Detection bot posts an "Omni LoRA tuning" spec

Google is preparing to let users further train its video generation AI "Gemini Omni" on their own data, according to a leak posted by the X account AI Tracker Bot (@aitrackerbot) on Nov. 13 (6:59:40 a.m. Nov. 14 Japan time). A spec called "Omni LoRA tuning" newly appeared in the API (v1beta1) of "Vertex AI," Google Cloud's developer service, according to the post. The post said there were no docs or announcement yet. Google has not announced the feature, and ORIENS has not been able to confirm it.

🎬 Google is getting LoRA fine-tuning ready for Gemini Omni, its video model. A new "Omni LoRA tuning" spec just appeared in the Vertex AI API (v1beta1): train on your own JSONL dataset for text-to-video or image-to-video, with adapter sizes 8, 16 or 32. No docs or announcement yet.

@aitrackerbot · November 14, 2026

@aitrackerbot is an account that automatically detects and posts AI model releases. Its profile names @SSHCodes as its operator. It is not an official Google account.

Post says training on own data for text- and image-to-video

LoRA is a further-training method that trains only a small add-on component (an adapter) while leaving most of the original model unchanged. According to the post, users can train on a dataset they prepare in JSONL format (a format with one data record per line), for making video from text and for making video from images. Adapter sizes can be chosen from 8, 16 and 32, the post said. All of these are the bot's observations, and the post does not state pricing, release timing or which version of Omni is covered.

No mention in Google Cloud's official documentation

In "Introduction to tuning," official Google Cloud documentation retrieved by ORIENS on Nov. 14, the models listed as supporting supervised further training were text-handling Gemini models such as "Gemini 3.5 Flash," and Gemini Omni was not listed. The same page also had no mention of a feature for further training Omni with LoRA. A name appearing in an API spec does not necessarily mean the feature will be offered as is.

Gemini Omni itself is a video generation AI that Google officially offers. On Nov. 5 (U.S. time), Google announced the new version "Gemini Omni 1.2 Flash" and made it generally available on the Gemini API and Google Cloud. In October, there was a separate leak about "Gemini Omni 1.5," said to be the next version, but Google has not announced 1.5.