YouTube has released a new artificial intelligence feature that allows creators to generate digital versions of themselves for use in short videos.
The update adds to a growing set of AI tools on the platform.
The feature allows users to create a lifelike avatar that can appear and speak like them. According to YouTube, the tool can be used to “generate videos that look and sound like you, safely and securely.” Once created, the avatar can be placed into short clips or used to produce entirely new scenes based on written prompts.
To use the feature, creators must be at least 18 years old and have a YouTube channel. They are required to record their face and voice through what the company describes as a “secure ‘live selfie’ capture process.” After this step, the system builds a digital model that can be reused in different videos.
For now, the tool is limited to YouTube Shorts and produces clips of up to eight seconds. The company has also introduced labels and watermarking to make it clear when a video has been generated using AI. Users can choose to delete their avatar data and control whether others can reuse their likeness in remixed content.
Similar tools are already available outside the platform, so by offering its own version, YouTube appears to be giving creators a way to experiment with AI-generated content without relying on external software. At the same time, the feature raises questions about how audiences will distinguish between real and generated appearances, especially as AI is becoming more convincing.
In a separate development, Apple is now facing a lawsuit from a group of YouTube creators.
The complaint, filed by channels including h3h3Productions and golf creators such as MrShortGame Golf and Golfholics, alleges that Apple used their videos without permission to train its AI systems. The plaintiffs say the company did so “without permission, payment, or even basic credit.”
At the center of the case is a dataset known as Panda-70M, which was mentioned in a 2025 research paper on video-generation AI. The dataset reportedly contains millions of YouTube clips, organized using links, timestamps, and other identifiers.
According to the lawsuit, compiling such a dataset would require extracting clips directly from YouTube, a process the creators argue bypasses the platform’s protections. They claim this violates the Digital Millennium Copyright Act, which restricts attempts to get around systems designed to protect copyrighted material.
The filing also states that some creators’ videos appear “hundreds of times” within the dataset. The plaintiffs are seeking damages and may ask the court to prevent further use of their content.
Apple has confirmed in its research that YouTube videos were used in training its models, but it has not provided detailed information on how the material was collected.
