YouTube brought two significant additions to its Shorts platform at Google I/O 2026 on May 19: a personal avatar system that generates video in a creator's own likeness, and an expanded remixing capability powered by Gemini Omni Flash. Both features were announced in the YouTube Community forum and detailed in updated Help Center documentation published around the same time. The rollout is gradual, and neither feature is yet available to all users.

The scale of what is being deployed is considerable. According to YouTube's updated Help Center documentation, the avatar system allows eligible creators to produce videos that "look and sound like you, safely and securely" - constructed from a single biometric recording session that captures both face and voice. Gemini Omni, meanwhile, gives users the ability to remix existing Shorts using text prompts, uploaded reference photos, or AI-suggested edits, transforming content created by others into something new while maintaining attribution links back to the source.

These are not minor feature additions. They represent a structural shift in how YouTube treats video creation: less recording, more generation.

How the avatar system works

The process starts with what YouTube describes as a "secure live selfie capture." Creators record their face and voice through either the main YouTube mobile app or the YouTube Create app. The recording can only be done once - or updated through a retake - and the resulting avatar is tied to the account that created it. Minors are explicitly excluded: to create an avatar, a person must be 18 or older and must own the channel they are creating content for.

Once the avatar exists, the creation workflow is straightforward. Opening the AI playground via the Create button in the YouTube app leads to a "Make a video with my avatar" option. The creator types a text prompt describing the scene - the Help Center gives the example "I am flying high in the sky, up above the clouds saying I can fly" - and the system generates a video featuring the avatar in that scenario. The same functionality is available inside the YouTube Create app, where the avatar appears as a selectable reference below the generation prompt field.

There is also a direct Shorts integration. On eligible Shorts in the feed, a Remix - Reimagine path leads to an "Add me to this scene" option, allowing a creator to insert their avatar into someone else's existing Short. This connects the avatar system directly to the broader remixing infrastructure rather than treating it as a standalone tool.

The guidelines around recording quality are specific. YouTube requires phone placement at eye level and a centered position in frame. Lighting must be neither too dim nor too bright. Eyes, nose, and mouth must be visible; standard glasses are acceptable, but sunglasses, masks, and hats are not. The audio environment must be quiet, with no other voices present. Backgrounds must contain no other people or images of faces. These constraints suggest the technical requirements for biometric capture are tight - the system is not designed for casual, imprecise input.

Data retention and deletion

The retention structure is strict in some respects and has important nuances in others. According to YouTube's documentation, if an avatar is not used to create new content for 3 years, it is automatically deleted. When a creator voluntarily deletes an avatar, the recorded selfie and voice data "is permanently removed from YouTube."

What is notable - and what creators should understand precisely - is that avatars and videos are managed as separate entities. Deleting an avatar does not automatically delete videos published using that avatar, and deleting a video with an avatar in it does not remove the avatar from the account. The two systems are independent. A creator who wants to fully remove their presence from the platform's AI systems must take separate actions for each.

Content created using the avatar is subject to standard remix rules. Creators can opt their Shorts out of video remixing through YouTube Studio, and doing so after remixes already exist will delete those derivative works. The opt-out path runs through YouTube Studio on desktop: Content - video title or thumbnail - Show more - Shorts remixing - Allow only audio remixing - Save.

Gemini Omni in Shorts remixing

The Gemini Omni Flash integration follows a path that YouTube had been building toward since at least September 2025, when it launched Extend with AI for Shorts. That earlier tool allowed creators to add AI-generated continuations to existing Shorts. The Gemini Omni integration goes further: it allows transformation of a Short's visual content through prompts, rather than just extension.

According to YouTube's Help Center documentation on remixed content, the system supports several input methods. A creator on an eligible Short taps Remix and can either enter a custom text prompt, select from AI-suggested prompts, or upload up to 3 reference photos from their gallery. Tapping Next generates the video. The generated output automatically creates new audio based on the prompt and the context of the original scene.

PPC Land's coverage of the May 22, 2026 Creator Insider announcements noted that creators can remix up to 10 seconds of an eligible Short through the three input methods. The feature currently supports English language prompts only, and the rollout excludes the European Union and the United Kingdom - a restriction consistent with how YouTube has treated several AI features over the past year, including the February 2026 test of AI-powered Reimagine and Add Object tools.

The Gemini Omni model itself is significant context. As PPC Land reported from Google Marketing Live 2026, Gemini Omni combines Gemini reasoning with generative media capabilities across text, images, and video simultaneously. It is the same model being integrated into Asset Studio for Google Ads creative this summer. Its deployment across both consumer Shorts and professional advertising tools on a similar timeline reflects how Google is positioning the model as a horizontal capability across its products.

Disclosure, watermarking, and the label framework

Every video generated using an avatar carries automatic AI disclosure. According to YouTube's documentation, content produced through the avatar feature includes "visible watermarks and digital labels like SynthID and C2PA to disclose it as AI-generated content." This applies regardless of whether the video is downloaded or published.

The SynthID watermark is an embedded signal developed by Google DeepMind that persists through editing and reformatting. C2PA - the Coalition for Content Provenance and Authenticity standard - provides a metadata-based provenance layer that a growing number of platforms have adopted. Google integrated SynthID verification into the Gemini app in December 2025, allowing users to check whether a video was created or modified using Google AI tools.

