YouTube adds AI avatars
YouTube is rolling out AI-generated avatars for Shorts that can be made to look and sound like the creator, which could let creators produce more versions of the same video quickly. That opens new low-friction content formats for brands—think multilingual variants or rapid A/B hook tests—but it also creates fresh licensing questions about who owns and can reuse those synthetic likenesses. (9to5google.com)
YouTube is starting to let creators make Shorts with an artificial intelligence version of their own face and voice, built from a “live selfie” recording inside the main YouTube app or the YouTube Create app. The rollout began on April 8, 2026, and YouTube says it is for adults with an existing channel. (support.google.com, 9to5google.com) The clip itself is short by design: reports on the launch say each generated segment can run up to about eight seconds. YouTube also lets creators drop that avatar into some existing Shorts through a Remix button and a tool called Reimagine. (engadget.com, 9to5google.com) This did not arrive out of nowhere. In February 2025, YouTube added Google DeepMind’s Veo 2 model to Dream Screen, which already let people type a prompt and generate backgrounds or standalone video clips for Shorts. (blog.youtube, blog.google) Then YouTube spent 2025 and early 2026 building the guardrails around identity. In September 2025, YouTube said its likeness detection tool would expand to more creators, and in March 2026 it said journalists, government officials, and political candidates would join the pilot for finding artificial intelligence videos that use their face without permission. (blog.youtube, blog.youtube) That explains the strange split in YouTube’s message now: it is making synthetic versions of creators easier to produce while also building tools to catch synthetic versions they did not approve. The company’s help page says only the account owner can use their avatar to create original videos. (support.google.com, blog.youtube) YouTube is also trying to mark the output so viewers and other platforms can tell it was machine-made. The company says avatar videos are labeled as artificial intelligence, and outside reports say the files carry SynthID and Coalition for Content Provenance and Authenticity markers, which work like tamper-evident tags for media. (support.google.com, phandroid.com) The practical use case is speed. A creator who used to re-record the same 15-second pitch three times for three hooks, or in three languages, can now test multiple versions with typed prompts instead of another camera session. (9to5google.com, engadget.com) The legal wrinkle is that a face and a voice are not just video assets; they are personal likeness rights. YouTube says your selfie video and voice are used to create the avatar, says others cannot use that avatar to make original Shorts, and says you can delete the avatar at any time. (support.google.com, 9to5google.com) Deletion is not a full rewind button. Reports on the launch say YouTube will automatically delete an unused avatar after three years, but any Shorts already published with that avatar stay up until the creator removes those individual videos. (phandroid.com, 9to5google.com) So the new product is not just “make a fake you.” It is YouTube turning a creator’s likeness into a reusable production tool, with labels, access controls, and detection systems bolted on at the same time because the same shortcut that helps a brand make ten versions of a Short can also make identity misuse much cheaper. (support.google.com, blog.youtube, blog.youtube)
