avatar’s details intact.
Google is also adding workflow features around generation. Wired says users can repeat custom instructions and create automated workflows that sort similarly styled clips into folders. Those changes fit the broader I/O push around AI agents and vibe coding, where Google is trying to make automated task-handling and natural-language software creation more mainstream.
For more on how Google framed AI at this year’s developer event, see MLXIO’s coverage of Singularity Claim Turns Google I/O Demos Into a Bet and Google I/O Puts Gemini on Trial as Claude Grabs Devs.
| Product or feature | How it handles AI likeness | Key limit or safeguard in source material |
|---|---|---|
| Flow avatars | Users create an AI version of themselves for generated clips | Initial focus is on users creating themselves, not other people |
| OpenAI Sora app | Users could generate videos of other users depending on settings | OpenAI wound it down after less than seven months |
| YouTube Shorts avatars | Creators can make avatars that “look and sound like you,” according to The Verge | Gradual rollout; creators must be at least 18 and have an existing YouTube channel |
AI Selfie Videos Raise Fresh Deepfake and Consent Questions
Google’s first safety distinction is scope. Unlike OpenAI’s Sora app, where users could generate clips of other users depending on that person’s settings, Google’s initial Flow avatar focus is on letting people generate AI versions of themselves only.
Every video generated with the Omni model, including avatar clips, includes Google’s SynthID watermark, Wired reports. On YouTube Shorts, The Verge reports that avatar videos will also carry visible AI-generated flags and digital labels including SynthID and C2PA.
Those labels matter because the feature sits close to the same problem Google is trying to contain elsewhere: synthetic clips that viewers may read as real. The Verge’s reporting on the Shorts rollout specifically cites YouTube’s struggle with AI slop, deepfake scams, and impersonations.
The consent model appears stronger than open-ended face cloning because it requires live capture of the user’s face and voice. But the hard question is not only whether Google can verify creation. It is whether viewers will reliably understand when a clip is synthetic, especially inside fast-moving social feeds.
Practical safeguards already described in the supplied reporting include:
- Self-capture: Flow users record their face and voice through a phone-based setup.
- User limitation: Google’s initial focus is self-avatars, not generating other people.
- Watermarking: Omni-generated videos include SynthID.
- YouTube controls: The Verge reports creators can delete their avatar or videos where it appears, and unused avatars are deleted after three years.
MLXIO analysis: The product tension is obvious. The more lifelike and easy the avatars become, the more useful they are for creators — and the harder platforms must work to keep disclosure visible and credible.
Developers, Creators, and Regulators Will Scrutinize Google’s Avatar Rollout
The next test is rollout detail. Wired confirms the Flow feature and its availability through Gemini and YouTube, but the supplied material does not specify Flow pricing, geographic limits, account eligibility, or usage caps.
Creators will likely judge the tool on four concrete things: likeness accuracy, voice quality, editability, and whether repeated generations preserve identity without visual drift. Google is clearly targeting that last issue with Omni Flash’s character-consistency improvements.
Brands and professional teams will have a different checklist. They will want clearer answers on rights, approvals, labeling, and whether avatar content can be managed safely across accounts and platforms.
The competitive signal is sharper than the demo. OpenAI tried an AI-first social app around synthetic video and shut it down quickly. Google is taking a more distributed route: put avatars into Flow, Gemini, and YouTube, then let creators test whether AI self-video belongs inside existing products.
The watch item now is not whether Google can make a convincing AI version of a user. Wired’s demo suggests it can. The real test is whether Google can make synthetic identity useful without making authenticity feel optional.
Impact Analysis
- Google is making personalized deepfake-style video creation a mainstream product feature rather than a specialist workflow.
- The tool lowers production barriers for creators who want to appear in videos without filming every take.
- Embedding avatars across Flow, Gemini, and YouTube could significantly expand access to AI-generated likeness and voice cloning.