The broader YouTube disclosure policy, detailed in the platform's altered content guidelines, draws a clear line between content that requires creator disclosure and content that does not. Applying beauty filters, using green screen, or improving audio with effects does not require disclosure. Digitally generating content to replace a real person's face, synthetically generating realistic footage of a real place, or cloning another person's voice does require it. Importantly, according to YouTube's documentation, "creators who make a post or YouTube Short using one of YouTube's generative artificial intelligence tools don't need to take extra steps to disclose. The tool will automatically disclose the use of AI for creators."

For sensitive categories - elections, ongoing conflicts, natural disasters, finance, and health - a more prominent label appears in the video player itself, not only in the expanded description. YouTube's documentation states that this type of information "can greatly affect the well-being, financial security, or safety of people and communities."

Disclosure, according to YouTube's policy, carries no monetisation penalty: "Disclosing content as altered or synthetic won't limit a video's audience or impact its eligibility to earn money." Non-disclosure carries the opposite risk. YouTube may proactively apply labels to undisclosed AI content, and "creators who consistently choose not to disclose this information may be subject to penalties from YouTube, including removal of content or suspension from the YouTube Partner Program."

Creator controls and the opt-out structure

YouTube's announcement, posted in Spanish in the YouTube Community by Carlos from TeamYouTube following the Google I/O announcement, addresses the question of creator control in explicit terms. According to the post, creators "decide how your work is used" and can disable visual remixing at any time, at both the per-video level and in bulk. Disabling remixing for a video after remixes already exist deletes those remixes.

Attribution is automatic. Shorts produced from another creator's video using AI tools link back to the original work, consistent with how existing remix features - Cut, Green Screen, Collab, and Sound - already handle attribution. The Sound Library page credits the source of remixed audio. The Shorts player displays a direct link to the source video for remixed visual content.

YouTube also noted in the community post that likeness detection tools are expanding. According to YouTube, "likeness detection will expand to all eligible creators over 18." The tool allows qualifying creators to find content on YouTube where their face appears to have been altered or generated by AI, review it, and take action. Anyone can request removal of relevant content through YouTube's privacy claim process.

The YouTube Create app

The YouTube Create app - distinct from the main YouTube mobile app - is the second path for both avatar creation and AI-assisted video production. Both apps share state: according to YouTube's documentation, "if you create, update, or delete your avatar in one app, those changes will automatically reflect in the other app."

The Create app has a dedicated homepage banner for the avatar feature and a separate You tab with a Manage Avatar option, allowing creators to update or delete their avatar without going through the main video creation flow. Feedback on generated videos can be submitted from within the app by tapping the share icon from the generated video preview.

The presence of two apps covering similar functionality - with the Create app positioned as a more production-focused environment and the main YouTube app as the consumer-facing entry point - creates some surface area for confusion. YouTube's documentation addresses this directly by specifying which interface each step applies to and noting the cross-app synchronisation of avatar state.

Context for the marketing community

The implications of these features extend beyond individual content creation. YouTube confirmed at Google NewFront 2026 in March that Shorts is generating 200 billion daily views, a figure that gives the platform's creative tooling decisions significant reach. Any format-level shift in how Shorts content is produced - from captured video toward generated or remixed video - changes the nature of the content inventory that advertisers are buying against.

The avatar feature in particular has potential advertising applications. If a creator can generate content at scale using their own likeness, the volume of original-format content on the platform could increase substantially without a proportional increase in production cost. That changes supply dynamics. For advertisers running contextual or brand-safety-based buying, the ratio of generated to captured content is a variable that did not previously exist at this scale.

The disclosure framework - automatic SynthID and C2PA labels on all avatar-generated content - provides a machine-readable signal that, in principle, could feed into brand safety classification systems. Whether buyers are currently filtering on that signal is a separate question, but the infrastructure for doing so now exists on the platform.

PPC Land tracked the February 2026 test of AI remix tools and noted at the time that the coupling between AI remix opt-out and general video remix opt-out attracted criticism. The current rollout appears to have separated those controls, though the full mechanics of the opt-out structure are worth reviewing by creators who have content on the platform that they wish to protect from generative reuse.

Timeline

Summary

Who: YouTube, the Google-owned video platform, with the announcement made at Google I/O 2026 and posted in the YouTube Community by Carlos from TeamYouTube. The features are directed at channel owners who are 18 or older.

What: Two connected AI creation capabilities for Shorts: a personal avatar system that generates video using a creator's biometric likeness, captured via a one-time face and voice recording; and Gemini Omni Flash remixing, which allows users to transform eligible Shorts using text prompts, AI suggestions, or up to 3 reference photos. All generated content carries automatic SynthID and C2PA watermarks. Creators retain opt-out controls at the per-video and bulk level.

When: Announced at Google I/O 2026 on May 19, 2026. The features are rolling out gradually and are not yet available to all users. English-language prompts only. EU and UK are excluded from the Gemini Omni remixing feature at launch.

Where: Available through the YouTube mobile app and the YouTube Create app on iOS and Android. The Gemini Omni remixing feature is also accessible directly from the Shorts feed via the Remix button on eligible Shorts.

Why: YouTube is expanding its creation tooling to reduce the friction between an idea and a publishable video, with generated and remixed content positioned as a complement to traditionally recorded material. The attribution and disclosure framework - automatic SynthID watermarks, C2PA labels, and remix links back to original creators - is designed to maintain transparency and creator control as AI-generated content scales on the platform.

Share this article
The link has been copied!