Google Vids now lets you create AI videos starring a digital version of yourself

[Quarry]

Google announced an update to Google Vids that lets users create a custom digital avatar, built from a selfie and a voice recording, that looks and sounds like them for use in AI generated videos. The update also brings Google's multimodal Gemini Omni model into Vids directly, letting people generate videos from a written prompt combined with reference images, with Omni blending those inputs together into the finished clip

Omni can also handle editing tasks on existing footage, swapping out backgrounds, fixing lighting on a video shot on a phone, or adding effects, and now supports step by step edits, meaning changes can be made incrementally as you go rather than needing to start the whole generation over from scratch

The update pushes Vids well beyond its original role as an AI assisted workplace presentation tool toward something closer to an all in one video creation platform, putting it in more direct competition with dedicated AI avatar and video tools like HeyGen, Synthesia, Captions and D-ID. Notably, this arrives not long after OpenAI shut down Sora, which had developed a reputation as one of the more unsettling AI apps around, partly due to how freely it let people generate videos of public figures including its own CEO

Google is building in guardrails specifically learned from that experience, personal avatars are tied to the account holder's own likeness and Google account, invisibly watermarked with SynthID, and access is restricted to users 18 or older in select regions, all of which reads as a fairly deliberate attempt to avoid the exact kind of impersonation controversy that dogged Sora during its short life

Create, edit and star in videos with two Google Vids updatesblog.google

[NeonPhantom32]

The SynthID watermarking and tying avatars strictly to your own account and likeness feels like a very deliberate lesson learned from watching Sora implode over impersonation issues

[MJF02]

Step by step editing instead of regenerating the whole video from scratch every time is honestly the feature that would actually get me to use this regularly, iterative editing matters so much more than raw generation quality